randomfolk commited on
Commit
86316a2
·
verified ·
1 Parent(s): 0faa10d

Add anonymous scratch checkpoints, synthetic data, and source

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. review_artifacts/ANONYMIZATION.json +237 -0
  2. review_artifacts/FULL_ARTIFACTS_STATUS.md +3 -36
  3. review_artifacts/MANIFEST.json +0 -0
  4. review_artifacts/README.md +76 -96
  5. review_artifacts/RELEASE_SCOPE.json +8 -0
  6. review_artifacts/archive/outputs/test_main_integrated_20260925/restless/A_alltrials_summary.json +422 -0
  7. review_artifacts/archive/outputs/test_main_integrated_20260925/restless/A_summary.json +1647 -0
  8. review_artifacts/archive/outputs/test_main_integrated_20260925/restless/B_summary.json +2515 -0
  9. review_artifacts/archive/outputs/test_main_integrated_20260925/restless/COMPLETE.json +27 -0
  10. review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/A_alltrials_summary.json +422 -0
  11. review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/A_summary.json +1402 -0
  12. review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/B_summary.json +2515 -0
  13. review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/COMPLETE.json +27 -0
  14. review_artifacts/data/llama_test_20260925/MANIFEST.json +69 -0
  15. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w000.jsonl +0 -0
  16. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w010.jsonl +0 -0
  17. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w030.jsonl +0 -0
  18. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w050.jsonl +0 -0
  19. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w070.jsonl +0 -0
  20. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w090.jsonl +0 -0
  21. review_artifacts/data/llama_test_20260925/task_a/choice/reward_w100.jsonl +0 -0
  22. review_artifacts/data/llama_test_20260925/task_a/full/reward_w000.jsonl +0 -0
  23. review_artifacts/data/llama_test_20260925/task_a/full/reward_w010.jsonl +0 -0
  24. review_artifacts/data/llama_test_20260925/task_a/full/reward_w030.jsonl +0 -0
  25. review_artifacts/data/llama_test_20260925/task_a/full/reward_w050.jsonl +0 -0
  26. review_artifacts/data/llama_test_20260925/task_a/full/reward_w070.jsonl +0 -0
  27. review_artifacts/data/llama_test_20260925/task_a/full/reward_w090.jsonl +0 -0
  28. review_artifacts/data/llama_test_20260925/task_a/full/reward_w100.jsonl +0 -0
  29. review_artifacts/data/llama_test_20260925/task_a/reward_w000_maps.npz +3 -0
  30. review_artifacts/data/llama_test_20260925/task_a/reward_w010_maps.npz +3 -0
  31. review_artifacts/data/llama_test_20260925/task_a/reward_w030_maps.npz +3 -0
  32. review_artifacts/data/llama_test_20260925/task_a/reward_w050_maps.npz +3 -0
  33. review_artifacts/data/llama_test_20260925/task_a/reward_w070_maps.npz +3 -0
  34. review_artifacts/data/llama_test_20260925/task_a/reward_w090_maps.npz +3 -0
  35. review_artifacts/data/llama_test_20260925/task_a/reward_w100_maps.npz +3 -0
  36. review_artifacts/data/llama_test_20260925/task_b/reward_w000.json +0 -0
  37. review_artifacts/data/llama_test_20260925/task_b/reward_w010.json +0 -0
  38. review_artifacts/data/llama_test_20260925/task_b/reward_w030.json +0 -0
  39. review_artifacts/data/llama_test_20260925/task_b/reward_w050.json +0 -0
  40. review_artifacts/data/llama_test_20260925/task_b/reward_w070.json +0 -0
  41. review_artifacts/data/llama_test_20260925/task_b/reward_w090.json +0 -0
  42. review_artifacts/data/llama_test_20260925/task_b/reward_w100.json +0 -0
  43. review_artifacts/outputs/llama_test_20260925/FROZEN_INPUTS.json +125 -0
  44. review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_17.out +39 -0
  45. review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_28.out +39 -0
  46. review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_4.out +35 -0
  47. review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/COMPLETE_shard0.json +269 -0
  48. review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_000.npz +3 -0
  49. review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_001.npz +3 -0
  50. review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_002.npz +3 -0
review_artifacts/ANONYMIZATION.json ADDED
@@ -0,0 +1,237 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_17.out": {
3
+ "original_sha256": "9f8e49a279f225196de83e555b8e3859595b2c7be119b85d21c5f2c5398127f4",
4
+ "anonymous_sha256": "e27c917e0268b328ecdfa8d74817d09873b717cc1f2e77dffa8154692d58a078",
5
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
6
+ },
7
+ "outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_28.out": {
8
+ "original_sha256": "70eda61e3d945affe1fea3d88337ffc4b1a9bead04af93f4ec0a7e60d1b52b45",
9
+ "anonymous_sha256": "e9f5f384f15e96933b391abce2460271b610faadfe988a09fe6fd07a93225968",
10
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
11
+ },
12
+ "outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_4.out": {
13
+ "original_sha256": "db8f24ea4cd04aa4c8efe87a82d8403a659085de723d7d64c96d479e4b4175c8",
14
+ "anonymous_sha256": "f5ba0e71a62b8ba63cddd188afcea46c952bf58896989d9a12b45284080572ea",
15
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
16
+ },
17
+ "outputs/llama_test_20260925/task_a/production/reward_w000/choice/COMPLETE_shard0.json": {
18
+ "original_sha256": "6dd8b9b841b21105e619b65ce45cd846b12113a7a0e525f3092e852e6a33453a",
19
+ "anonymous_sha256": "ca75a2f6fb48001d9cd6f6ab3d683871595f78001c5018a1e24cd12fd48a288b",
20
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
21
+ },
22
+ "outputs/llama_test_20260925/task_a/production/reward_w000/full/COMPLETE_shard0.json": {
23
+ "original_sha256": "fd92733b96c883e2fd45366df72bb5193d3b201cc78fe6d8afa716f2a1a7766b",
24
+ "anonymous_sha256": "36678830be4c7d8661fde56e1bc3bbc7b8cb96e6fd7ca42116c1e5e90ea79748",
25
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
26
+ },
27
+ "outputs/llama_test_20260925/task_a/production/reward_w000/full/COMPLETE_shard1.json": {
28
+ "original_sha256": "7cc89eba53aa68a98155402cf9e86bcaf7cb22a0ab4fffc14ea51ebf938cdb04",
29
+ "anonymous_sha256": "99ce3ad2d82e3c8bb1c897878d08737c881741047a1f646333915259723e0ab1",
30
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
31
+ },
32
+ "outputs/llama_test_20260925/task_a/production/reward_w000/full/COMPLETE_shard2.json": {
33
+ "original_sha256": "61b8f0dea0bb75dfcab94c7362d1ca5abd57ebee50fe0e4ad230bb593ca72983",
34
+ "anonymous_sha256": "87ff2c836cb863b0cedbee29470abcf3b44a93a191b0e6cc8099edd845b1f047",
35
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
36
+ },
37
+ "outputs/llama_test_20260925/task_a/production/reward_w000/full/COMPLETE_shard3.json": {
38
+ "original_sha256": "6baf254516af7657f44d9e16252e3042720ac33c078b2d54b54b7f16d2a04e55",
39
+ "anonymous_sha256": "8c77ff92cc3c6d38e7c7adf54e1859a49ca718aeb18f320697461fe61443818e",
40
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
41
+ },
42
+ "outputs/llama_test_20260925/task_a/production/reward_w000/full/COMPLETE_shard4.json": {
43
+ "original_sha256": "d1c45c5f86dd5c68002fb6919a0c8979bdf73f861780ccaeb75250a0d8fb4af2",
44
+ "anonymous_sha256": "2cbf0b35f57450623a6ff4b0e2ac324281089fefaa8ee2084a8a92601da74624",
45
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
46
+ },
47
+ "outputs/llama_test_20260925/task_a/production/reward_w010/choice/COMPLETE_shard0.json": {
48
+ "original_sha256": "92b6c1499be68a73e6c51aa33be13e58a7f0d7ffa1db4b80b5fa9ff54401a854",
49
+ "anonymous_sha256": "7a7771bef2df42c92f71e0f9d8d342b87143751859c512776047f4af7953c23d",
50
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
51
+ },
52
+ "outputs/llama_test_20260925/task_a/production/reward_w010/full/COMPLETE_shard0.json": {
53
+ "original_sha256": "aa66e9831e22a31cf812ba35b2653da7c0ba7f5e71bc184ad4887eff377febf8",
54
+ "anonymous_sha256": "74fb6482f1809bbd77fd690a14239a4a9fb138d9d1de895f11c8c77b2a6e60e8",
55
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
56
+ },
57
+ "outputs/llama_test_20260925/task_a/production/reward_w010/full/COMPLETE_shard1.json": {
58
+ "original_sha256": "181f4664c5662477715e65d2ad060e94f878543e5bd1ed201cf61a0e8f079aae",
59
+ "anonymous_sha256": "822b923ff3333de7d76e3a94b56c3c4d183f179b030fe9d48b2ce6c884ee5627",
60
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
61
+ },
62
+ "outputs/llama_test_20260925/task_a/production/reward_w010/full/COMPLETE_shard2.json": {
63
+ "original_sha256": "ad6dfeb281a01abcd156a9770d5b7e58ce6d657de9aa7cde34041d31cd4cb2f7",
64
+ "anonymous_sha256": "9fb786e2c8d91cf4455a9b2f709485f4ed924c833ecc0c03b2127241b4bc1802",
65
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
66
+ },
67
+ "outputs/llama_test_20260925/task_a/production/reward_w010/full/COMPLETE_shard3.json": {
68
+ "original_sha256": "449f7ab1636fa6c2a12fb315a95f4106aee4fbce26c5f446a8acfe9567e720ed",
69
+ "anonymous_sha256": "dee6d3dc80b4747d68235b2b012dde84097857b7c725c2e79b11ac464c3e5d4c",
70
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
71
+ },
72
+ "outputs/llama_test_20260925/task_a/production/reward_w010/full/COMPLETE_shard4.json": {
73
+ "original_sha256": "0c73d8d83b46bf87f5655407afe1f335a6a261f1b4004799798cb7fb2d45469b",
74
+ "anonymous_sha256": "be70e279c2d5ad0cb83c01b77fe7789f1f6ef491642c0d13de24e45d17b4a808",
75
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
76
+ },
77
+ "outputs/llama_test_20260925/task_a/production/reward_w030/choice/COMPLETE_shard0.json": {
78
+ "original_sha256": "ef8c2adc1b15326ebff40752b0171e27800a209f57d577a3ed24d073b92b1d1e",
79
+ "anonymous_sha256": "e94149dcaf7f6a95eedaaa6978c43820b95cf3858b0e7143d69480db12f97771",
80
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
81
+ },
82
+ "outputs/llama_test_20260925/task_a/production/reward_w030/full/COMPLETE_shard0.json": {
83
+ "original_sha256": "e65480c44080962c410b2cbd1a4d6e85f8827bfb3e04957b607350afdf1a4551",
84
+ "anonymous_sha256": "ffe503b08926a4e59df3d8b93d16b2cfc9339fa2eb4601fab9c9de05edb29e32",
85
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
86
+ },
87
+ "outputs/llama_test_20260925/task_a/production/reward_w030/full/COMPLETE_shard1.json": {
88
+ "original_sha256": "8b9f055db12fb53fca8237b2b64ecaad755a4729190bd82a2e81e783717e5c81",
89
+ "anonymous_sha256": "2b703e7327d799fbd5a4c307ec648c17446ea131810d343fc90e598516823978",
90
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
91
+ },
92
+ "outputs/llama_test_20260925/task_a/production/reward_w030/full/COMPLETE_shard2.json": {
93
+ "original_sha256": "ffa06bc6e1e8b160516975cb1bcd1f8c51de3f99c563299c79fac0b498243935",
94
+ "anonymous_sha256": "cbb372c8f098e24cf8a117e53a15078f40dfdd680d3d6644de10b9fb68977cd8",
95
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
96
+ },
97
+ "outputs/llama_test_20260925/task_a/production/reward_w030/full/COMPLETE_shard3.json": {
98
+ "original_sha256": "5210fb6710e98178debfbc57a66d4de91da1a9e7ff855d09620da8bdb33a3d41",
99
+ "anonymous_sha256": "ed51fd7562b9200e98c02f46940d38125ea65b99908734468308902f3f4132ce",
100
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
101
+ },
102
+ "outputs/llama_test_20260925/task_a/production/reward_w030/full/COMPLETE_shard4.json": {
103
+ "original_sha256": "79fe2998af61eaeaf3b17508041c50efcc60f87000f9b8ea47caa199d217f975",
104
+ "anonymous_sha256": "f184de16489f626da7c4b12804a253ac17d5da43c5ed95cd7aa06ae2674ce6cc",
105
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
106
+ },
107
+ "outputs/llama_test_20260925/task_a/production/reward_w050/choice/COMPLETE_shard0.json": {
108
+ "original_sha256": "e99ee26e5c7be75a5b3b3253cc0e935dc5e9e908ca2fc7ea94f1ede59e8a9906",
109
+ "anonymous_sha256": "37d3e0a6902a822da72e6e2a64c0392835204322f3589f7e7532a52c5041162d",
110
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
111
+ },
112
+ "outputs/llama_test_20260925/task_a/production/reward_w050/full/COMPLETE_shard0.json": {
113
+ "original_sha256": "4ba54d6816b65c6f1f3dbbbec0cda41be927753883d320f7f420d2469b6c95e6",
114
+ "anonymous_sha256": "7d6e7de8ef19368f2fe5deb9dadbfb1842c37dbd4bfcb6c4e157ee3b012e3521",
115
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
116
+ },
117
+ "outputs/llama_test_20260925/task_a/production/reward_w050/full/COMPLETE_shard1.json": {
118
+ "original_sha256": "1a54b95da936a4c1c21bd0eec06f61e43a306e708477a17c6eeca861b5777ea8",
119
+ "anonymous_sha256": "3dcdde35ec6faea6c06d72c83d3ed8662f360182a0c38b919e643b04477a6cb0",
120
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
121
+ },
122
+ "outputs/llama_test_20260925/task_a/production/reward_w050/full/COMPLETE_shard2.json": {
123
+ "original_sha256": "50714136334953644b297163dab88e46e1e696be8a4592e704731efc8252bfff",
124
+ "anonymous_sha256": "83643590eb5defcf676da377ad4a15c74e46d7b5de5f91f00d0f6b0faf32e167",
125
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
126
+ },
127
+ "outputs/llama_test_20260925/task_a/production/reward_w050/full/COMPLETE_shard3.json": {
128
+ "original_sha256": "e5e4fdd46e750e3db2e9414e6db0e79fac3800b36047455f03bf7895ed5c2c48",
129
+ "anonymous_sha256": "2fca4ee4cd039d776c2f770ea1da94fbad96963ae3fe7b2b112da0a6d4e93003",
130
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
131
+ },
132
+ "outputs/llama_test_20260925/task_a/production/reward_w050/full/COMPLETE_shard4.json": {
133
+ "original_sha256": "94c940b4d6ba3f358777727879ad517edae4d71490f7dffedcb9b94b9d9a390d",
134
+ "anonymous_sha256": "80d92bb662996bdb94fc4fc19e6f20cf7c7d437639ed43236194ce882876a9dc",
135
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
136
+ },
137
+ "outputs/llama_test_20260925/task_a/production/reward_w070/choice/COMPLETE_shard0.json": {
138
+ "original_sha256": "0af576f3283c6a4720ae7274779b8502f7cd32c49c3015baaf3fa9541102d9fc",
139
+ "anonymous_sha256": "ddb871d8a39063d010f05a99738fcff0436f0617cbd677a4b7815e0823e1f7e2",
140
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
141
+ },
142
+ "outputs/llama_test_20260925/task_a/production/reward_w070/full/COMPLETE_shard0.json": {
143
+ "original_sha256": "61db7ecb3418d7ec5d64d548b708145d33434c32e178b75e8d3ea6632cbab967",
144
+ "anonymous_sha256": "31a42fe421c24574c64d091c2fd317e2c12ff135ba5aad3dd51a6700529ff541",
145
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
146
+ },
147
+ "outputs/llama_test_20260925/task_a/production/reward_w070/full/COMPLETE_shard1.json": {
148
+ "original_sha256": "3808a68568e2bbd0349a08f007c39e85250e7076e36c8de96d87b39bb7337148",
149
+ "anonymous_sha256": "3e344b32d25254ac008b9f24028f997215c39a1090d857bcd6fffd05c8e2fb49",
150
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
151
+ },
152
+ "outputs/llama_test_20260925/task_a/production/reward_w070/full/COMPLETE_shard2.json": {
153
+ "original_sha256": "c399415140385896732d5cb84fd09e1cb6edaab4cde4e76f486eab199f5f0465",
154
+ "anonymous_sha256": "8bb319449983449792da18d6118ef49e782ad45874d5eb2191ac55caa5302059",
155
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
156
+ },
157
+ "outputs/llama_test_20260925/task_a/production/reward_w070/full/COMPLETE_shard3.json": {
158
+ "original_sha256": "2789aa223695a21c4827e26d747cddab5abe317cc5c2c48d404ff3f4ffc4bd4e",
159
+ "anonymous_sha256": "2cb1cd98b3c3b2c2472c0c35cd8c305adcb4df2881044ace72c0f98ac76ff622",
160
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
161
+ },
162
+ "outputs/llama_test_20260925/task_a/production/reward_w070/full/COMPLETE_shard4.json": {
163
+ "original_sha256": "19b1918cd2daaf7bafd81ea4ced76e015be00a92002655b88e8c45e2b0804970",
164
+ "anonymous_sha256": "ffacafa82b93a5303a89e5c6355cfa1d6a4e271250e0541df00d9c2d7034a039",
165
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
166
+ },
167
+ "outputs/llama_test_20260925/task_a/production/reward_w090/choice/COMPLETE_shard0.json": {
168
+ "original_sha256": "2318c6392bae7304076cea0bf79ae62774a89c5138a1123599ebf93c4ab05c80",
169
+ "anonymous_sha256": "07235976ea030c2574df48472fadb4ac12986c6c9f1c9cfebf605575436c8506",
170
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
171
+ },
172
+ "outputs/llama_test_20260925/task_a/production/reward_w090/full/COMPLETE_shard0.json": {
173
+ "original_sha256": "cb2f74b54669d5e06a1392d0e663cd714aa315b1aba620a576f39b9cebb0ea72",
174
+ "anonymous_sha256": "75a57a0d4fc690b19afbd637e2ef9898e43d7374f102de54a7e761f0a06d3f50",
175
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
176
+ },
177
+ "outputs/llama_test_20260925/task_a/production/reward_w090/full/COMPLETE_shard1.json": {
178
+ "original_sha256": "c529e0b231ff1a2dc517e14cf2ef81efc8f44b29936b203b06e2b3deec4eeba6",
179
+ "anonymous_sha256": "0b4c41c252a636f4a52e04f75a79f98220383e39b71e7f9e1920137a5f1f6eb3",
180
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
181
+ },
182
+ "outputs/llama_test_20260925/task_a/production/reward_w090/full/COMPLETE_shard2.json": {
183
+ "original_sha256": "e64ce83e185dcb985c39ffd5d770d4616d74eaf8f7f4e59079b17e9e4a7d26e4",
184
+ "anonymous_sha256": "11b7d5fa43d13ceb6ca1773e56b83d9d54f23fabeab8fb96f71378071c1a0473",
185
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
186
+ },
187
+ "outputs/llama_test_20260925/task_a/production/reward_w090/full/COMPLETE_shard3.json": {
188
+ "original_sha256": "5a44e36a283511923038d44c903fc4ffa98ed8c3be7a684d4c251c8c58e887e8",
189
+ "anonymous_sha256": "0f12380bdb79e21aabd5da5c94362f5668f2d2545ca8882a602baec8499cc1a6",
190
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
191
+ },
192
+ "outputs/llama_test_20260925/task_a/production/reward_w090/full/COMPLETE_shard4.json": {
193
+ "original_sha256": "c81c3fdabd49584f1c1894dbcb245efb78792452086278cd45f5a94151107df4",
194
+ "anonymous_sha256": "776c486901c9d82f510405e81895f5bd4f563e49f39a82385df82a050afec7e6",
195
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
196
+ },
197
+ "outputs/llama_test_20260925/task_a/production/reward_w100/choice/COMPLETE_shard0.json": {
198
+ "original_sha256": "ef48d27dec62f5fe82e3aec3db8bbf7b5b020abe11118598b7f96882cd09f7d5",
199
+ "anonymous_sha256": "6b378c4c4ff700a868837d8d1112174431dcbce5992aed2d5b3af98a8a6d56bb",
200
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
201
+ },
202
+ "outputs/llama_test_20260925/task_a/production/reward_w100/full/COMPLETE_shard0.json": {
203
+ "original_sha256": "33e6667faf40979f20e76bfc5a40031c393fe7186fd8e99f95d7a1ae3c00ff43",
204
+ "anonymous_sha256": "fc66b65f36d64227c93d1c4798fcbd64bbfc2c1de6e97c83ec7b69fdbbd2b4d3",
205
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
206
+ },
207
+ "outputs/llama_test_20260925/task_a/production/reward_w100/full/COMPLETE_shard1.json": {
208
+ "original_sha256": "a3e19c935bfc60e3d9116601011c62988fad982a2a0f21620434bf8a14bd879a",
209
+ "anonymous_sha256": "98aa244f1f9a1007e7ef077ff897b6acc2eacbca4ef49339a6c97c73e48f16dc",
210
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
211
+ },
212
+ "outputs/llama_test_20260925/task_a/production/reward_w100/full/COMPLETE_shard2.json": {
213
+ "original_sha256": "57974399faca12df77c37708cc603c102116eb71efca2206b56562fb43bfe032",
214
+ "anonymous_sha256": "97604b6c22613a1cf533e6a2269181581881fca9e350f40d3417adcada28144b",
215
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
216
+ },
217
+ "outputs/llama_test_20260925/task_a/production/reward_w100/full/COMPLETE_shard3.json": {
218
+ "original_sha256": "fc06001ea9b466d69748a0d3008d60ebe27fe67d90b4c1e559638600feb7a843",
219
+ "anonymous_sha256": "3af28e37018ee5253cb9141d2e7077fbec4afb545ae40ddbfcc4b8e995f6ffd4",
220
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
221
+ },
222
+ "outputs/llama_test_20260925/task_a/production/reward_w100/full/COMPLETE_shard4.json": {
223
+ "original_sha256": "d8ba545719c0f1e18d46898ee8fbe3cd709c9d377046731c8c4d0383452c519d",
224
+ "anonymous_sha256": "9eb1fea017769e48fe1888050131745e587be735c7d48791e3c74925418230ea",
225
+ "transformation": "Anonymize execution paths and log metadata; scientific values unchanged."
226
+ },
227
+ "outputs/test_main_integrated_20260925/restless/audit.json": {
228
+ "original_sha256": "2c8eec497a0ba91c8f478c791ddd968b958a171f243541d878c6d283b6aa4458",
229
+ "anonymous_sha256": "9c55c942ef799b19bc8c69eab7193cfea719839afe60195de5ce3ac534a4846a",
230
+ "transformation": "Recomputed input hashes by rerunning the unchanged collector on anonymized released inputs; all three numerical summary files match original SHA-256."
231
+ },
232
+ "outputs/test_main_integrated_20260925/spatial/audit.json": {
233
+ "original_sha256": "cd72684260fa0f5d1a3a563f9ce5ab1791ac4ee0af764acf52af8f2ead31a96c",
234
+ "anonymous_sha256": "cd72684260fa0f5d1a3a563f9ce5ab1791ac4ee0af764acf52af8f2ead31a96c",
235
+ "transformation": "Recomputed input hashes by rerunning the unchanged collector on anonymized released inputs; all three numerical summary files match original SHA-256."
236
+ }
237
+ }
review_artifacts/FULL_ARTIFACTS_STATUS.md CHANGED
@@ -1,36 +1,3 @@
1
- # Full anonymous scratch-model artifacts
2
-
3
- This directory extends the anonymous source package with the synthetic data
4
- and scratch-network checkpoints used in the paper. It is hosted in the
5
- `review_artifacts/` directory of the accompanying anonymous Hugging Face
6
- repository. It is not the compact OpenReview code ZIP.
7
-
8
- ## Included
9
-
10
- - Task A: all seven weights of train/validation raw NPZ trajectories, plus
11
- validation oracle-assay arrays and the generation protocol.
12
- - Task B: all seven weights of train/validation observable JSONL records and
13
- generator-audit NPZ arrays, plus the generation protocol and audit report.
14
- - Prepared train/validation arrays and JSONL transcripts for the scratch and
15
- LLaMA pipelines.
16
- - Every reported GRU and scratch Transformer best checkpoint: two tasks,
17
- full-input and choice-only training, seven weights, three seeds per setting
18
- (168 checkpoints), with matching configurations, metrics, and histories.
19
- - The code, small analysis snapshots, figures, and tests from the anonymous
20
- review package. `MANIFEST.json` lists every file and its SHA-256 digest.
21
-
22
- ## Not yet included
23
-
24
- - The 28 LLaMA SFT adapters are not duplicated in this directory. They are
25
- available at the root of the same anonymous Hugging Face repository. Each
26
- final adapter is about 414 MB. They require the compatible Llama-3.1-70B
27
- base model and GPU software environment to load.
28
- - Reserved test-split trajectories. The paper reports development-validation
29
- results, and the test splits have not been used for its analyses.
30
- - Intermediate training checkpoints and full participant-level probability
31
- exports. Only selected best scratch checkpoints are included.
32
-
33
- The `*_audit.npz` files contain synthetic latent variables for generator
34
- checks. Predictors must receive only observable inputs. The existing README
35
- describes the evaluation routes; this extension does not claim a complete
36
- 70B training rerun or an independently verified installation on a fresh host.
 
1
+ # Release scope
2
+
3
+ Includes independent test data, all main test probability exports and summaries, original train/validation data, and 168 scratch checkpoints. LLaMA adapters remain in the repository task_a/task_b folders. Base-model weights and intermediate training checkpoints are not included. See README.md.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
review_artifacts/MANIFEST.json CHANGED
The diff for this file is too large to render. See raw diff
 
review_artifacts/README.md CHANGED
@@ -1,108 +1,88 @@
1
- # Anonymous full-artifact package
2
 
3
- This folder extends the anonymous ICLR 2027 code supplement with synthetic
4
- train/validation data and GRU/Transformer best checkpoints. It contains no
5
- author metadata, Git history, institutional paths, or reserved test data.
 
6
 
7
- The manuscript PDF and appendices are submitted separately. Final LLaMA SFT
8
- adapters are in the root of the
9
- [same anonymous artifact repository](https://huggingface.co/randomfolk/behavioral-model-ablations-review).
10
- Do not use an author-identifying repository link during review.
11
 
 
 
 
 
 
 
 
 
 
 
12
 
13
- # Behavioral model ablations: sequential-choice benchmark
14
 
15
- This code accompanies *Better Behavioral Prediction, More Faithful Model
16
- Ablations? Evidence from Sequential Choice*. It tests whether a predictor's
17
- response to a reward-history ablation matches the response of the known
18
- policy that generated its training data. The experiments are synthetic; they
19
- do not identify a mechanism in human participants.
20
 
21
- ## Paper-to-code map
22
-
23
- | Paper component | Data and analysis route | Bundled lightweight evidence |
24
- | --- | --- | --- |
25
- | Task A: four-armed restless bandit | `src/mechcal/generators/four_armed_restless.py`, `scripts/run_weight_curve_a.py` | `archive/outputs/weight_curve_a_20260915/reward_w050/val_000.npz` (Figure 1 example) |
26
- | Task B: spatially correlated bandit | `src/mechcal/generators/spatial_mixture.py`, `scripts/generate_spatial_pilot.py`, `scripts/prepare_spatial_neural.py` | `archive/outputs/spatial_pilot_v1/reward_w050/val_audit.npz` (Figure 1 example) |
27
- | Full-input and choice-only GRU/Transformer fits | `src/mechcal/models/`, `scripts/run_weight_neural_suite.py`, `scripts/run_spatial_neural.py` | Figure 2–4 validation summaries |
28
- | LLaMA supervised fine-tuning and evaluation | `src/mechcal/training/finetune_centaur_original.py`, `scripts/prepare_weight_llama.py`, `scripts/eval_weight_llama.py`, `scripts/eval_spatial_llama.py` | Integrated summaries; final adapters are hosted separately |
29
- | Oracle, fitted generator, and pooled RW | `scripts/run_reference_followup.py`, `scripts/ablate_spatial_references.py`, `scripts/run_rw_baseline.py` | Integrated summaries |
30
- | Matched donor-reward ablation | `scripts/evaluate_weight_neural.py`, `scripts/analyze_spatial_neural.py`, `scripts/eval_weight_llama_operations_fixedshape.py` | Figures 3–4 and operation-sweep summary |
31
- | Neural-generator and operation sensitivity | `scripts/run_neural_generator_sweep.py`, `scripts/run_trained_transformer_generator.py`, `scripts/run_random_generator_controls.py`, `scripts/plot_weight_operation_sweep_complete.py` | Supplementary figure PDFs in `paper_figures/` |
32
 
33
- For donor replacement, observed past choices, target choices, trained model
34
- parameters, and donor assignments are held fixed. The response is the change
35
- in valid-choice probabilities or choice NLL after replacing past chosen
36
- rewards. Choice-only training is a *different* comparison: its network never
37
- sees rewards during fitting. The oracle uses true generating parameters;
38
- the fitted-generator reference estimates participant parameters from an
39
- earlier prefix. These three operations must not be conflated.
 
40
 
41
- ## Quick check and figure rebuild
42
 
43
- From the package root, in an environment with Python 3.10+:
44
 
45
- ```bash
46
- python -m pip install -r requirements.txt
47
- python -m pytest -q -p no:cacheprovider tests/test_four_armed_restless.py tests/test_spatial_mixture.py tests/test_reward_ablation.py
48
- python scripts/reproduce_figures.py
49
  ```
50
 
51
- `requirements.txt` is a convenience installer, not a lockfile for the original
52
- GPU jobs. It installs the local package and its analysis/test dependencies plus
53
- PyTorch for the quick checks and scratch networks. For GPU training, choose a
54
- PyTorch build compatible with the local CUDA driver before running the training
55
- scripts. The 70B LLaMA path additionally requires a separately configured
56
- Unsloth/TRL/Transformers environment and access to the base model; those
57
- hardware- and platform-specific dependencies are not claimed to be covered by
58
- this requirements file.
59
-
60
- The last command recreates Figures 1–4 and S1 from the frozen development-set
61
- summary records in `archive/`; it does not rerun fitting. The exact figure
62
- PDFs used in the manuscript are also in `paper_figures/`. The source-to-figure
63
- and inclusion hashes are listed in `MANIFEST.json`.
64
-
65
- ## Full experimental route
66
-
67
- 1. Generate independent train/validation/test synthetic trajectories with
68
- `scripts/run_weight_curve_a.py` (Task A) and
69
- `scripts/generate_spatial_pilot.py` (Task B). Start with `--help` for output
70
- locations and use a new empty output directory. Generation writes a test
71
- split, but the reported analyses use the development validation split.
72
- 2. Fit weight-specific, full-input and choice-only scratch networks with
73
- `scripts/run_weight_neural_suite.py` and `scripts/run_spatial_neural.py`.
74
- The main training code and configs are under `src/mechcal/training/` and
75
- `configs/`.
76
- 3. Fit symbolic references and evaluate the same held-out choices and donor
77
- assignments. Use the scripts in the map above; do not treat the oracle as
78
- a fitted model or choose donor mappings separately by predictor.
79
- 4. For LLaMA, prepare JSONL transcripts, fine-tune the pretrained
80
- Llama-3.1-70B *base* model with LoRA, then evaluate with the fixed-shape
81
- uncached causal procedure. This path requires access to the base model,
82
- Unsloth, TRL, GPU resources, and a separately configured environment.
83
- No pretrained base weights, adapters, or private cluster paths are bundled
84
- in this folder. Final adapters are in the repository root.
85
- 5. Integrate audited outputs with `scripts/integrate_llama_main.py` and
86
- `scripts/integrate_spatial_llama.py`; use the figure scripts to regenerate
87
- the complete plots. The accompanying paper specifies counts, windows,
88
- optimization, and uncertainty conventions.
89
-
90
- The bundled `archive/` is intentionally compact: selected synthetic Figure 1
91
- inputs and aggregate development-set summaries, **not** the 10+ GB of
92
- intermediate checkpoints, all participant-level probability exports, or a
93
- turnkey rerun of the 70B training. Selected scratch checkpoints are under
94
- `outputs/`. The test split was not used for the reported claims. The
95
- summary-based figure rebuild is verified separately from full training.
96
-
97
- ## Structure
98
-
99
- - `src/mechcal/`: generators, transcript builders, predictors, and ablations.
100
- - `scripts/`: generation, training, evaluation, integration, and plotting.
101
- - `configs/`: recorded scratch-network and generator configurations.
102
- - `tests/`: unit and small integration checks.
103
- - `archive/`: frozen figure inputs and audited aggregate summaries.
104
- - `paper_figures/`: PDF figures matching the manuscript revision.
105
- - `requirements.txt`: quick-check and scratch-network installer (not a GPU lockfile).
106
- - `MANIFEST.json`: relative-file SHA-256 inventory of this package.
107
-
108
- No human data are included. All bundled trajectories are synthetic.
 
1
+ # Behavioral model ablations: anonymous review materials
2
 
3
+ This release accompanies the frozen independent-test manuscript. The code ZIP
4
+ and anonymous repository share the same scientific code and main-figure inputs.
5
+ No human data, credentials, Git history, or author-identifying manuscript source
6
+ is included. The separate private repository candidate has the same payload.
7
 
8
+ ## What is where
 
 
 
9
 
10
+ - Supplementary ZIP: source, configurations, requirements, tests, frozen test
11
+ summaries, figure scripts, and the exact manuscript figure PDFs.
12
+ - Anonymous repository: https://huggingface.co/randomfolk/behavioral-model-ablations-review
13
+ `review_artifacts/` contains the code plus full synthetic test records,
14
+ prepared test prompts, participant-level probability exports, cognitive fits,
15
+ audit manifests, and the previously released train/validation data and 168
16
+ validation-selected GRU/Transformer checkpoints.
17
+ - `task_a/` and `task_b/` in that repository contain all 28 final LLaMA adapters.
18
+ Base LLaMA weights are not redistributed. Use the specified base model under
19
+ its applicable access and licensing terms.
20
 
21
+ ## Quick checks and main-figure rebuild (no inference)
22
 
23
+ From the package root, with Python 3.10 or newer:
 
 
 
 
24
 
25
+ ```sh
26
+ python -m pip install -r requirements.txt
27
+ python -m pytest -q -p no:cacheprovider tests/test_four_armed_restless.py tests/test_spatial_mixture.py tests/test_reward_ablation.py
28
+ python scripts/reproduce_figures.py
29
+ ```
 
 
 
 
 
 
30
 
31
+ The last command renders Figures 1 to 4 and the choice-only behavioral-baseline
32
+ supplement (Figure S3 in the manuscript). It uses the independent-test summaries
33
+ in `archive/outputs/test_main_integrated_20260925/`, not old validation curves.
34
+ Figure 1A retains the manuscript's labeled illustrative validation trial;
35
+ Figure 1C uses test means. `paper_figures/` contains the exact frozen PDFs for
36
+ all main and supplementary figures. Other supplementary generator/operation
37
+ analyses remain validation or training analyses, as labeled in the manuscript.
38
+ Saved-summary plotting is not a rerun of model training.
39
 
40
+ ## Full test reaggregation (download review_artifacts first)
41
 
42
+ Run from the root containing `src/`, `data/`, `outputs/`, and `scripts/`:
43
 
44
+ ```sh
45
+ python scripts/collect_llama_test_20260925.py --self-test
46
+ python scripts/collect_llama_test_20260925.py --task both --output reproduced_test_summaries
 
47
  ```
48
 
49
+ This uses the released participant exports and checks frozen checkpoint/script
50
+ provenance, subject IDs, actual choices, donor mappings, conditional probability
51
+ normalization, causal probes and metric identities before aggregation. It does
52
+ not run LLaMA inference. Original absolute compute paths have been replaced
53
+ with package-relative paths. Three resumed shards use anonymized extracts of
54
+ the original successful causal-probe logs; their source hashes are retained in
55
+ `ANONYMIZATION.json`. The release inventory hashes the sanitized files.
56
+
57
+ ## Experimental routes and GPU requirements
58
+
59
+ | Paper component | Code |
60
+ | --- | --- |
61
+ | Restless and spatial generators | `src/mechcal/generators/`, `scripts/run_weight_curve_a.py`, `scripts/generate_spatial_pilot.py` |
62
+ | Scratch neural training | `scripts/run_weight_neural_suite.py`, `scripts/run_spatial_neural.py` |
63
+ | Non-LLM test evaluation and reference fitting | `scripts/test_nonllm_20260925.py`, `scripts/test_nonllm_audit_20260925.py` |
64
+ | Frozen LLaMA test payloads | `data/llama_test_20260925/`, `scripts/prepare_llama_test_20260925.py` |
65
+ | LLaMA test inference | `scripts/eval_weight_llama.py --test-main`, `scripts/eval_spatial_llama.py --test-main` |
66
+ | Test collection | `scripts/collect_llama_test_20260925.py` |
67
+ | Main-figure reconstruction | `scripts/reproduce_figures.py` |
68
+
69
+ Use each script's `--help` for task, mode and weight arguments. The non-LLM
70
+ pipeline phases are prepare, neural, spatial_fits, references, and aggregate;
71
+ the released caches allow aggregation without rerunning fits. Never overwrite
72
+ the archived test results when exploring new settings. Models retain their
73
+ validation-selected checkpoints, with no test-based tuning or retraining.
74
+ Individual fitted generators use only the first 150 trials or first six maps;
75
+ scored targets are trials 151-200 or maps 7-8. Both tasks use seven weights and
76
+ 20 shared donor derangements, with actual choices retained.
77
+
78
+ The requirements file supports the analysis and scratch-network paths, not a
79
+ locked 70B GPU environment. LLaMA additionally needs compatible Unsloth, TRL,
80
+ Transformers, bitsandbytes, CUDA, and substantial GPU memory. To rerun inference,
81
+ place adapters in the relative paths recorded in `FROZEN_INPUTS.json`; their
82
+ Hub locations are `task_a/{full,choice_only}/reward_wXXX` and
83
+ `task_b/{full,choice_only}/reward_wXXX`.
84
+ No full retraining or 70B inference rerun is claimed for this release check.
85
+
86
+ `MANIFEST.json` gives SHA-256 hashes. `RELEASE_SCOPE.json` records which materials
87
+ are test versus development. The manuscript itself is frozen and distributed
88
+ separately. No new code/data license is granted by this review snapshot.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
review_artifacts/RELEASE_SCOPE.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "main_figures": "independent test",
3
+ "choice_baselines": "independent test",
4
+ "spatial_prediction_table": "independent test",
5
+ "figure1_A": "validation illustration",
6
+ "other_supplements": "original training/validation analyses",
7
+ "manuscript_modified": false
8
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/restless/A_alltrials_summary.json ADDED
@@ -0,0 +1,422 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "all": {
4
+ "oracle_nll": {
5
+ "mean": 0.7615397192025329,
6
+ "ci95": [
7
+ 0.7087965605773747,
8
+ 0.8125769691052986
9
+ ]
10
+ },
11
+ "rw_nll": {
12
+ "mean": 1.3862943611198908,
13
+ "ci95": [
14
+ 1.3862943611198908,
15
+ 1.3862943611198908
16
+ ]
17
+ },
18
+ "gru_full_nll": {
19
+ "mean": 0.7666141390800476,
20
+ "ci95": [
21
+ 0.7139537677168846,
22
+ 0.8174964696168899
23
+ ]
24
+ },
25
+ "gru_choice_nll": {
26
+ "mean": 0.765352189540863,
27
+ "ci95": [
28
+ 0.7126925647258758,
29
+ 0.8162335872650146
30
+ ]
31
+ },
32
+ "transformer_full_nll": {
33
+ "mean": 0.8390780091285706,
34
+ "ci95": [
35
+ 0.7780588790774345,
36
+ 0.8982668563723564
37
+ ]
38
+ },
39
+ "transformer_choice_nll": {
40
+ "mean": 0.7693579196929932,
41
+ "ci95": [
42
+ 0.7172033414244652,
43
+ 0.8200799733400345
44
+ ]
45
+ },
46
+ "llama_full_nll": {
47
+ "mean": 0.7662451775347893,
48
+ "ci95": [
49
+ 0.7135033650741311,
50
+ 0.8171215551273225
51
+ ]
52
+ },
53
+ "llama_choice_nll": {
54
+ "mean": 0.7665152796540999,
55
+ "ci95": [
56
+ 0.7136117825521615,
57
+ 0.817566579761824
58
+ ]
59
+ }
60
+ }
61
+ },
62
+ "0.1": {
63
+ "all": {
64
+ "oracle_nll": {
65
+ "mean": 0.35870988828155703,
66
+ "ci95": [
67
+ 0.35137164868033566,
68
+ 0.3667316486038875
69
+ ]
70
+ },
71
+ "rw_nll": {
72
+ "mean": 1.3312757648089386,
73
+ "ci95": [
74
+ 1.3122212688350483,
75
+ 1.3510878929808616
76
+ ]
77
+ },
78
+ "gru_full_nll": {
79
+ "mean": 0.4017099440097809,
80
+ "ci95": [
81
+ 0.3923706367611885,
82
+ 0.4115542151033878
83
+ ]
84
+ },
85
+ "gru_choice_nll": {
86
+ "mean": 0.44611719250679016,
87
+ "ci95": [
88
+ 0.43175888806581497,
89
+ 0.4598716519773006
90
+ ]
91
+ },
92
+ "transformer_full_nll": {
93
+ "mean": 0.49322184920310974,
94
+ "ci95": [
95
+ 0.4738965094089508,
96
+ 0.5117886886000633
97
+ ]
98
+ },
99
+ "transformer_choice_nll": {
100
+ "mean": 0.46376892924308777,
101
+ "ci95": [
102
+ 0.44772387370467187,
103
+ 0.47943436428904534
104
+ ]
105
+ },
106
+ "llama_full_nll": {
107
+ "mean": 0.38808746656092313,
108
+ "ci95": [
109
+ 0.37951394922040477,
110
+ 0.39703506995977317
111
+ ]
112
+ },
113
+ "llama_choice_nll": {
114
+ "mean": 0.4436390406101226,
115
+ "ci95": [
116
+ 0.42963034279429196,
117
+ 0.4570294804439361
118
+ ]
119
+ }
120
+ }
121
+ },
122
+ "0.3": {
123
+ "all": {
124
+ "oracle_nll": {
125
+ "mean": 0.35878210663980986,
126
+ "ci95": [
127
+ 0.3511238881793803,
128
+ 0.36665133044372394
129
+ ]
130
+ },
131
+ "rw_nll": {
132
+ "mean": 1.2564486111888118,
133
+ "ci95": [
134
+ 1.2355956336607112,
135
+ 1.2771085784378613
136
+ ]
137
+ },
138
+ "gru_full_nll": {
139
+ "mean": 0.4201349914073944,
140
+ "ci95": [
141
+ 0.41033915653824804,
142
+ 0.43012534752488135
143
+ ]
144
+ },
145
+ "gru_choice_nll": {
146
+ "mean": 0.4665714204311371,
147
+ "ci95": [
148
+ 0.45299574360251427,
149
+ 0.48032378405332565
150
+ ]
151
+ },
152
+ "transformer_full_nll": {
153
+ "mean": 0.5237559080123901,
154
+ "ci95": [
155
+ 0.5043867245316506,
156
+ 0.5423404946923256
157
+ ]
158
+ },
159
+ "transformer_choice_nll": {
160
+ "mean": 0.48638489842414856,
161
+ "ci95": [
162
+ 0.47092897072434425,
163
+ 0.5018887862563133
164
+ ]
165
+ },
166
+ "llama_full_nll": {
167
+ "mean": 0.39776558776061893,
168
+ "ci95": [
169
+ 0.3890470136299825,
170
+ 0.4064857188887151
171
+ ]
172
+ },
173
+ "llama_choice_nll": {
174
+ "mean": 0.4626961844524759,
175
+ "ci95": [
176
+ 0.44954073198487793,
177
+ 0.4759124844008198
178
+ ]
179
+ }
180
+ }
181
+ },
182
+ "0.5": {
183
+ "all": {
184
+ "oracle_nll": {
185
+ "mean": 0.35913403593617477,
186
+ "ci95": [
187
+ 0.35170791159006376,
188
+ 0.366819944844785
189
+ ]
190
+ },
191
+ "rw_nll": {
192
+ "mean": 0.9238715607071197,
193
+ "ci95": [
194
+ 0.8981959717451431,
195
+ 0.950433208083594
196
+ ]
197
+ },
198
+ "gru_full_nll": {
199
+ "mean": 0.4410792887210846,
200
+ "ci95": [
201
+ 0.4309645421802998,
202
+ 0.4508512854576111
203
+ ]
204
+ },
205
+ "gru_choice_nll": {
206
+ "mean": 0.5135248303413391,
207
+ "ci95": [
208
+ 0.4995715029537678,
209
+ 0.526460325717926
210
+ ]
211
+ },
212
+ "transformer_full_nll": {
213
+ "mean": 0.5814924240112305,
214
+ "ci95": [
215
+ 0.563471020758152,
216
+ 0.5990168735384941
217
+ ]
218
+ },
219
+ "transformer_choice_nll": {
220
+ "mean": 0.5406873226165771,
221
+ "ci95": [
222
+ 0.5256843939423561,
223
+ 0.5549581110477447
224
+ ]
225
+ },
226
+ "llama_full_nll": {
227
+ "mean": 0.41266504845863944,
228
+ "ci95": [
229
+ 0.4037584289914706,
230
+ 0.42207593869385274
231
+ ]
232
+ },
233
+ "llama_choice_nll": {
234
+ "mean": 0.5112831870463022,
235
+ "ci95": [
236
+ 0.49731910312490196,
237
+ 0.524102012648826
238
+ ]
239
+ }
240
+ }
241
+ },
242
+ "0.7": {
243
+ "all": {
244
+ "oracle_nll": {
245
+ "mean": 0.36202517487995506,
246
+ "ci95": [
247
+ 0.35390386146085817,
248
+ 0.3699065386930659
249
+ ]
250
+ },
251
+ "rw_nll": {
252
+ "mean": 0.6791247748808362,
253
+ "ci95": [
254
+ 0.6613463430446613,
255
+ 0.6978278673861364
256
+ ]
257
+ },
258
+ "gru_full_nll": {
259
+ "mean": 0.4510907530784607,
260
+ "ci95": [
261
+ 0.439642346650362,
262
+ 0.4623833209276199
263
+ ]
264
+ },
265
+ "gru_choice_nll": {
266
+ "mean": 0.5267288684844971,
267
+ "ci95": [
268
+ 0.511530464887619,
269
+ 0.5414909422397614
270
+ ]
271
+ },
272
+ "transformer_full_nll": {
273
+ "mean": 0.5742741227149963,
274
+ "ci95": [
275
+ 0.5542959734797478,
276
+ 0.5929679587483405
277
+ ]
278
+ },
279
+ "transformer_choice_nll": {
280
+ "mean": 0.5560355186462402,
281
+ "ci95": [
282
+ 0.5389567390084267,
283
+ 0.5725994914770126
284
+ ]
285
+ },
286
+ "llama_full_nll": {
287
+ "mean": 0.41690015245573986,
288
+ "ci95": [
289
+ 0.4074843792006891,
290
+ 0.4266876236137902
291
+ ]
292
+ },
293
+ "llama_choice_nll": {
294
+ "mean": 0.5242757868270241,
295
+ "ci95": [
296
+ 0.5091511251788423,
297
+ 0.5389106728228897
298
+ ]
299
+ }
300
+ }
301
+ },
302
+ "0.9": {
303
+ "all": {
304
+ "oracle_nll": {
305
+ "mean": 0.3601490293752272,
306
+ "ci95": [
307
+ 0.352755721802086,
308
+ 0.36788549607884136
309
+ ]
310
+ },
311
+ "rw_nll": {
312
+ "mean": 0.5891644033799003,
313
+ "ci95": [
314
+ 0.5751712216297498,
315
+ 0.6035675240041773
316
+ ]
317
+ },
318
+ "gru_full_nll": {
319
+ "mean": 0.44056054949760437,
320
+ "ci95": [
321
+ 0.43067443892359736,
322
+ 0.45060925334692004
323
+ ]
324
+ },
325
+ "gru_choice_nll": {
326
+ "mean": 0.5046015977859497,
327
+ "ci95": [
328
+ 0.49213571324944494,
329
+ 0.5163752377033234
330
+ ]
331
+ },
332
+ "transformer_full_nll": {
333
+ "mean": 0.49121472239494324,
334
+ "ci95": [
335
+ 0.4772181138396263,
336
+ 0.5046241715550422
337
+ ]
338
+ },
339
+ "transformer_choice_nll": {
340
+ "mean": 0.5215232968330383,
341
+ "ci95": [
342
+ 0.5076067343354225,
343
+ 0.5344561904668808
344
+ ]
345
+ },
346
+ "llama_full_nll": {
347
+ "mean": 0.4102427099357991,
348
+ "ci95": [
349
+ 0.4014488634769003,
350
+ 0.419269722284018
351
+ ]
352
+ },
353
+ "llama_choice_nll": {
354
+ "mean": 0.5031357212720928,
355
+ "ci95": [
356
+ 0.49092453918158224,
357
+ 0.514781433991205
358
+ ]
359
+ }
360
+ }
361
+ },
362
+ "1.0": {
363
+ "all": {
364
+ "oracle_nll": {
365
+ "mean": 0.3620594985830462,
366
+ "ci95": [
367
+ 0.35444677710644357,
368
+ 0.3698726163956932
369
+ ]
370
+ },
371
+ "rw_nll": {
372
+ "mean": 0.582847877116578,
373
+ "ci95": [
374
+ 0.5697373541317889,
375
+ 0.5966154148776828
376
+ ]
377
+ },
378
+ "gru_full_nll": {
379
+ "mean": 0.42268866300582886,
380
+ "ci95": [
381
+ 0.4130487136542797,
382
+ 0.4326356008648872
383
+ ]
384
+ },
385
+ "gru_choice_nll": {
386
+ "mean": 0.4987739026546478,
387
+ "ci95": [
388
+ 0.48679227083921434,
389
+ 0.510681937634945
390
+ ]
391
+ },
392
+ "transformer_full_nll": {
393
+ "mean": 0.4692218601703644,
394
+ "ci95": [
395
+ 0.4572054646909237,
396
+ 0.481750264018774
397
+ ]
398
+ },
399
+ "transformer_choice_nll": {
400
+ "mean": 0.5105881690979004,
401
+ "ci95": [
402
+ 0.4981814078986645,
403
+ 0.5231763079762459
404
+ ]
405
+ },
406
+ "llama_full_nll": {
407
+ "mean": 0.4032484353783656,
408
+ "ci95": [
409
+ 0.3947084346898632,
410
+ 0.41207386575583166
411
+ ]
412
+ },
413
+ "llama_choice_nll": {
414
+ "mean": 0.4981825190274488,
415
+ "ci95": [
416
+ 0.4862601051845965,
417
+ 0.5099427641669416
418
+ ]
419
+ }
420
+ }
421
+ }
422
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/restless/A_summary.json ADDED
@@ -0,0 +1,1647 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "all": {
4
+ "oracle_intact_nll": {
5
+ "mean": 0.7622831832040385,
6
+ "ci95": [
7
+ 0.7059837172557903,
8
+ 0.817094916137887
9
+ ]
10
+ },
11
+ "oracle_donor_nll": {
12
+ "mean": 0.7622831832040387,
13
+ "ci95": [
14
+ 0.7059837172557903,
15
+ 0.8170949161378871
16
+ ]
17
+ },
18
+ "oracle_delta_nll": {
19
+ "mean": 1.121325254871408e-16,
20
+ "ci95": [
21
+ 8.94784246696645e-17,
22
+ 1.3467282844459305e-16
23
+ ]
24
+ },
25
+ "oracle_response_error": {
26
+ "mean": 0.0,
27
+ "ci95": [
28
+ 0.0,
29
+ 0.0
30
+ ]
31
+ },
32
+ "fit_intact_nll": {
33
+ "mean": 0.8483674662414769,
34
+ "ci95": [
35
+ 0.7854019202620498,
36
+ 0.9080846057346155
37
+ ]
38
+ },
39
+ "fit_donor_nll": {
40
+ "mean": 0.7847864236356374,
41
+ "ci95": [
42
+ 0.7295404441916973,
43
+ 0.8383608372209361
44
+ ]
45
+ },
46
+ "fit_delta_nll": {
47
+ "mean": -0.06358104260583938,
48
+ "ci95": [
49
+ -0.1038799697863227,
50
+ -0.028367010029941655
51
+ ]
52
+ },
53
+ "fit_response_error": {
54
+ "mean": 0.0281268,
55
+ "ci95": [
56
+ 0.01593846,
57
+ 0.041662709999999985
58
+ ]
59
+ },
60
+ "rw_intact_nll": {
61
+ "mean": 1.3862943611198908,
62
+ "ci95": [
63
+ 1.3862943611198908,
64
+ 1.3862943611198908
65
+ ]
66
+ },
67
+ "rw_donor_nll": {
68
+ "mean": 1.3862943611198908,
69
+ "ci95": [
70
+ 1.3862943611198908,
71
+ 1.3862943611198908
72
+ ]
73
+ },
74
+ "rw_delta_nll": {
75
+ "mean": -2.220446049250313e-16,
76
+ "ci95": [
77
+ -2.220446049250313e-16,
78
+ -2.220446049250313e-16
79
+ ]
80
+ },
81
+ "rw_response_error": {
82
+ "mean": 0.0,
83
+ "ci95": [
84
+ 0.0,
85
+ 0.0
86
+ ]
87
+ },
88
+ "baseline_uniform_nll": {
89
+ "mean": 1.3862943611198908,
90
+ "ci95": [
91
+ 1.3862943611198908,
92
+ 1.3862943611198908
93
+ ]
94
+ },
95
+ "baseline_static_arm_frequency_nll": {
96
+ "mean": 1.385546742084214,
97
+ "ci95": [
98
+ 1.3849631356489776,
99
+ 1.3861351594596039
100
+ ]
101
+ },
102
+ "baseline_static_stay_probability_nll": {
103
+ "mean": 1.2526949219191008,
104
+ "ci95": [
105
+ 1.1946047230170138,
106
+ 1.308789359487401
107
+ ]
108
+ },
109
+ "baseline_static_transition_nll": {
110
+ "mean": 1.2519496137930815,
111
+ "ci95": [
112
+ 1.1937931037752816,
113
+ 1.3085098754702005
114
+ ]
115
+ },
116
+ "baseline_train_selected_lag_nll": {
117
+ "mean": 1.1241737276488215,
118
+ "ci95": [
119
+ 1.0698406027241103,
120
+ 1.175234332020613
121
+ ]
122
+ },
123
+ "gru_intact_nll": {
124
+ "mean": 0.7630717979613043,
125
+ "ci95": [
126
+ 0.7069652555585906,
127
+ 0.8178074233535216
128
+ ]
129
+ },
130
+ "gru_donor_nll": {
131
+ "mean": 0.763411439820356,
132
+ "ci95": [
133
+ 0.7073599961361051,
134
+ 0.8180875486809018
135
+ ]
136
+ },
137
+ "gru_delta_nll": {
138
+ "mean": 0.00033964185905163214,
139
+ "ci95": [
140
+ -0.00014853869217058664,
141
+ 0.0008500300455197925
142
+ ]
143
+ },
144
+ "gru_response_error": {
145
+ "mean": 0.00837545404664359,
146
+ "ci95": [
147
+ 0.0077797547158419255,
148
+ 0.009000036130773954
149
+ ]
150
+ },
151
+ "gru_full_nll": {
152
+ "mean": 0.7630717979613043,
153
+ "ci95": [
154
+ 0.7069652555585906,
155
+ 0.8178074233535216
156
+ ]
157
+ },
158
+ "gru_choice_nll": {
159
+ "mean": 0.7624148767056697,
160
+ "ci95": [
161
+ 0.7060824576577139,
162
+ 0.8172178009425823
163
+ ]
164
+ },
165
+ "transformer_intact_nll": {
166
+ "mean": 0.8271710782106564,
167
+ "ci95": [
168
+ 0.7634170516371505,
169
+ 0.888664995515858
170
+ ]
171
+ },
172
+ "transformer_donor_nll": {
173
+ "mean": 0.8269989130554898,
174
+ "ci95": [
175
+ 0.7629776556548822,
176
+ 0.8882924109869897
177
+ ]
178
+ },
179
+ "transformer_delta_nll": {
180
+ "mean": -0.0001721651551665042,
181
+ "ci95": [
182
+ -0.0005256870249537961,
183
+ 0.00020807734246566442
184
+ ]
185
+ },
186
+ "transformer_response_error": {
187
+ "mean": 0.008717824416243138,
188
+ "ci95": [
189
+ 0.008404769848386695,
190
+ 0.009033003046961053
191
+ ]
192
+ },
193
+ "transformer_full_nll": {
194
+ "mean": 0.8271710782106564,
195
+ "ci95": [
196
+ 0.7634170516371505,
197
+ 0.888664995515858
198
+ ]
199
+ },
200
+ "transformer_choice_nll": {
201
+ "mean": 0.7662887955148586,
202
+ "ci95": [
203
+ 0.7103074068328358,
204
+ 0.8206538647105449
205
+ ]
206
+ },
207
+ "oracle_nll": {
208
+ "mean": 0.7622831832040385,
209
+ "ci95": [
210
+ 0.7059837172557903,
211
+ 0.817094916137887
212
+ ]
213
+ },
214
+ "llama_full_nll": {
215
+ "mean": 0.7630837702473433,
216
+ "ci95": [
217
+ 0.7081551329450082,
218
+ 0.8166046093560824
219
+ ]
220
+ },
221
+ "llama_choice_nll": {
222
+ "mean": 0.7625497153058011,
223
+ "ci95": [
224
+ 0.7080384854284038,
225
+ 0.8159663827425214
226
+ ]
227
+ },
228
+ "llama_reward_predictive_gain": {
229
+ "mean": -0.0005340549415423553,
230
+ "ci95": [
231
+ -0.0013722305514553377,
232
+ 0.0003064117202134289
233
+ ]
234
+ }
235
+ }
236
+ },
237
+ "0.1": {
238
+ "all": {
239
+ "oracle_intact_nll": {
240
+ "mean": 0.35094693566561525,
241
+ "ci95": [
242
+ 0.33274790938588833,
243
+ 0.3697309306471904
244
+ ]
245
+ },
246
+ "oracle_donor_nll": {
247
+ "mean": 1.337810801191282,
248
+ "ci95": [
249
+ 1.2007934377206229,
250
+ 1.4647320381781874
251
+ ]
252
+ },
253
+ "oracle_delta_nll": {
254
+ "mean": 0.9868638655256665,
255
+ "ci95": [
256
+ 0.8510915465512873,
257
+ 1.1118760102121572
258
+ ]
259
+ },
260
+ "oracle_response_error": {
261
+ "mean": 0.0,
262
+ "ci95": [
263
+ 0.0,
264
+ 0.0
265
+ ]
266
+ },
267
+ "fit_intact_nll": {
268
+ "mean": 0.44800840915749196,
269
+ "ci95": [
270
+ 0.4023664067416689,
271
+ 0.4968280193364418
272
+ ]
273
+ },
274
+ "fit_donor_nll": {
275
+ "mean": 1.3569775534715973,
276
+ "ci95": [
277
+ 1.224403785584238,
278
+ 1.4814011185383267
279
+ ]
280
+ },
281
+ "fit_delta_nll": {
282
+ "mean": 0.9089691443141052,
283
+ "ci95": [
284
+ 0.7617527431991828,
285
+ 1.0464237899439852
286
+ ]
287
+ },
288
+ "fit_response_error": {
289
+ "mean": 0.043848,
290
+ "ci95": [
291
+ 0.031063769999999997,
292
+ 0.05833673999999999
293
+ ]
294
+ },
295
+ "rw_intact_nll": {
296
+ "mean": 1.3303796867230553,
297
+ "ci95": [
298
+ 1.3058373981393419,
299
+ 1.357189234173727
300
+ ]
301
+ },
302
+ "rw_donor_nll": {
303
+ "mean": 1.4153938308975889,
304
+ "ci95": [
305
+ 1.4116780209577373,
306
+ 1.4192131143716538
307
+ ]
308
+ },
309
+ "rw_delta_nll": {
310
+ "mean": 0.08501414417453378,
311
+ "ci95": [
312
+ 0.05807879587990905,
313
+ 0.10952223843493773
314
+ ]
315
+ },
316
+ "rw_response_error": {
317
+ "mean": 0.387172420067802,
318
+ "ci95": [
319
+ 0.3580775588302532,
320
+ 0.4143837410618941
321
+ ]
322
+ },
323
+ "baseline_uniform_nll": {
324
+ "mean": 1.3862943611198908,
325
+ "ci95": [
326
+ 1.3862943611198908,
327
+ 1.3862943611198908
328
+ ]
329
+ },
330
+ "baseline_static_arm_frequency_nll": {
331
+ "mean": 1.3848891461239807,
332
+ "ci95": [
333
+ 1.3827500108058508,
334
+ 1.3870151128376051
335
+ ]
336
+ },
337
+ "baseline_static_stay_probability_nll": {
338
+ "mean": 1.2552518401712833,
339
+ "ci95": [
340
+ 1.1967124403127034,
341
+ 1.3110042183403439
342
+ ]
343
+ },
344
+ "baseline_static_transition_nll": {
345
+ "mean": 1.255555000234112,
346
+ "ci95": [
347
+ 1.197949683469721,
348
+ 1.3118753594860326
349
+ ]
350
+ },
351
+ "baseline_train_selected_lag_nll": {
352
+ "mean": 0.6292175672517345,
353
+ "ci95": [
354
+ 0.6005990478124624,
355
+ 0.6578254121220253
356
+ ]
357
+ },
358
+ "gru_intact_nll": {
359
+ "mean": 0.38209832533331256,
360
+ "ci95": [
361
+ 0.3621853450495232,
362
+ 0.401618977606335
363
+ ]
364
+ },
365
+ "gru_donor_nll": {
366
+ "mean": 0.5700262028629846,
367
+ "ci95": [
368
+ 0.5347388137375804,
369
+ 0.604377743721662
370
+ ]
371
+ },
372
+ "gru_delta_nll": {
373
+ "mean": 0.18792787752967197,
374
+ "ci95": [
375
+ 0.16081534282580304,
376
+ 0.21499383616117984
377
+ ]
378
+ },
379
+ "gru_response_error": {
380
+ "mean": 0.2525889948475027,
381
+ "ci95": [
382
+ 0.21958437609473636,
383
+ 0.28311460593739646
384
+ ]
385
+ },
386
+ "gru_full_nll": {
387
+ "mean": 0.38209832533331256,
388
+ "ci95": [
389
+ 0.3621853450495232,
390
+ 0.401618977606335
391
+ ]
392
+ },
393
+ "gru_choice_nll": {
394
+ "mean": 0.4152231739529124,
395
+ "ci95": [
396
+ 0.3927392942739844,
397
+ 0.43803218971073554
398
+ ]
399
+ },
400
+ "transformer_intact_nll": {
401
+ "mean": 0.46183476641842724,
402
+ "ci95": [
403
+ 0.4346699471273844,
404
+ 0.48976248652904897
405
+ ]
406
+ },
407
+ "transformer_donor_nll": {
408
+ "mean": 0.46657601467364657,
409
+ "ci95": [
410
+ 0.43921089645020744,
411
+ 0.49468999289676974
412
+ ]
413
+ },
414
+ "transformer_delta_nll": {
415
+ "mean": 0.004741248255219306,
416
+ "ci95": [
417
+ 0.002986597877574243,
418
+ 0.0064506400055594154
419
+ ]
420
+ },
421
+ "transformer_response_error": {
422
+ "mean": 0.2899770411458139,
423
+ "ci95": [
424
+ 0.2515092936878267,
425
+ 0.3255082649280953
426
+ ]
427
+ },
428
+ "transformer_full_nll": {
429
+ "mean": 0.46183476641842724,
430
+ "ci95": [
431
+ 0.4346699471273844,
432
+ 0.48976248652904897
433
+ ]
434
+ },
435
+ "transformer_choice_nll": {
436
+ "mean": 0.43139459532984037,
437
+ "ci95": [
438
+ 0.4075386498026013,
439
+ 0.455432339096367
440
+ ]
441
+ },
442
+ "oracle_nll": {
443
+ "mean": 0.35094693566561525,
444
+ "ci95": [
445
+ 0.33274790938588833,
446
+ 0.3697309306471904
447
+ ]
448
+ },
449
+ "llama_full_nll": {
450
+ "mean": 0.3712323003892026,
451
+ "ci95": [
452
+ 0.3523632580491595,
453
+ 0.39130260876466777
454
+ ]
455
+ },
456
+ "llama_choice_nll": {
457
+ "mean": 0.41473857945378156,
458
+ "ci95": [
459
+ 0.3941077767099306,
460
+ 0.43661868695294415
461
+ ]
462
+ },
463
+ "llama_reward_predictive_gain": {
464
+ "mean": 0.04350627906457892,
465
+ "ci95": [
466
+ 0.034753753987408084,
467
+ 0.05295621903463085
468
+ ]
469
+ }
470
+ }
471
+ },
472
+ "0.3": {
473
+ "all": {
474
+ "oracle_intact_nll": {
475
+ "mean": 0.35210242939766134,
476
+ "ci95": [
477
+ 0.3336145296849229,
478
+ 0.3708792025434112
479
+ ]
480
+ },
481
+ "oracle_donor_nll": {
482
+ "mean": 1.378354187514451,
483
+ "ci95": [
484
+ 1.2449881842370578,
485
+ 1.5042307859492283
486
+ ]
487
+ },
488
+ "oracle_delta_nll": {
489
+ "mean": 1.0262517581167896,
490
+ "ci95": [
491
+ 0.8940574977020498,
492
+ 1.1503149535758372
493
+ ]
494
+ },
495
+ "oracle_response_error": {
496
+ "mean": 0.0,
497
+ "ci95": [
498
+ 0.0,
499
+ 0.0
500
+ ]
501
+ },
502
+ "fit_intact_nll": {
503
+ "mean": 0.4896061835111534,
504
+ "ci95": [
505
+ 0.44106822492938974,
506
+ 0.5427661170211016
507
+ ]
508
+ },
509
+ "fit_donor_nll": {
510
+ "mean": 1.394704423822904,
511
+ "ci95": [
512
+ 1.2626838485597622,
513
+ 1.5143586885106135
514
+ ]
515
+ },
516
+ "fit_delta_nll": {
517
+ "mean": 0.9050982403117507,
518
+ "ci95": [
519
+ 0.7629880382170984,
520
+ 1.0391878715388374
521
+ ]
522
+ },
523
+ "fit_response_error": {
524
+ "mean": 0.057430800000000004,
525
+ "ci95": [
526
+ 0.04407462,
527
+ 0.07232642999999998
528
+ ]
529
+ },
530
+ "rw_intact_nll": {
531
+ "mean": 1.2242914606985695,
532
+ "ci95": [
533
+ 1.1950203432111972,
534
+ 1.2539018142430842
535
+ ]
536
+ },
537
+ "rw_donor_nll": {
538
+ "mean": 1.437007960043988,
539
+ "ci95": [
540
+ 1.4319140422608279,
541
+ 1.4427214624270381
542
+ ]
543
+ },
544
+ "rw_delta_nll": {
545
+ "mean": 0.2127164993454183,
546
+ "ci95": [
547
+ 0.18326928850535126,
548
+ 0.2417489239295584
549
+ ]
550
+ },
551
+ "rw_response_error": {
552
+ "mean": 0.42784062991079863,
553
+ "ci95": [
554
+ 0.4023893697322823,
555
+ 0.4508794729227661
556
+ ]
557
+ },
558
+ "baseline_uniform_nll": {
559
+ "mean": 1.3862943611198908,
560
+ "ci95": [
561
+ 1.3862943611198908,
562
+ 1.3862943611198908
563
+ ]
564
+ },
565
+ "baseline_static_arm_frequency_nll": {
566
+ "mean": 1.386206787122509,
567
+ "ci95": [
568
+ 1.3837822901485612,
569
+ 1.388634069972134
570
+ ]
571
+ },
572
+ "baseline_static_stay_probability_nll": {
573
+ "mean": 1.2441673948641805,
574
+ "ci95": [
575
+ 1.1866701314025636,
576
+ 1.2990698518091786
577
+ ]
578
+ },
579
+ "baseline_static_transition_nll": {
580
+ "mean": 1.2435886505660323,
581
+ "ci95": [
582
+ 1.1855407213032967,
583
+ 1.298007934582763
584
+ ]
585
+ },
586
+ "baseline_train_selected_lag_nll": {
587
+ "mean": 0.6476650184453606,
588
+ "ci95": [
589
+ 0.6191878596090458,
590
+ 0.6759260018489243
591
+ ]
592
+ },
593
+ "gru_intact_nll": {
594
+ "mean": 0.40672615669395856,
595
+ "ci95": [
596
+ 0.38528014426506296,
597
+ 0.4276997376507525
598
+ ]
599
+ },
600
+ "gru_donor_nll": {
601
+ "mean": 0.5987031797002234,
602
+ "ci95": [
603
+ 0.5609218736803137,
604
+ 0.6346544145602319
605
+ ]
606
+ },
607
+ "gru_delta_nll": {
608
+ "mean": 0.19197702300626474,
609
+ "ci95": [
610
+ 0.16490749850116282,
611
+ 0.21831516029137363
612
+ ]
613
+ },
614
+ "gru_response_error": {
615
+ "mean": 0.27157453789534314,
616
+ "ci95": [
617
+ 0.23941685639378218,
618
+ 0.3015227994090808
619
+ ]
620
+ },
621
+ "gru_full_nll": {
622
+ "mean": 0.40672615669395856,
623
+ "ci95": [
624
+ 0.38528014426506296,
625
+ 0.4276997376507525
626
+ ]
627
+ },
628
+ "gru_choice_nll": {
629
+ "mean": 0.44105740624071493,
630
+ "ci95": [
631
+ 0.4182955649157589,
632
+ 0.4642907047032306
633
+ ]
634
+ },
635
+ "transformer_intact_nll": {
636
+ "mean": 0.49686909495512305,
637
+ "ci95": [
638
+ 0.4686865709007146,
639
+ 0.5247881967408486
640
+ ]
641
+ },
642
+ "transformer_donor_nll": {
643
+ "mean": 0.5009551909940255,
644
+ "ci95": [
645
+ 0.47245381303420375,
646
+ 0.5286070914975086
647
+ ]
648
+ },
649
+ "transformer_delta_nll": {
650
+ "mean": 0.004086096038902498,
651
+ "ci95": [
652
+ 0.0025259870376069258,
653
+ 0.005639122733196082
654
+ ]
655
+ },
656
+ "transformer_response_error": {
657
+ "mean": 0.30301871779365386,
658
+ "ci95": [
659
+ 0.2655893755823427,
660
+ 0.3382120522968002
661
+ ]
662
+ },
663
+ "transformer_full_nll": {
664
+ "mean": 0.49686909495512305,
665
+ "ci95": [
666
+ 0.4686865709007146,
667
+ 0.5247881967408486
668
+ ]
669
+ },
670
+ "transformer_choice_nll": {
671
+ "mean": 0.4584787729801795,
672
+ "ci95": [
673
+ 0.43331636168871257,
674
+ 0.48349471569884783
675
+ ]
676
+ },
677
+ "oracle_nll": {
678
+ "mean": 0.35210242939766134,
679
+ "ci95": [
680
+ 0.3336145296849229,
681
+ 0.3708792025434112
682
+ ]
683
+ },
684
+ "llama_full_nll": {
685
+ "mean": 0.38488092391922707,
686
+ "ci95": [
687
+ 0.36633522118137063,
688
+ 0.40466352992523025
689
+ ]
690
+ },
691
+ "llama_choice_nll": {
692
+ "mean": 0.4386108408841783,
693
+ "ci95": [
694
+ 0.41715411393602325,
695
+ 0.4612097146793197
696
+ ]
697
+ },
698
+ "llama_reward_predictive_gain": {
699
+ "mean": 0.053729916964951246,
700
+ "ci95": [
701
+ 0.04448048185789844,
702
+ 0.06321365515106948
703
+ ]
704
+ }
705
+ }
706
+ },
707
+ "0.5": {
708
+ "all": {
709
+ "oracle_intact_nll": {
710
+ "mean": 0.35296904969669596,
711
+ "ci95": [
712
+ 0.33534777028299206,
713
+ 0.3708792025434113
714
+ ]
715
+ },
716
+ "oracle_donor_nll": {
717
+ "mean": 1.7604470773588121,
718
+ "ci95": [
719
+ 1.639070765918563,
720
+ 1.8773859317762132
721
+ ]
722
+ },
723
+ "oracle_delta_nll": {
724
+ "mean": 1.4074780276621164,
725
+ "ci95": [
726
+ 1.2866119389229238,
727
+ 1.5210973652424233
728
+ ]
729
+ },
730
+ "oracle_response_error": {
731
+ "mean": 0.0,
732
+ "ci95": [
733
+ 0.0,
734
+ 0.0
735
+ ]
736
+ },
737
+ "fit_intact_nll": {
738
+ "mean": 0.4728515243964842,
739
+ "ci95": [
740
+ 0.4266173314429875,
741
+ 0.5257225844734209
742
+ ]
743
+ },
744
+ "fit_donor_nll": {
745
+ "mean": 1.802853697324906,
746
+ "ci95": [
747
+ 1.6913857446369527,
748
+ 1.9186825555759597
749
+ ]
750
+ },
751
+ "fit_delta_nll": {
752
+ "mean": 1.3300021729284217,
753
+ "ci95": [
754
+ 1.2136862018509522,
755
+ 1.4516074165643327
756
+ ]
757
+ },
758
+ "fit_response_error": {
759
+ "mean": 0.0719568,
760
+ "ci95": [
761
+ 0.05918301,
762
+ 0.08561060999999998
763
+ ]
764
+ },
765
+ "rw_intact_nll": {
766
+ "mean": 0.84433300429568,
767
+ "ci95": [
768
+ 0.8072664182478049,
769
+ 0.8808272133078769
770
+ ]
771
+ },
772
+ "rw_donor_nll": {
773
+ "mean": 1.54841550954771,
774
+ "ci95": [
775
+ 1.5319224623099645,
776
+ 1.5647956943133168
777
+ ]
778
+ },
779
+ "rw_delta_nll": {
780
+ "mean": 0.7040825052520303,
781
+ "ci95": [
782
+ 0.6643037131121271,
783
+ 0.7440399554702459
784
+ ]
785
+ },
786
+ "rw_response_error": {
787
+ "mean": 0.5124604277504972,
788
+ "ci95": [
789
+ 0.5022777914010076,
790
+ 0.5226934129920436
791
+ ]
792
+ },
793
+ "baseline_uniform_nll": {
794
+ "mean": 1.3862943611198908,
795
+ "ci95": [
796
+ 1.3862943611198908,
797
+ 1.3862943611198908
798
+ ]
799
+ },
800
+ "baseline_static_arm_frequency_nll": {
801
+ "mean": 1.3891646412353267,
802
+ "ci95": [
803
+ 1.3853338750081559,
804
+ 1.3927336855198238
805
+ ]
806
+ },
807
+ "baseline_static_stay_probability_nll": {
808
+ "mean": 1.1051973320006765,
809
+ "ci95": [
810
+ 1.0432267582265229,
811
+ 1.1693076487207807
812
+ ]
813
+ },
814
+ "baseline_static_transition_nll": {
815
+ "mean": 1.1047703074936117,
816
+ "ci95": [
817
+ 1.0425278413785393,
818
+ 1.169084269758087
819
+ ]
820
+ },
821
+ "baseline_train_selected_lag_nll": {
822
+ "mean": 0.6716050179152989,
823
+ "ci95": [
824
+ 0.6445413126030036,
825
+ 0.6980486307559435
826
+ ]
827
+ },
828
+ "gru_intact_nll": {
829
+ "mean": 0.4239003927002035,
830
+ "ci95": [
831
+ 0.4039072294108178,
832
+ 0.4438283940250908
833
+ ]
834
+ },
835
+ "gru_donor_nll": {
836
+ "mean": 0.7939825358098876,
837
+ "ci95": [
838
+ 0.7399255289150708,
839
+ 0.8474773464752646
840
+ ]
841
+ },
842
+ "gru_delta_nll": {
843
+ "mean": 0.3700821431096841,
844
+ "ci95": [
845
+ 0.3228127236933769,
846
+ 0.4182303060750939
847
+ ]
848
+ },
849
+ "gru_response_error": {
850
+ "mean": 0.34638604662118166,
851
+ "ci95": [
852
+ 0.32069860146358337,
853
+ 0.37095656402907423
854
+ ]
855
+ },
856
+ "gru_full_nll": {
857
+ "mean": 0.4239003927002035,
858
+ "ci95": [
859
+ 0.4039072294108178,
860
+ 0.4438283940250908
861
+ ]
862
+ },
863
+ "gru_choice_nll": {
864
+ "mean": 0.48624380729834094,
865
+ "ci95": [
866
+ 0.46312726970635076,
867
+ 0.5085444523779327
868
+ ]
869
+ },
870
+ "transformer_intact_nll": {
871
+ "mean": 0.5602828206511774,
872
+ "ci95": [
873
+ 0.5318364537874596,
874
+ 0.5885385491709137
875
+ ]
876
+ },
877
+ "transformer_donor_nll": {
878
+ "mean": 0.582972208854322,
879
+ "ci95": [
880
+ 0.5543366621717734,
881
+ 0.6111613082920802
882
+ ]
883
+ },
884
+ "transformer_delta_nll": {
885
+ "mean": 0.022689388203144655,
886
+ "ci95": [
887
+ 0.019269561739748916,
888
+ 0.02602509734763524
889
+ ]
890
+ },
891
+ "transformer_response_error": {
892
+ "mean": 0.42997184065303506,
893
+ "ci95": [
894
+ 0.3974607780017984,
895
+ 0.46170873135830315
896
+ ]
897
+ },
898
+ "transformer_full_nll": {
899
+ "mean": 0.5602828206511774,
900
+ "ci95": [
901
+ 0.5318364537874596,
902
+ 0.5885385491709137
903
+ ]
904
+ },
905
+ "transformer_choice_nll": {
906
+ "mean": 0.5113768472994935,
907
+ "ci95": [
908
+ 0.48676175148961504,
909
+ 0.5349671850635938
910
+ ]
911
+ },
912
+ "oracle_nll": {
913
+ "mean": 0.35296904969669596,
914
+ "ci95": [
915
+ 0.33534777028299206,
916
+ 0.3708792025434113
917
+ ]
918
+ },
919
+ "llama_full_nll": {
920
+ "mean": 0.3987451196263668,
921
+ "ci95": [
922
+ 0.3795272060539283,
923
+ 0.4174766444148147
924
+ ]
925
+ },
926
+ "llama_choice_nll": {
927
+ "mean": 0.4839780912141663,
928
+ "ci95": [
929
+ 0.4618802124913104,
930
+ 0.5071861501510945
931
+ ]
932
+ },
933
+ "llama_reward_predictive_gain": {
934
+ "mean": 0.08523297158779956,
935
+ "ci95": [
936
+ 0.07585017583638187,
937
+ 0.09465791068877909
938
+ ]
939
+ }
940
+ }
941
+ },
942
+ "0.7": {
943
+ "all": {
944
+ "oracle_intact_nll": {
945
+ "mean": 0.35354679656271903,
946
+ "ci95": [
947
+ 0.3367921374480498,
948
+ 0.3700125822443766
949
+ ]
950
+ },
951
+ "oracle_donor_nll": {
952
+ "mean": 2.2652967325614264,
953
+ "ci95": [
954
+ 2.179960270623697,
955
+ 2.350091917904051
956
+ ]
957
+ },
958
+ "oracle_delta_nll": {
959
+ "mean": 1.9117499359987074,
960
+ "ci95": [
961
+ 1.8224829899130648,
962
+ 1.995500843880995
963
+ ]
964
+ },
965
+ "oracle_response_error": {
966
+ "mean": 0.0,
967
+ "ci95": [
968
+ 0.0,
969
+ 0.0
970
+ ]
971
+ },
972
+ "fit_intact_nll": {
973
+ "mean": 0.4933615381403033,
974
+ "ci95": [
975
+ 0.44049769989919196,
976
+ 0.5554765480736092
977
+ ]
978
+ },
979
+ "fit_donor_nll": {
980
+ "mean": 2.2282776021209982,
981
+ "ci95": [
982
+ 2.1372589997559315,
983
+ 2.317044797167531
984
+ ]
985
+ },
986
+ "fit_delta_nll": {
987
+ "mean": 1.7349160639806946,
988
+ "ci95": [
989
+ 1.6240005817333756,
990
+ 1.8365713472410372
991
+ ]
992
+ },
993
+ "fit_response_error": {
994
+ "mean": 0.10029600000000001,
995
+ "ci95": [
996
+ 0.08329257,
997
+ 0.12025655999999998
998
+ ]
999
+ },
1000
+ "rw_intact_nll": {
1001
+ "mean": 0.6367851249131584,
1002
+ "ci95": [
1003
+ 0.6059779823696988,
1004
+ 0.6690869970074427
1005
+ ]
1006
+ },
1007
+ "rw_donor_nll": {
1008
+ "mean": 1.7542504882058652,
1009
+ "ci95": [
1010
+ 1.7245179652925853,
1011
+ 1.784242717269756
1012
+ ]
1013
+ },
1014
+ "rw_delta_nll": {
1015
+ "mean": 1.1174653632927065,
1016
+ "ci95": [
1017
+ 1.073613864337225,
1018
+ 1.1604707393076024
1019
+ ]
1020
+ },
1021
+ "rw_response_error": {
1022
+ "mean": 0.46508049211239433,
1023
+ "ci95": [
1024
+ 0.45716462662636775,
1025
+ 0.47281200417128716
1026
+ ]
1027
+ },
1028
+ "baseline_uniform_nll": {
1029
+ "mean": 1.3862943611198908,
1030
+ "ci95": [
1031
+ 1.3862943611198908,
1032
+ 1.3862943611198908
1033
+ ]
1034
+ },
1035
+ "baseline_static_arm_frequency_nll": {
1036
+ "mean": 1.3891729105723136,
1037
+ "ci95": [
1038
+ 1.3851654398489182,
1039
+ 1.3931416025271623
1040
+ ]
1041
+ },
1042
+ "baseline_static_stay_probability_nll": {
1043
+ "mean": 0.8644965987836469,
1044
+ "ci95": [
1045
+ 0.8124792112422624,
1046
+ 0.9181352125107066
1047
+ ]
1048
+ },
1049
+ "baseline_static_transition_nll": {
1050
+ "mean": 0.8647343044135412,
1051
+ "ci95": [
1052
+ 0.8130905471962365,
1053
+ 0.9184358855550762
1054
+ ]
1055
+ },
1056
+ "baseline_train_selected_lag_nll": {
1057
+ "mean": 0.6728193957670667,
1058
+ "ci95": [
1059
+ 0.6455297454957415,
1060
+ 0.6980571926345327
1061
+ ]
1062
+ },
1063
+ "gru_intact_nll": {
1064
+ "mean": 0.43352238996782294,
1065
+ "ci95": [
1066
+ 0.4138113304489863,
1067
+ 0.4525574196692029
1068
+ ]
1069
+ },
1070
+ "gru_donor_nll": {
1071
+ "mean": 0.8305084236511499,
1072
+ "ci95": [
1073
+ 0.7850481193625473,
1074
+ 0.8766347834391328
1075
+ ]
1076
+ },
1077
+ "gru_delta_nll": {
1078
+ "mean": 0.3969860336833268,
1079
+ "ci95": [
1080
+ 0.3603409405940758,
1081
+ 0.4349362723933048
1082
+ ]
1083
+ },
1084
+ "gru_response_error": {
1085
+ "mean": 0.47436250322923085,
1086
+ "ci95": [
1087
+ 0.4557596783350788,
1088
+ 0.4917105428630023
1089
+ ]
1090
+ },
1091
+ "gru_full_nll": {
1092
+ "mean": 0.43352238996782294,
1093
+ "ci95": [
1094
+ 0.4138113304489863,
1095
+ 0.4525574196692029
1096
+ ]
1097
+ },
1098
+ "gru_choice_nll": {
1099
+ "mean": 0.495903679754697,
1100
+ "ci95": [
1101
+ 0.47215593645114573,
1102
+ 0.5197811366598344
1103
+ ]
1104
+ },
1105
+ "transformer_intact_nll": {
1106
+ "mean": 0.5486615970356825,
1107
+ "ci95": [
1108
+ 0.5204271245068792,
1109
+ 0.5765164960739434
1110
+ ]
1111
+ },
1112
+ "transformer_donor_nll": {
1113
+ "mean": 0.5974257733244831,
1114
+ "ci95": [
1115
+ 0.5663911416258028,
1116
+ 0.6283860161809668
1117
+ ]
1118
+ },
1119
+ "transformer_delta_nll": {
1120
+ "mean": 0.04876417628880067,
1121
+ "ci95": [
1122
+ 0.0436706073986821,
1123
+ 0.05418430119006499
1124
+ ]
1125
+ },
1126
+ "transformer_response_error": {
1127
+ "mean": 0.5682790094067541,
1128
+ "ci95": [
1129
+ 0.5445340336128616,
1130
+ 0.5916767173504924
1131
+ ]
1132
+ },
1133
+ "transformer_full_nll": {
1134
+ "mean": 0.5486615970356825,
1135
+ "ci95": [
1136
+ 0.5204271245068792,
1137
+ 0.5765164960739434
1138
+ ]
1139
+ },
1140
+ "transformer_choice_nll": {
1141
+ "mean": 0.5262255678096864,
1142
+ "ci95": [
1143
+ 0.49873070844732226,
1144
+ 0.5539373510890337
1145
+ ]
1146
+ },
1147
+ "oracle_nll": {
1148
+ "mean": 0.35354679656271903,
1149
+ "ci95": [
1150
+ 0.3367921374480498,
1151
+ 0.3700125822443766
1152
+ ]
1153
+ },
1154
+ "llama_full_nll": {
1155
+ "mean": 0.3974785488950231,
1156
+ "ci95": [
1157
+ 0.380080654508458,
1158
+ 0.4144363998542966
1159
+ ]
1160
+ },
1161
+ "llama_choice_nll": {
1162
+ "mean": 0.49387579370840684,
1163
+ "ci95": [
1164
+ 0.4712477815121117,
1165
+ 0.517045396465575
1166
+ ]
1167
+ },
1168
+ "llama_reward_predictive_gain": {
1169
+ "mean": 0.09639724481338369,
1170
+ "ci95": [
1171
+ 0.08522192019083413,
1172
+ 0.10836390890938048
1173
+ ]
1174
+ }
1175
+ }
1176
+ },
1177
+ "0.9": {
1178
+ "all": {
1179
+ "oracle_intact_nll": {
1180
+ "mean": 0.3538356699957305,
1181
+ "ci95": [
1182
+ 0.3367921374480498,
1183
+ 0.370597550946225
1184
+ ]
1185
+ },
1186
+ "oracle_donor_nll": {
1187
+ "mean": 2.41985846289425,
1188
+ "ci95": [
1189
+ 2.3698107795332097,
1190
+ 2.4667747582414448
1191
+ ]
1192
+ },
1193
+ "oracle_delta_nll": {
1194
+ "mean": 2.0660227928985195,
1195
+ "ci95": [
1196
+ 2.016492554074361,
1197
+ 2.113377092588518
1198
+ ]
1199
+ },
1200
+ "oracle_response_error": {
1201
+ "mean": 0.0,
1202
+ "ci95": [
1203
+ 0.0,
1204
+ 0.0
1205
+ ]
1206
+ },
1207
+ "fit_intact_nll": {
1208
+ "mean": 0.4725626509634726,
1209
+ "ci95": [
1210
+ 0.42633567984580123,
1211
+ 0.5251448376073978
1212
+ ]
1213
+ },
1214
+ "fit_donor_nll": {
1215
+ "mean": 2.3385839225164538,
1216
+ "ci95": [
1217
+ 2.2559635930326145,
1218
+ 2.4107492002760225
1219
+ ]
1220
+ },
1221
+ "fit_delta_nll": {
1222
+ "mean": 1.8660212715529811,
1223
+ "ci95": [
1224
+ 1.777752383178393,
1225
+ 1.9485531335479598
1226
+ ]
1227
+ },
1228
+ "fit_response_error": {
1229
+ "mean": 0.10665,
1230
+ "ci95": [
1231
+ 0.08918568000000002,
1232
+ 0.12582224999999997
1233
+ ]
1234
+ },
1235
+ "rw_intact_nll": {
1236
+ "mean": 0.5646761974613452,
1237
+ "ci95": [
1238
+ 0.5368416569344175,
1239
+ 0.592434864108435
1240
+ ]
1241
+ },
1242
+ "rw_donor_nll": {
1243
+ "mean": 1.8486147261739798,
1244
+ "ci95": [
1245
+ 1.8127806587170125,
1246
+ 1.8841942795764481
1247
+ ]
1248
+ },
1249
+ "rw_delta_nll": {
1250
+ "mean": 1.2839385287126341,
1251
+ "ci95": [
1252
+ 1.239298381422582,
1253
+ 1.3271210708591765
1254
+ ]
1255
+ },
1256
+ "rw_response_error": {
1257
+ "mean": 0.4112662210155444,
1258
+ "ci95": [
1259
+ 0.403177579536395,
1260
+ 0.41939179149765365
1261
+ ]
1262
+ },
1263
+ "baseline_uniform_nll": {
1264
+ "mean": 1.3862943611198908,
1265
+ "ci95": [
1266
+ 1.3862943611198908,
1267
+ 1.3862943611198908
1268
+ ]
1269
+ },
1270
+ "baseline_static_arm_frequency_nll": {
1271
+ "mean": 1.3919394249096622,
1272
+ "ci95": [
1273
+ 1.3865122093252278,
1274
+ 1.3975226755075831
1275
+ ]
1276
+ },
1277
+ "baseline_static_stay_probability_nll": {
1278
+ "mean": 0.6817777197803314,
1279
+ "ci95": [
1280
+ 0.6509234221147523,
1281
+ 0.7140527053831514
1282
+ ]
1283
+ },
1284
+ "baseline_static_transition_nll": {
1285
+ "mean": 0.6825401617375507,
1286
+ "ci95": [
1287
+ 0.6514550700562811,
1288
+ 0.7141958725164214
1289
+ ]
1290
+ },
1291
+ "baseline_train_selected_lag_nll": {
1292
+ "mean": 0.675262574311137,
1293
+ "ci95": [
1294
+ 0.6479264696313185,
1295
+ 0.701776695012611
1296
+ ]
1297
+ },
1298
+ "gru_intact_nll": {
1299
+ "mean": 0.42817653542883566,
1300
+ "ci95": [
1301
+ 0.40763914878664675,
1302
+ 0.44783797596751024
1303
+ ]
1304
+ },
1305
+ "gru_donor_nll": {
1306
+ "mean": 0.7773886712318161,
1307
+ "ci95": [
1308
+ 0.750643722699596,
1309
+ 0.8039591114987192
1310
+ ]
1311
+ },
1312
+ "gru_delta_nll": {
1313
+ "mean": 0.3492121358029805,
1314
+ "ci95": [
1315
+ 0.3338261728956548,
1316
+ 0.36406010816886447
1317
+ ]
1318
+ },
1319
+ "gru_response_error": {
1320
+ "mean": 0.5252097769509073,
1321
+ "ci95": [
1322
+ 0.5128079189424689,
1323
+ 0.5365733743816665
1324
+ ]
1325
+ },
1326
+ "gru_full_nll": {
1327
+ "mean": 0.42817653542883566,
1328
+ "ci95": [
1329
+ 0.40763914878664675,
1330
+ 0.44783797596751024
1331
+ ]
1332
+ },
1333
+ "gru_choice_nll": {
1334
+ "mean": 0.48175588727990154,
1335
+ "ci95": [
1336
+ 0.45739784345259715,
1337
+ 0.5064396067255538
1338
+ ]
1339
+ },
1340
+ "transformer_intact_nll": {
1341
+ "mean": 0.47869842622950065,
1342
+ "ci95": [
1343
+ 0.45423021619117987,
1344
+ 0.5030158911592408
1345
+ ]
1346
+ },
1347
+ "transformer_donor_nll": {
1348
+ "mean": 0.5376308068473252,
1349
+ "ci95": [
1350
+ 0.5095779604207772,
1351
+ 0.5667927758341778
1352
+ ]
1353
+ },
1354
+ "transformer_delta_nll": {
1355
+ "mean": 0.058932380617824506,
1356
+ "ci95": [
1357
+ 0.051348951959877306,
1358
+ 0.0668927411508699
1359
+ ]
1360
+ },
1361
+ "transformer_response_error": {
1362
+ "mean": 0.5924415928248987,
1363
+ "ci95": [
1364
+ 0.578484843551241,
1365
+ 0.6054317958210641
1366
+ ]
1367
+ },
1368
+ "transformer_full_nll": {
1369
+ "mean": 0.47869842622950065,
1370
+ "ci95": [
1371
+ 0.45423021619117987,
1372
+ 0.5030158911592408
1373
+ ]
1374
+ },
1375
+ "transformer_choice_nll": {
1376
+ "mean": 0.5006007779105845,
1377
+ "ci95": [
1378
+ 0.47461560580884665,
1379
+ 0.527362045213413
1380
+ ]
1381
+ },
1382
+ "oracle_nll": {
1383
+ "mean": 0.3538356699957305,
1384
+ "ci95": [
1385
+ 0.3367921374480498,
1386
+ 0.370597550946225
1387
+ ]
1388
+ },
1389
+ "llama_full_nll": {
1390
+ "mean": 0.39667870658271903,
1391
+ "ci95": [
1392
+ 0.37927451894877817,
1393
+ 0.4139721763957136
1394
+ ]
1395
+ },
1396
+ "llama_choice_nll": {
1397
+ "mean": 0.48026610224555905,
1398
+ "ci95": [
1399
+ 0.4588816590884829,
1400
+ 0.502987530858407
1401
+ ]
1402
+ },
1403
+ "llama_reward_predictive_gain": {
1404
+ "mean": 0.08358739566284004,
1405
+ "ci95": [
1406
+ 0.07248513776142348,
1407
+ 0.09623014834951416
1408
+ ]
1409
+ }
1410
+ }
1411
+ },
1412
+ "1.0": {
1413
+ "all": {
1414
+ "oracle_intact_nll": {
1415
+ "mean": 0.3555689105937998,
1416
+ "ci95": [
1417
+ 0.3391031249121421,
1418
+ 0.37146417124525966
1419
+ ]
1420
+ },
1421
+ "oracle_donor_nll": {
1422
+ "mean": 2.422530542149607,
1423
+ "ci95": [
1424
+ 2.3792703011889635,
1425
+ 2.4649248849947973
1426
+ ]
1427
+ },
1428
+ "oracle_delta_nll": {
1429
+ "mean": 2.066961631555807,
1430
+ "ci95": [
1431
+ 2.024444517501586,
1432
+ 2.109949609305836
1433
+ ]
1434
+ },
1435
+ "oracle_response_error": {
1436
+ "mean": 0.0,
1437
+ "ci95": [
1438
+ 0.0,
1439
+ 0.0
1440
+ ]
1441
+ },
1442
+ "fit_intact_nll": {
1443
+ "mean": 0.4893173100781419,
1444
+ "ci95": [
1445
+ 0.43846836403228584,
1446
+ 0.5465286934860768
1447
+ ]
1448
+ },
1449
+ "fit_donor_nll": {
1450
+ "mean": 2.3547897221084013,
1451
+ "ci95": [
1452
+ 2.279978002610656,
1453
+ 2.422191115865818
1454
+ ]
1455
+ },
1456
+ "fit_delta_nll": {
1457
+ "mean": 1.865472412030259,
1458
+ "ci95": [
1459
+ 1.7838898603545144,
1460
+ 1.9464302749071163
1461
+ ]
1462
+ },
1463
+ "fit_response_error": {
1464
+ "mean": 0.1060848,
1465
+ "ci95": [
1466
+ 0.08813106,
1467
+ 0.12556844999999997
1468
+ ]
1469
+ },
1470
+ "rw_intact_nll": {
1471
+ "mean": 0.5564620276621879,
1472
+ "ci95": [
1473
+ 0.5303074604317995,
1474
+ 0.5825734385894253
1475
+ ]
1476
+ },
1477
+ "rw_donor_nll": {
1478
+ "mean": 1.843404323107438,
1479
+ "ci95": [
1480
+ 1.808157825799863,
1481
+ 1.8785521841710209
1482
+ ]
1483
+ },
1484
+ "rw_delta_nll": {
1485
+ "mean": 1.2869422954452503,
1486
+ "ci95": [
1487
+ 1.243610022593297,
1488
+ 1.3288379270210136
1489
+ ]
1490
+ },
1491
+ "rw_response_error": {
1492
+ "mean": 0.4067648792552328,
1493
+ "ci95": [
1494
+ 0.39874349032177636,
1495
+ 0.4144070957360606
1496
+ ]
1497
+ },
1498
+ "baseline_uniform_nll": {
1499
+ "mean": 1.3862943611198908,
1500
+ "ci95": [
1501
+ 1.3862943611198908,
1502
+ 1.3862943611198908
1503
+ ]
1504
+ },
1505
+ "baseline_static_arm_frequency_nll": {
1506
+ "mean": 1.3917323312509218,
1507
+ "ci95": [
1508
+ 1.386258633958181,
1509
+ 1.3973493130494838
1510
+ ]
1511
+ },
1512
+ "baseline_static_stay_probability_nll": {
1513
+ "mean": 0.6533980502634067,
1514
+ "ci95": [
1515
+ 0.6265057651892878,
1516
+ 0.6794448641606202
1517
+ ]
1518
+ },
1519
+ "baseline_static_transition_nll": {
1520
+ "mean": 0.6541015689795677,
1521
+ "ci95": [
1522
+ 0.6270122398578049,
1523
+ 0.6799042794329158
1524
+ ]
1525
+ },
1526
+ "baseline_train_selected_lag_nll": {
1527
+ "mean": 0.6533980502634067,
1528
+ "ci95": [
1529
+ 0.6265057651892878,
1530
+ 0.6794448641606202
1531
+ ]
1532
+ },
1533
+ "gru_intact_nll": {
1534
+ "mean": 0.4122413658801879,
1535
+ "ci95": [
1536
+ 0.3924156504261923,
1537
+ 0.4314481299530263
1538
+ ]
1539
+ },
1540
+ "gru_donor_nll": {
1541
+ "mean": 0.8702557120266952,
1542
+ "ci95": [
1543
+ 0.8428284942577027,
1544
+ 0.8977218686793744
1545
+ ]
1546
+ },
1547
+ "gru_delta_nll": {
1548
+ "mean": 0.4580143461465074,
1549
+ "ci95": [
1550
+ 0.43938654936765564,
1551
+ 0.4773695430538984
1552
+ ]
1553
+ },
1554
+ "gru_response_error": {
1555
+ "mean": 0.500764796355698,
1556
+ "ci95": [
1557
+ 0.4902894189621314,
1558
+ 0.5111828144448192
1559
+ ]
1560
+ },
1561
+ "gru_full_nll": {
1562
+ "mean": 0.4122413658801879,
1563
+ "ci95": [
1564
+ 0.3924156504261923,
1565
+ 0.4314481299530263
1566
+ ]
1567
+ },
1568
+ "gru_choice_nll": {
1569
+ "mean": 0.47484978552653206,
1570
+ "ci95": [
1571
+ 0.45096606361704156,
1572
+ 0.4981496893958654
1573
+ ]
1574
+ },
1575
+ "transformer_intact_nll": {
1576
+ "mean": 0.4626244695171602,
1577
+ "ci95": [
1578
+ 0.43952595224950564,
1579
+ 0.48491468249785163
1580
+ ]
1581
+ },
1582
+ "transformer_donor_nll": {
1583
+ "mean": 0.5467612604003217,
1584
+ "ci95": [
1585
+ 0.516834423241357,
1586
+ 0.5756310228409485
1587
+ ]
1588
+ },
1589
+ "transformer_delta_nll": {
1590
+ "mean": 0.08413679088316148,
1591
+ "ci95": [
1592
+ 0.07302276167256361,
1593
+ 0.09550627897550838
1594
+ ]
1595
+ },
1596
+ "transformer_response_error": {
1597
+ "mean": 0.5924755479283029,
1598
+ "ci95": [
1599
+ 0.5801959963155888,
1600
+ 0.6043576856168339
1601
+ ]
1602
+ },
1603
+ "transformer_full_nll": {
1604
+ "mean": 0.4626244695171602,
1605
+ "ci95": [
1606
+ 0.43952595224950564,
1607
+ 0.48491468249785163
1608
+ ]
1609
+ },
1610
+ "transformer_choice_nll": {
1611
+ "mean": 0.4868411339728783,
1612
+ "ci95": [
1613
+ 0.4622481042139416,
1614
+ 0.5110058921499612
1615
+ ]
1616
+ },
1617
+ "oracle_nll": {
1618
+ "mean": 0.3555689105937998,
1619
+ "ci95": [
1620
+ 0.3391031249121421,
1621
+ 0.37146417124525966
1622
+ ]
1623
+ },
1624
+ "llama_full_nll": {
1625
+ "mean": 0.39184374347422907,
1626
+ "ci95": [
1627
+ 0.37459694013477624,
1628
+ 0.4088950564098682
1629
+ ]
1630
+ },
1631
+ "llama_choice_nll": {
1632
+ "mean": 0.4751047001399281,
1633
+ "ci95": [
1634
+ 0.4528385337205593,
1635
+ 0.49719918769813887
1636
+ ]
1637
+ },
1638
+ "llama_reward_predictive_gain": {
1639
+ "mean": 0.083260956665699,
1640
+ "ci95": [
1641
+ 0.0715406502095134,
1642
+ 0.09523279384042423
1643
+ ]
1644
+ }
1645
+ }
1646
+ }
1647
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/restless/B_summary.json ADDED
@@ -0,0 +1,2515 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "oracle_intact_nll": {
4
+ "mean": 0.7622831832040385,
5
+ "ci95": [
6
+ 0.7077135251837541,
7
+ 0.8159035892782495
8
+ ]
9
+ },
10
+ "oracle_donor_nll": {
11
+ "mean": 0.7622831832040387,
12
+ "ci95": [
13
+ 0.7077135251837543,
14
+ 0.8159035892782496
15
+ ]
16
+ },
17
+ "oracle_delta_nll": {
18
+ "mean": 1.121325254871408e-16,
19
+ "ci95": [
20
+ 9.037215420448774e-17,
21
+ 1.3511691765444312e-16
22
+ ]
23
+ },
24
+ "oracle_response_error": {
25
+ "mean": 0.0,
26
+ "ci95": [
27
+ 0.0,
28
+ 0.0
29
+ ]
30
+ },
31
+ "fit_intact_nll": {
32
+ "mean": 0.8483674662414769,
33
+ "ci95": [
34
+ 0.7885645365311226,
35
+ 0.910060043019351
36
+ ]
37
+ },
38
+ "fit_donor_nll": {
39
+ "mean": 0.7847864236356374,
40
+ "ci95": [
41
+ 0.7319429521168608,
42
+ 0.8362034846932674
43
+ ]
44
+ },
45
+ "fit_delta_nll": {
46
+ "mean": -0.06358104260583938,
47
+ "ci95": [
48
+ -0.10433205670898575,
49
+ -0.029146246115490267
50
+ ]
51
+ },
52
+ "fit_response_error": {
53
+ "mean": 0.0281268,
54
+ "ci95": [
55
+ 0.016461270000000004,
56
+ 0.041578199999999996
57
+ ]
58
+ },
59
+ "rw_intact_nll": {
60
+ "mean": 1.3862943611198908,
61
+ "ci95": [
62
+ 1.3862943611198908,
63
+ 1.3862943611198908
64
+ ]
65
+ },
66
+ "rw_donor_nll": {
67
+ "mean": 1.3862943611198908,
68
+ "ci95": [
69
+ 1.3862943611198908,
70
+ 1.3862943611198908
71
+ ]
72
+ },
73
+ "rw_delta_nll": {
74
+ "mean": -2.220446049250313e-16,
75
+ "ci95": [
76
+ -2.220446049250313e-16,
77
+ -2.220446049250313e-16
78
+ ]
79
+ },
80
+ "rw_response_error": {
81
+ "mean": 0.0,
82
+ "ci95": [
83
+ 0.0,
84
+ 0.0
85
+ ]
86
+ },
87
+ "gru_intact_nll": {
88
+ "mean": 0.7630717979613043,
89
+ "ci95": [
90
+ 0.7078486512600141,
91
+ 0.8164198409461291
92
+ ]
93
+ },
94
+ "gru_donor_nll": {
95
+ "mean": 0.763411439820356,
96
+ "ci95": [
97
+ 0.7082909148065478,
98
+ 0.8165612710825744
99
+ ]
100
+ },
101
+ "gru_delta_nll": {
102
+ "mean": 0.00033964185905163214,
103
+ "ci95": [
104
+ -0.00017061420642414948,
105
+ 0.0008591160613694002
106
+ ]
107
+ },
108
+ "gru_response_error": {
109
+ "mean": 0.00837545404664359,
110
+ "ci95": [
111
+ 0.007792406016524281,
112
+ 0.008952487292638861
113
+ ]
114
+ },
115
+ "gru_full_nll": {
116
+ "mean": 0.7630717979613043,
117
+ "ci95": [
118
+ 0.7078486512600141,
119
+ 0.8164198409461291
120
+ ]
121
+ },
122
+ "gru_choice_nll": {
123
+ "mean": 0.7624148767056697,
124
+ "ci95": [
125
+ 0.707409101648277,
126
+ 0.8156433074385794
127
+ ]
128
+ },
129
+ "transformer_intact_nll": {
130
+ "mean": 0.8271710782106564,
131
+ "ci95": [
132
+ 0.7643411545292396,
133
+ 0.8877736162777501
134
+ ]
135
+ },
136
+ "transformer_donor_nll": {
137
+ "mean": 0.8269989130554898,
138
+ "ci95": [
139
+ 0.7641808156609311,
140
+ 0.8876039408191849
141
+ ]
142
+ },
143
+ "transformer_delta_nll": {
144
+ "mean": -0.0001721651551665042,
145
+ "ci95": [
146
+ -0.0005563447354454075,
147
+ 0.00019730675805899794
148
+ ]
149
+ },
150
+ "transformer_response_error": {
151
+ "mean": 0.008717824416243138,
152
+ "ci95": [
153
+ 0.008404667781309327,
154
+ 0.009017450141388649
155
+ ]
156
+ },
157
+ "transformer_full_nll": {
158
+ "mean": 0.8271710782106564,
159
+ "ci95": [
160
+ 0.7643411545292396,
161
+ 0.8877736162777501
162
+ ]
163
+ },
164
+ "transformer_choice_nll": {
165
+ "mean": 0.7662887955148586,
166
+ "ci95": [
167
+ 0.7120669772326466,
168
+ 0.8203644290312895
169
+ ]
170
+ },
171
+ "oracle_nll": {
172
+ "mean": 0.7622831832040385,
173
+ "ci95": [
174
+ 0.7077135251837541,
175
+ 0.8159035892782495
176
+ ]
177
+ },
178
+ "llama_intact_nll": {
179
+ "mean": 0.7630837702473433,
180
+ "ci95": [
181
+ 0.7081551329450082,
182
+ 0.8166046093560824
183
+ ]
184
+ },
185
+ "llama_donor_nll": {
186
+ "mean": 0.7628997326845911,
187
+ "ci95": [
188
+ 0.7079717463611989,
189
+ 0.8165061102335179
190
+ ]
191
+ },
192
+ "llama_delta_nll": {
193
+ "mean": -0.00018403756275230565,
194
+ "ci95": [
195
+ -0.0006521315657943006,
196
+ 0.0002927113655689931
197
+ ]
198
+ },
199
+ "llama_response_error": {
200
+ "mean": 0.010879141125847236,
201
+ "ci95": [
202
+ 0.010226052713296185,
203
+ 0.01153126529138923
204
+ ]
205
+ },
206
+ "llama_full_nll": {
207
+ "mean": 0.7630837702473433,
208
+ "ci95": [
209
+ 0.7081551329450082,
210
+ 0.8166046093560824
211
+ ]
212
+ },
213
+ "llama_choice_nll": {
214
+ "mean": 0.7625497153058011,
215
+ "ci95": [
216
+ 0.7080384854284038,
217
+ 0.8159663827425214
218
+ ]
219
+ },
220
+ "llama_reward_predictive_gain": {
221
+ "mean": -0.0005340549415423553,
222
+ "ci95": [
223
+ -0.0013722305514553377,
224
+ 0.0003064117202134289
225
+ ]
226
+ },
227
+ "llama_full_token_nll": {
228
+ "mean": 0.7631381750106812,
229
+ "ci95": [
230
+ 0.7082111746072769,
231
+ 0.8166580036282539
232
+ ]
233
+ },
234
+ "llama_full_valid_mass": {
235
+ "mean": 0.9999455809593201,
236
+ "ci95": [
237
+ 0.999943196773529,
238
+ 0.9999479055404663
239
+ ]
240
+ },
241
+ "llama_choice_token_nll": {
242
+ "mean": 0.7626761198043823,
243
+ "ci95": [
244
+ 0.7081539928913116,
245
+ 0.816103033721447
246
+ ]
247
+ },
248
+ "llama_choice_valid_mass": {
249
+ "mean": 0.9998735785484314,
250
+ "ci95": [
251
+ 0.9998615980148315,
252
+ 0.9998859778046608
253
+ ]
254
+ },
255
+ "llama_intact_nll_minus_oracle": {
256
+ "mean": 0.0008005870433048315,
257
+ "ci95": [
258
+ 3.824942152617241e-05,
259
+ 0.0015519903808523523
260
+ ]
261
+ },
262
+ "llama_response_error_minus_oracle": {
263
+ "mean": 0.010879141125847236,
264
+ "ci95": [
265
+ 0.010226052713296185,
266
+ 0.01153126529138923
267
+ ]
268
+ },
269
+ "llama_delta_nll_minus_oracle": {
270
+ "mean": -0.00018403756275241779,
271
+ "ci95": [
272
+ -0.0006521315657944107,
273
+ 0.00029271136556889175
274
+ ]
275
+ },
276
+ "llama_intact_nll_minus_fit": {
277
+ "mean": -0.08528369599413346,
278
+ "ci95": [
279
+ -0.13069124610567126,
280
+ -0.04573904278982295
281
+ ]
282
+ },
283
+ "llama_response_error_minus_fit": {
284
+ "mean": -0.017247658874152763,
285
+ "ci95": [
286
+ -0.030879008704263567,
287
+ -0.005567730472547319
288
+ ]
289
+ },
290
+ "llama_delta_nll_minus_fit": {
291
+ "mean": 0.06339700504308708,
292
+ "ci95": [
293
+ 0.02905902894536918,
294
+ 0.10426235788179315
295
+ ]
296
+ },
297
+ "llama_intact_nll_minus_rw": {
298
+ "mean": -0.6232105908725472,
299
+ "ci95": [
300
+ -0.6781392281748824,
301
+ -0.5696897517638083
302
+ ]
303
+ },
304
+ "llama_response_error_minus_rw": {
305
+ "mean": 0.010879141125847236,
306
+ "ci95": [
307
+ 0.010226052713296185,
308
+ 0.01153126529138923
309
+ ]
310
+ },
311
+ "llama_delta_nll_minus_rw": {
312
+ "mean": -0.0001840375627520836,
313
+ "ci95": [
314
+ -0.0006521315657940786,
315
+ 0.00029271136556921517
316
+ ]
317
+ },
318
+ "llama_intact_nll_minus_gru": {
319
+ "mean": 1.1972286039041435e-05,
320
+ "ci95": [
321
+ -0.0009605667537321261,
322
+ 0.0010677800666640076
323
+ ]
324
+ },
325
+ "llama_response_error_minus_gru": {
326
+ "mean": 0.0025036870792036468,
327
+ "ci95": [
328
+ 0.0012667932797940937,
329
+ 0.003735921928850303
330
+ ]
331
+ },
332
+ "llama_delta_nll_minus_gru": {
333
+ "mean": -0.0005236794218039379,
334
+ "ci95": [
335
+ -0.0012084787606479405,
336
+ 0.0001330963581034913
337
+ ]
338
+ },
339
+ "llama_intact_nll_minus_transformer": {
340
+ "mean": -0.06408730796331293,
341
+ "ci95": [
342
+ -0.07248742432984658,
343
+ -0.055475922615317155
344
+ ]
345
+ },
346
+ "llama_response_error_minus_transformer": {
347
+ "mean": 0.0021613167096040985,
348
+ "ci95": [
349
+ 0.0017278048017899063,
350
+ 0.002614769122647078
351
+ ]
352
+ },
353
+ "llama_delta_nll_minus_transformer": {
354
+ "mean": -1.187240758580152e-05,
355
+ "ci95": [
356
+ -0.0006409756710907896,
357
+ 0.0006169215823478406
358
+ ]
359
+ }
360
+ },
361
+ "0.1": {
362
+ "oracle_intact_nll": {
363
+ "mean": 0.35094693566561525,
364
+ "ci95": [
365
+ 0.3330295609830745,
366
+ 0.36944205721417883
367
+ ]
368
+ },
369
+ "oracle_donor_nll": {
370
+ "mean": 1.337810801191282,
371
+ "ci95": [
372
+ 1.208915114289742,
373
+ 1.4646439317811188
374
+ ]
375
+ },
376
+ "oracle_delta_nll": {
377
+ "mean": 0.9868638655256665,
378
+ "ci95": [
379
+ 0.8599440729059259,
380
+ 1.1170908978615977
381
+ ]
382
+ },
383
+ "oracle_response_error": {
384
+ "mean": 0.0,
385
+ "ci95": [
386
+ 0.0,
387
+ 0.0
388
+ ]
389
+ },
390
+ "fit_intact_nll": {
391
+ "mean": 0.44800840915749196,
392
+ "ci95": [
393
+ 0.402944153607692,
394
+ 0.4976946396354764
395
+ ]
396
+ },
397
+ "fit_donor_nll": {
398
+ "mean": 1.3569775534715973,
399
+ "ci95": [
400
+ 1.2320459322545585,
401
+ 1.480940004320882
402
+ ]
403
+ },
404
+ "fit_delta_nll": {
405
+ "mean": 0.9089691443141052,
406
+ "ci95": [
407
+ 0.7735571949474082,
408
+ 1.0473290470646848
409
+ ]
410
+ },
411
+ "fit_response_error": {
412
+ "mean": 0.043848,
413
+ "ci95": [
414
+ 0.031715730000000004,
415
+ 0.05741406
416
+ ]
417
+ },
418
+ "rw_intact_nll": {
419
+ "mean": 1.3303796867230553,
420
+ "ci95": [
421
+ 1.305819348750301,
422
+ 1.3562553147695362
423
+ ]
424
+ },
425
+ "rw_donor_nll": {
426
+ "mean": 1.4153938308975889,
427
+ "ci95": [
428
+ 1.4115843340049306,
429
+ 1.419093293989195
430
+ ]
431
+ },
432
+ "rw_delta_nll": {
433
+ "mean": 0.08501414417453378,
434
+ "ci95": [
435
+ 0.05791561415992192,
436
+ 0.11000932837550137
437
+ ]
438
+ },
439
+ "rw_response_error": {
440
+ "mean": 0.387172420067802,
441
+ "ci95": [
442
+ 0.3594914706588719,
443
+ 0.41428433494295225
444
+ ]
445
+ },
446
+ "gru_intact_nll": {
447
+ "mean": 0.38209832533331256,
448
+ "ci95": [
449
+ 0.3637523851666046,
450
+ 0.4020710538958957
451
+ ]
452
+ },
453
+ "gru_donor_nll": {
454
+ "mean": 0.5700262028629846,
455
+ "ci95": [
456
+ 0.5352831593485384,
457
+ 0.6060507540317192
458
+ ]
459
+ },
460
+ "gru_delta_nll": {
461
+ "mean": 0.18792787752967197,
462
+ "ci95": [
463
+ 0.1619008110088027,
464
+ 0.21494585027512503
465
+ ]
466
+ },
467
+ "gru_response_error": {
468
+ "mean": 0.2525889948475027,
469
+ "ci95": [
470
+ 0.22182542914059922,
471
+ 0.2830708373258202
472
+ ]
473
+ },
474
+ "gru_full_nll": {
475
+ "mean": 0.38209832533331256,
476
+ "ci95": [
477
+ 0.3637523851666046,
478
+ 0.4020710538958957
479
+ ]
480
+ },
481
+ "gru_choice_nll": {
482
+ "mean": 0.4152231739529124,
483
+ "ci95": [
484
+ 0.3940926724709079,
485
+ 0.43765332110398314
486
+ ]
487
+ },
488
+ "transformer_intact_nll": {
489
+ "mean": 0.46183476641842724,
490
+ "ci95": [
491
+ 0.43598920046513306,
492
+ 0.4881816682453971
493
+ ]
494
+ },
495
+ "transformer_donor_nll": {
496
+ "mean": 0.46657601467364657,
497
+ "ci95": [
498
+ 0.4402695947206693,
499
+ 0.4938758784872021
500
+ ]
501
+ },
502
+ "transformer_delta_nll": {
503
+ "mean": 0.004741248255219306,
504
+ "ci95": [
505
+ 0.0029478322002229945,
506
+ 0.006574285309236003
507
+ ]
508
+ },
509
+ "transformer_response_error": {
510
+ "mean": 0.2899770411458139,
511
+ "ci95": [
512
+ 0.25410776555243647,
513
+ 0.3261458611704361
514
+ ]
515
+ },
516
+ "transformer_full_nll": {
517
+ "mean": 0.46183476641842724,
518
+ "ci95": [
519
+ 0.43598920046513306,
520
+ 0.4881816682453971
521
+ ]
522
+ },
523
+ "transformer_choice_nll": {
524
+ "mean": 0.43139459532984037,
525
+ "ci95": [
526
+ 0.40878319672550306,
527
+ 0.45557778469965376
528
+ ]
529
+ },
530
+ "oracle_nll": {
531
+ "mean": 0.35094693566561525,
532
+ "ci95": [
533
+ 0.3330295609830745,
534
+ 0.36944205721417883
535
+ ]
536
+ },
537
+ "llama_intact_nll": {
538
+ "mean": 0.3712323003892026,
539
+ "ci95": [
540
+ 0.3523632580491595,
541
+ 0.39130260876466777
542
+ ]
543
+ },
544
+ "llama_donor_nll": {
545
+ "mean": 0.5106519355880039,
546
+ "ci95": [
547
+ 0.4815514978112045,
548
+ 0.5420601925346038
549
+ ]
550
+ },
551
+ "llama_delta_nll": {
552
+ "mean": 0.13941963519880127,
553
+ "ci95": [
554
+ 0.11743128204144467,
555
+ 0.16146984362915792
556
+ ]
557
+ },
558
+ "llama_response_error": {
559
+ "mean": 0.24759311717611734,
560
+ "ci95": [
561
+ 0.21650006027478727,
562
+ 0.27936766620093734
563
+ ]
564
+ },
565
+ "llama_full_nll": {
566
+ "mean": 0.3712323003892026,
567
+ "ci95": [
568
+ 0.3523632580491595,
569
+ 0.39130260876466777
570
+ ]
571
+ },
572
+ "llama_choice_nll": {
573
+ "mean": 0.41473857945378156,
574
+ "ci95": [
575
+ 0.3941077767099306,
576
+ 0.43661868695294415
577
+ ]
578
+ },
579
+ "llama_reward_predictive_gain": {
580
+ "mean": 0.04350627906457892,
581
+ "ci95": [
582
+ 0.034753753987408084,
583
+ 0.05295621903463085
584
+ ]
585
+ },
586
+ "llama_full_token_nll": {
587
+ "mean": 0.37130677700042725,
588
+ "ci95": [
589
+ 0.3524359978735447,
590
+ 0.3913790702819824
591
+ ]
592
+ },
593
+ "llama_full_valid_mass": {
594
+ "mean": 0.9999255537986755,
595
+ "ci95": [
596
+ 0.9999234676361084,
597
+ 0.9999274015426636
598
+ ]
599
+ },
600
+ "llama_choice_token_nll": {
601
+ "mean": 0.41478368639945984,
602
+ "ci95": [
603
+ 0.39415267780423163,
604
+ 0.43666457384824753
605
+ ]
606
+ },
607
+ "llama_choice_valid_mass": {
608
+ "mean": 0.999954879283905,
609
+ "ci95": [
610
+ 0.9999539852142334,
611
+ 0.9999558329582214
612
+ ]
613
+ },
614
+ "llama_intact_nll_minus_oracle": {
615
+ "mean": 0.02028536472358742,
616
+ "ci95": [
617
+ 0.015426079479963965,
618
+ 0.025733971446694544
619
+ ]
620
+ },
621
+ "llama_response_error_minus_oracle": {
622
+ "mean": 0.24759311717611734,
623
+ "ci95": [
624
+ 0.21650006027478727,
625
+ 0.27936766620093734
626
+ ]
627
+ },
628
+ "llama_delta_nll_minus_oracle": {
629
+ "mean": -0.8474442303268652,
630
+ "ci95": [
631
+ -0.9588021666125863,
632
+ -0.7362386976332812
633
+ ]
634
+ },
635
+ "llama_intact_nll_minus_fit": {
636
+ "mean": -0.07677610876828933,
637
+ "ci95": [
638
+ -0.12330007165065913,
639
+ -0.03701621428406879
640
+ ]
641
+ },
642
+ "llama_response_error_minus_fit": {
643
+ "mean": 0.20374511717611735,
644
+ "ci95": [
645
+ 0.16862424158895456,
646
+ 0.23746889560111065
647
+ ]
648
+ },
649
+ "llama_delta_nll_minus_fit": {
650
+ "mean": -0.7695495091153041,
651
+ "ci95": [
652
+ -0.8910850610064643,
653
+ -0.6507917567803869
654
+ ]
655
+ },
656
+ "llama_intact_nll_minus_rw": {
657
+ "mean": -0.9591473863338527,
658
+ "ci95": [
659
+ -0.9930060283687095,
660
+ -0.9267668431152938
661
+ ]
662
+ },
663
+ "llama_response_error_minus_rw": {
664
+ "mean": -0.13957930289168474,
665
+ "ci95": [
666
+ -0.14647717108841332,
667
+ -0.13343551644272877
668
+ ]
669
+ },
670
+ "llama_delta_nll_minus_rw": {
671
+ "mean": 0.05440549102426749,
672
+ "ci95": [
673
+ 0.03035514347352625,
674
+ 0.07768775762720086
675
+ ]
676
+ },
677
+ "llama_intact_nll_minus_gru": {
678
+ "mean": -0.010866024944109954,
679
+ "ci95": [
680
+ -0.015475602375327214,
681
+ -0.006839348820591253
682
+ ]
683
+ },
684
+ "llama_response_error_minus_gru": {
685
+ "mean": -0.004995877671385343,
686
+ "ci95": [
687
+ -0.007551890162822119,
688
+ -0.002593131304043414
689
+ ]
690
+ },
691
+ "llama_delta_nll_minus_gru": {
692
+ "mean": -0.0485082423308707,
693
+ "ci95": [
694
+ -0.05896645292913483,
695
+ -0.03874847974349387
696
+ ]
697
+ },
698
+ "llama_intact_nll_minus_transformer": {
699
+ "mean": -0.09060246602922467,
700
+ "ci95": [
701
+ -0.10528854609645798,
702
+ -0.07639215601411735
703
+ ]
704
+ },
705
+ "llama_response_error_minus_transformer": {
706
+ "mean": -0.04238392396969661,
707
+ "ci95": [
708
+ -0.048063866941845515,
709
+ -0.03670873025401626
710
+ ]
711
+ },
712
+ "llama_delta_nll_minus_transformer": {
713
+ "mean": 0.13467838694358197,
714
+ "ci95": [
715
+ 0.11342649979568831,
716
+ 0.15636751136571614
717
+ ]
718
+ }
719
+ },
720
+ "0.3": {
721
+ "oracle_intact_nll": {
722
+ "mean": 0.35210242939766134,
723
+ "ci95": [
724
+ 0.3336145296849229,
725
+ 0.37059755094622504
726
+ ]
727
+ },
728
+ "oracle_donor_nll": {
729
+ "mean": 1.378354187514451,
730
+ "ci95": [
731
+ 1.2535565313519716,
732
+ 1.5019367697993256
733
+ ]
734
+ },
735
+ "oracle_delta_nll": {
736
+ "mean": 1.0262517581167896,
737
+ "ci95": [
738
+ 0.8989585965848816,
739
+ 1.150601660458101
740
+ ]
741
+ },
742
+ "oracle_response_error": {
743
+ "mean": 0.0,
744
+ "ci95": [
745
+ 0.0,
746
+ 0.0
747
+ ]
748
+ },
749
+ "fit_intact_nll": {
750
+ "mean": 0.4896061835111534,
751
+ "ci95": [
752
+ 0.4419348452284243,
753
+ 0.5410400982588576
754
+ ]
755
+ },
756
+ "fit_donor_nll": {
757
+ "mean": 1.394704423822904,
758
+ "ci95": [
759
+ 1.275138987715846,
760
+ 1.5163096674588148
761
+ ]
762
+ },
763
+ "fit_delta_nll": {
764
+ "mean": 0.9050982403117507,
765
+ "ci95": [
766
+ 0.7687608126840426,
767
+ 1.0376416764886434
768
+ ]
769
+ },
770
+ "fit_response_error": {
771
+ "mean": 0.057430800000000004,
772
+ "ci95": [
773
+ 0.04508622,
774
+ 0.07118954999999996
775
+ ]
776
+ },
777
+ "rw_intact_nll": {
778
+ "mean": 1.2242914606985695,
779
+ "ci95": [
780
+ 1.1962488598456653,
781
+ 1.2532178613115506
782
+ ]
783
+ },
784
+ "rw_donor_nll": {
785
+ "mean": 1.437007960043988,
786
+ "ci95": [
787
+ 1.4316061250584189,
788
+ 1.4423473286531951
789
+ ]
790
+ },
791
+ "rw_delta_nll": {
792
+ "mean": 0.2127164993454183,
793
+ "ci95": [
794
+ 0.18341734263568935,
795
+ 0.24083285394712275
796
+ ]
797
+ },
798
+ "rw_response_error": {
799
+ "mean": 0.42784062991079863,
800
+ "ci95": [
801
+ 0.40413970498216,
802
+ 0.4510055435782805
803
+ ]
804
+ },
805
+ "gru_intact_nll": {
806
+ "mean": 0.40672615669395856,
807
+ "ci95": [
808
+ 0.38673649117880615,
809
+ 0.428749748746146
810
+ ]
811
+ },
812
+ "gru_donor_nll": {
813
+ "mean": 0.5987031797002234,
814
+ "ci95": [
815
+ 0.5626802621405961,
816
+ 0.634079166840625
817
+ ]
818
+ },
819
+ "gru_delta_nll": {
820
+ "mean": 0.19197702300626474,
821
+ "ci95": [
822
+ 0.1650535683938426,
823
+ 0.21861007858791848
824
+ ]
825
+ },
826
+ "gru_response_error": {
827
+ "mean": 0.27157453789534314,
828
+ "ci95": [
829
+ 0.24094550584495597,
830
+ 0.30222556288931063
831
+ ]
832
+ },
833
+ "gru_full_nll": {
834
+ "mean": 0.40672615669395856,
835
+ "ci95": [
836
+ 0.38673649117880615,
837
+ 0.428749748746146
838
+ ]
839
+ },
840
+ "gru_choice_nll": {
841
+ "mean": 0.44105740624071493,
842
+ "ci95": [
843
+ 0.4193601177588154,
844
+ 0.46436614237781837
845
+ ]
846
+ },
847
+ "transformer_intact_nll": {
848
+ "mean": 0.49686909495512305,
849
+ "ci95": [
850
+ 0.4693766814112672,
851
+ 0.5249696007397026
852
+ ]
853
+ },
854
+ "transformer_donor_nll": {
855
+ "mean": 0.5009551909940255,
856
+ "ci95": [
857
+ 0.47323124604874517,
858
+ 0.529296390832026
859
+ ]
860
+ },
861
+ "transformer_delta_nll": {
862
+ "mean": 0.004086096038902498,
863
+ "ci95": [
864
+ 0.0024780783650246306,
865
+ 0.005708253372472617
866
+ ]
867
+ },
868
+ "transformer_response_error": {
869
+ "mean": 0.30301871779365386,
870
+ "ci95": [
871
+ 0.26703232726302284,
872
+ 0.3384407965443171
873
+ ]
874
+ },
875
+ "transformer_full_nll": {
876
+ "mean": 0.49686909495512305,
877
+ "ci95": [
878
+ 0.4693766814112672,
879
+ 0.5249696007397026
880
+ ]
881
+ },
882
+ "transformer_choice_nll": {
883
+ "mean": 0.4584787729801795,
884
+ "ci95": [
885
+ 0.43448693396867477,
886
+ 0.4833706978723548
887
+ ]
888
+ },
889
+ "oracle_nll": {
890
+ "mean": 0.35210242939766134,
891
+ "ci95": [
892
+ 0.3336145296849229,
893
+ 0.37059755094622504
894
+ ]
895
+ },
896
+ "llama_intact_nll": {
897
+ "mean": 0.38488092391922707,
898
+ "ci95": [
899
+ 0.36633522118137063,
900
+ 0.40466352992523025
901
+ ]
902
+ },
903
+ "llama_donor_nll": {
904
+ "mean": 0.551995735235289,
905
+ "ci95": [
906
+ 0.5191817795797989,
907
+ 0.5843210807748787
908
+ ]
909
+ },
910
+ "llama_delta_nll": {
911
+ "mean": 0.16711481131606185,
912
+ "ci95": [
913
+ 0.1427420959917755,
914
+ 0.19171349973880025
915
+ ]
916
+ },
917
+ "llama_response_error": {
918
+ "mean": 0.2528817587157585,
919
+ "ci95": [
920
+ 0.22244073381508578,
921
+ 0.2827141828920568
922
+ ]
923
+ },
924
+ "llama_full_nll": {
925
+ "mean": 0.38488092391922707,
926
+ "ci95": [
927
+ 0.36633522118137063,
928
+ 0.40466352992523025
929
+ ]
930
+ },
931
+ "llama_choice_nll": {
932
+ "mean": 0.4386108408841783,
933
+ "ci95": [
934
+ 0.41715411393602325,
935
+ 0.4612097146793197
936
+ ]
937
+ },
938
+ "llama_reward_predictive_gain": {
939
+ "mean": 0.053729916964951246,
940
+ "ci95": [
941
+ 0.04448048185789844,
942
+ 0.06321365515106948
943
+ ]
944
+ },
945
+ "llama_full_token_nll": {
946
+ "mean": 0.3849515914916992,
947
+ "ci95": [
948
+ 0.36640542894601824,
949
+ 0.4047354623675346
950
+ ]
951
+ },
952
+ "llama_full_valid_mass": {
953
+ "mean": 0.9999293088912964,
954
+ "ci95": [
955
+ 0.9999277591705322,
956
+ 0.999930739402771
957
+ ]
958
+ },
959
+ "llama_choice_token_nll": {
960
+ "mean": 0.43867120146751404,
961
+ "ci95": [
962
+ 0.4172144539654255,
963
+ 0.4612710237503051
964
+ ]
965
+ },
966
+ "llama_choice_valid_mass": {
967
+ "mean": 0.9999396204948425,
968
+ "ci95": [
969
+ 0.999938428401947,
970
+ 0.9999408721923828
971
+ ]
972
+ },
973
+ "llama_intact_nll_minus_oracle": {
974
+ "mean": 0.03277849452156578,
975
+ "ci95": [
976
+ 0.026452546206126373,
977
+ 0.03949223680997706
978
+ ]
979
+ },
980
+ "llama_response_error_minus_oracle": {
981
+ "mean": 0.2528817587157585,
982
+ "ci95": [
983
+ 0.22244073381508578,
984
+ 0.2827141828920568
985
+ ]
986
+ },
987
+ "llama_delta_nll_minus_oracle": {
988
+ "mean": -0.8591369468007278,
989
+ "ci95": [
990
+ -0.9634673899752468,
991
+ -0.7520830104397563
992
+ ]
993
+ },
994
+ "llama_intact_nll_minus_fit": {
995
+ "mean": -0.10472525959192626,
996
+ "ci95": [
997
+ -0.1545844548718939,
998
+ -0.06215241868717741
999
+ ]
1000
+ },
1001
+ "llama_response_error_minus_fit": {
1002
+ "mean": 0.19545095871575852,
1003
+ "ci95": [
1004
+ 0.16088595533069072,
1005
+ 0.2272515045084239
1006
+ ]
1007
+ },
1008
+ "llama_delta_nll_minus_fit": {
1009
+ "mean": -0.7379834289956889,
1010
+ "ci95": [
1011
+ -0.8501489173382416,
1012
+ -0.6213265675978454
1013
+ ]
1014
+ },
1015
+ "llama_intact_nll_minus_rw": {
1016
+ "mean": -0.8394105367793427,
1017
+ "ci95": [
1018
+ -0.8730704696728944,
1019
+ -0.8074135693521395
1020
+ ]
1021
+ },
1022
+ "llama_response_error_minus_rw": {
1023
+ "mean": -0.17495887119504014,
1024
+ "ci95": [
1025
+ -0.1850637378877384,
1026
+ -0.16571810354122402
1027
+ ]
1028
+ },
1029
+ "llama_delta_nll_minus_rw": {
1030
+ "mean": -0.04560168802935647,
1031
+ "ci95": [
1032
+ -0.07616730155558132,
1033
+ -0.017588680277109865
1034
+ ]
1035
+ },
1036
+ "llama_intact_nll_minus_gru": {
1037
+ "mean": -0.02184523277473145,
1038
+ "ci95": [
1039
+ -0.02796226582771772,
1040
+ -0.015897968166281817
1041
+ ]
1042
+ },
1043
+ "llama_response_error_minus_gru": {
1044
+ "mean": -0.018692779179584663,
1045
+ "ci95": [
1046
+ -0.021882112174503365,
1047
+ -0.015755130231753133
1048
+ ]
1049
+ },
1050
+ "llama_delta_nll_minus_gru": {
1051
+ "mean": -0.024862211690202942,
1052
+ "ci95": [
1053
+ -0.03389971664947549,
1054
+ -0.016321781476854192
1055
+ ]
1056
+ },
1057
+ "llama_intact_nll_minus_transformer": {
1058
+ "mean": -0.11198817103589596,
1059
+ "ci95": [
1060
+ -0.12731590745371837,
1061
+ -0.0969021472717774
1062
+ ]
1063
+ },
1064
+ "llama_response_error_minus_transformer": {
1065
+ "mean": -0.050136959077895386,
1066
+ "ci95": [
1067
+ -0.0564486625529101,
1068
+ -0.04378221510117189
1069
+ ]
1070
+ },
1071
+ "llama_delta_nll_minus_transformer": {
1072
+ "mean": 0.16302871527715934,
1073
+ "ci95": [
1074
+ 0.13884985361096064,
1075
+ 0.18730529007325716
1076
+ ]
1077
+ }
1078
+ },
1079
+ "0.5": {
1080
+ "oracle_intact_nll": {
1081
+ "mean": 0.35296904969669596,
1082
+ "ci95": [
1083
+ 0.33534777028299206,
1084
+ 0.37030867751321345
1085
+ ]
1086
+ },
1087
+ "oracle_donor_nll": {
1088
+ "mean": 1.7604470773588121,
1089
+ "ci95": [
1090
+ 1.6429015887320872,
1091
+ 1.8788747132315962
1092
+ ]
1093
+ },
1094
+ "oracle_delta_nll": {
1095
+ "mean": 1.4074780276621164,
1096
+ "ci95": [
1097
+ 1.2922222220837987,
1098
+ 1.526593543397259
1099
+ ]
1100
+ },
1101
+ "oracle_response_error": {
1102
+ "mean": 0.0,
1103
+ "ci95": [
1104
+ 0.0,
1105
+ 0.0
1106
+ ]
1107
+ },
1108
+ "fit_intact_nll": {
1109
+ "mean": 0.4728515243964842,
1110
+ "ci95": [
1111
+ 0.42894276257873043,
1112
+ 0.5231227235763171
1113
+ ]
1114
+ },
1115
+ "fit_donor_nll": {
1116
+ "mean": 1.802853697324906,
1117
+ "ci95": [
1118
+ 1.6943051717693254,
1119
+ 1.9130181086463949
1120
+ ]
1121
+ },
1122
+ "fit_delta_nll": {
1123
+ "mean": 1.3300021729284217,
1124
+ "ci95": [
1125
+ 1.2116312284668664,
1126
+ 1.4518071003249016
1127
+ ]
1128
+ },
1129
+ "fit_response_error": {
1130
+ "mean": 0.0719568,
1131
+ "ci95": [
1132
+ 0.06022152000000001,
1133
+ 0.08597213999999999
1134
+ ]
1135
+ },
1136
+ "rw_intact_nll": {
1137
+ "mean": 0.84433300429568,
1138
+ "ci95": [
1139
+ 0.8074712181553327,
1140
+ 0.881730027992606
1141
+ ]
1142
+ },
1143
+ "rw_donor_nll": {
1144
+ "mean": 1.54841550954771,
1145
+ "ci95": [
1146
+ 1.5308405514493986,
1147
+ 1.5639211302375513
1148
+ ]
1149
+ },
1150
+ "rw_delta_nll": {
1151
+ "mean": 0.7040825052520303,
1152
+ "ci95": [
1153
+ 0.6639001779897213,
1154
+ 0.742819256930886
1155
+ ]
1156
+ },
1157
+ "rw_response_error": {
1158
+ "mean": 0.5124604277504972,
1159
+ "ci95": [
1160
+ 0.5020080886421157,
1161
+ 0.5228493490985301
1162
+ ]
1163
+ },
1164
+ "gru_intact_nll": {
1165
+ "mean": 0.4239003927002035,
1166
+ "ci95": [
1167
+ 0.4044784331312756,
1168
+ 0.44449516947989026
1169
+ ]
1170
+ },
1171
+ "gru_donor_nll": {
1172
+ "mean": 0.7939825358098876,
1173
+ "ci95": [
1174
+ 0.7396016463645056,
1175
+ 0.8480251633772004
1176
+ ]
1177
+ },
1178
+ "gru_delta_nll": {
1179
+ "mean": 0.3700821431096841,
1180
+ "ci95": [
1181
+ 0.3225968490408368,
1182
+ 0.41760340011591346
1183
+ ]
1184
+ },
1185
+ "gru_response_error": {
1186
+ "mean": 0.34638604662118166,
1187
+ "ci95": [
1188
+ 0.3216506780240329,
1189
+ 0.3719628153741485
1190
+ ]
1191
+ },
1192
+ "gru_full_nll": {
1193
+ "mean": 0.4239003927002035,
1194
+ "ci95": [
1195
+ 0.4044784331312756,
1196
+ 0.44449516947989026
1197
+ ]
1198
+ },
1199
+ "gru_choice_nll": {
1200
+ "mean": 0.48624380729834094,
1201
+ "ci95": [
1202
+ 0.4644259395310131,
1203
+ 0.5090617307058393
1204
+ ]
1205
+ },
1206
+ "transformer_intact_nll": {
1207
+ "mean": 0.5602828206511774,
1208
+ "ci95": [
1209
+ 0.5330132524303004,
1210
+ 0.5877640749649157
1211
+ ]
1212
+ },
1213
+ "transformer_donor_nll": {
1214
+ "mean": 0.582972208854322,
1215
+ "ci95": [
1216
+ 0.5567214856489335,
1217
+ 0.6105053197155013
1218
+ ]
1219
+ },
1220
+ "transformer_delta_nll": {
1221
+ "mean": 0.022689388203144655,
1222
+ "ci95": [
1223
+ 0.019165301597907098,
1224
+ 0.02622493699970618
1225
+ ]
1226
+ },
1227
+ "transformer_response_error": {
1228
+ "mean": 0.42997184065303506,
1229
+ "ci95": [
1230
+ 0.3978538152802478,
1231
+ 0.46210426095038976
1232
+ ]
1233
+ },
1234
+ "transformer_full_nll": {
1235
+ "mean": 0.5602828206511774,
1236
+ "ci95": [
1237
+ 0.5330132524303004,
1238
+ 0.5877640749649157
1239
+ ]
1240
+ },
1241
+ "transformer_choice_nll": {
1242
+ "mean": 0.5113768472994935,
1243
+ "ci95": [
1244
+ 0.4887612396374821,
1245
+ 0.5352216072023641
1246
+ ]
1247
+ },
1248
+ "oracle_nll": {
1249
+ "mean": 0.35296904969669596,
1250
+ "ci95": [
1251
+ 0.33534777028299206,
1252
+ 0.37030867751321345
1253
+ ]
1254
+ },
1255
+ "llama_intact_nll": {
1256
+ "mean": 0.3987451196263668,
1257
+ "ci95": [
1258
+ 0.3795272060539283,
1259
+ 0.4174766444148147
1260
+ ]
1261
+ },
1262
+ "llama_donor_nll": {
1263
+ "mean": 0.6578082667469846,
1264
+ "ci95": [
1265
+ 0.6251918771369919,
1266
+ 0.6924845669774135
1267
+ ]
1268
+ },
1269
+ "llama_delta_nll": {
1270
+ "mean": 0.25906314712061784,
1271
+ "ci95": [
1272
+ 0.23491743712551183,
1273
+ 0.28339860169715614
1274
+ ]
1275
+ },
1276
+ "llama_response_error": {
1277
+ "mean": 0.34346799077972573,
1278
+ "ci95": [
1279
+ 0.31560118989718355,
1280
+ 0.3720033861070285
1281
+ ]
1282
+ },
1283
+ "llama_full_nll": {
1284
+ "mean": 0.3987451196263668,
1285
+ "ci95": [
1286
+ 0.3795272060539283,
1287
+ 0.4174766444148147
1288
+ ]
1289
+ },
1290
+ "llama_choice_nll": {
1291
+ "mean": 0.4839780912141663,
1292
+ "ci95": [
1293
+ 0.4618802124913104,
1294
+ 0.5071861501510945
1295
+ ]
1296
+ },
1297
+ "llama_reward_predictive_gain": {
1298
+ "mean": 0.08523297158779956,
1299
+ "ci95": [
1300
+ 0.07585017583638187,
1301
+ 0.09465791068877909
1302
+ ]
1303
+ },
1304
+ "llama_full_token_nll": {
1305
+ "mean": 0.39880475401878357,
1306
+ "ci95": [
1307
+ 0.37958583533763884,
1308
+ 0.4175353497266769
1309
+ ]
1310
+ },
1311
+ "llama_full_valid_mass": {
1312
+ "mean": 0.9999403953552246,
1313
+ "ci95": [
1314
+ 0.9999386072158813,
1315
+ 0.9999420657753945
1316
+ ]
1317
+ },
1318
+ "llama_choice_token_nll": {
1319
+ "mean": 0.48402056097984314,
1320
+ "ci95": [
1321
+ 0.46192228719592093,
1322
+ 0.5072284817695618
1323
+ ]
1324
+ },
1325
+ "llama_choice_valid_mass": {
1326
+ "mean": 0.9999575018882751,
1327
+ "ci95": [
1328
+ 0.999956488609314,
1329
+ 0.9999586343765259
1330
+ ]
1331
+ },
1332
+ "llama_intact_nll_minus_oracle": {
1333
+ "mean": 0.04577606992967082,
1334
+ "ci95": [
1335
+ 0.03926879655093441,
1336
+ 0.052720601383560065
1337
+ ]
1338
+ },
1339
+ "llama_response_error_minus_oracle": {
1340
+ "mean": 0.34346799077972573,
1341
+ "ci95": [
1342
+ 0.31560118989718355,
1343
+ 0.3720033861070285
1344
+ ]
1345
+ },
1346
+ "llama_delta_nll_minus_oracle": {
1347
+ "mean": -1.1484148805414987,
1348
+ "ci95": [
1349
+ -1.247769469515086,
1350
+ -1.0495695008559418
1351
+ ]
1352
+ },
1353
+ "llama_intact_nll_minus_fit": {
1354
+ "mean": -0.07410640477011742,
1355
+ "ci95": [
1356
+ -0.1233130595576454,
1357
+ -0.03528465760684591
1358
+ ]
1359
+ },
1360
+ "llama_response_error_minus_fit": {
1361
+ "mean": 0.27151119077972574,
1362
+ "ci95": [
1363
+ 0.23908827143141892,
1364
+ 0.3037675233960233
1365
+ ]
1366
+ },
1367
+ "llama_delta_nll_minus_fit": {
1368
+ "mean": -1.0709390258078042,
1369
+ "ci95": [
1370
+ -1.1728408809095585,
1371
+ -0.9705878516756828
1372
+ ]
1373
+ },
1374
+ "llama_intact_nll_minus_rw": {
1375
+ "mean": -0.4455878846693131,
1376
+ "ci95": [
1377
+ -0.4838515897940322,
1378
+ -0.4104475509720684
1379
+ ]
1380
+ },
1381
+ "llama_response_error_minus_rw": {
1382
+ "mean": -0.1689924369707714,
1383
+ "ci95": [
1384
+ -0.19014066558251644,
1385
+ -0.14664546459376823
1386
+ ]
1387
+ },
1388
+ "llama_delta_nll_minus_rw": {
1389
+ "mean": -0.44501935813141247,
1390
+ "ci95": [
1391
+ -0.4888386935802328,
1392
+ -0.40198073247091254
1393
+ ]
1394
+ },
1395
+ "llama_intact_nll_minus_gru": {
1396
+ "mean": -0.025155273073836745,
1397
+ "ci95": [
1398
+ -0.029978908946826935,
1399
+ -0.02015752292453398
1400
+ ]
1401
+ },
1402
+ "llama_response_error_minus_gru": {
1403
+ "mean": -0.0029180558414559498,
1404
+ "ci95": [
1405
+ -0.007855505290075695,
1406
+ 0.0021493338478629784
1407
+ ]
1408
+ },
1409
+ "llama_delta_nll_minus_gru": {
1410
+ "mean": -0.11101899598906631,
1411
+ "ci95": [
1412
+ -0.1421524779034286,
1413
+ -0.08121782677104071
1414
+ ]
1415
+ },
1416
+ "llama_intact_nll_minus_transformer": {
1417
+ "mean": -0.16153770102481058,
1418
+ "ci95": [
1419
+ -0.1764153692596757,
1420
+ -0.14719912494680829
1421
+ ]
1422
+ },
1423
+ "llama_response_error_minus_transformer": {
1424
+ "mean": -0.0865038498733094,
1425
+ "ci95": [
1426
+ -0.0923243371810329,
1427
+ -0.08094139169911427
1428
+ ]
1429
+ },
1430
+ "llama_delta_nll_minus_transformer": {
1431
+ "mean": 0.23637375891747317,
1432
+ "ci95": [
1433
+ 0.2122926776852641,
1434
+ 0.2605071777041206
1435
+ ]
1436
+ }
1437
+ },
1438
+ "0.7": {
1439
+ "oracle_intact_nll": {
1440
+ "mean": 0.35354679656271903,
1441
+ "ci95": [
1442
+ 0.33708101088106135,
1443
+ 0.36973093064719037
1444
+ ]
1445
+ },
1446
+ "oracle_donor_nll": {
1447
+ "mean": 2.2652967325614264,
1448
+ "ci95": [
1449
+ 2.182707095879845,
1450
+ 2.347232793100819
1451
+ ]
1452
+ },
1453
+ "oracle_delta_nll": {
1454
+ "mean": 1.9117499359987074,
1455
+ "ci95": [
1456
+ 1.829737323999567,
1457
+ 1.994946206889613
1458
+ ]
1459
+ },
1460
+ "oracle_response_error": {
1461
+ "mean": 0.0,
1462
+ "ci95": [
1463
+ 0.0,
1464
+ 0.0
1465
+ ]
1466
+ },
1467
+ "fit_intact_nll": {
1468
+ "mean": 0.4933615381403033,
1469
+ "ci95": [
1470
+ 0.4419348452284243,
1471
+ 0.5505584778765878
1472
+ ]
1473
+ },
1474
+ "fit_donor_nll": {
1475
+ "mean": 2.2282776021209982,
1476
+ "ci95": [
1477
+ 2.1387232269695082,
1478
+ 2.314518598995845
1479
+ ]
1480
+ },
1481
+ "fit_delta_nll": {
1482
+ "mean": 1.7349160639806946,
1483
+ "ci95": [
1484
+ 1.6357078997897507,
1485
+ 1.8323866544720737
1486
+ ]
1487
+ },
1488
+ "fit_response_error": {
1489
+ "mean": 0.10029600000000001,
1490
+ "ci95": [
1491
+ 0.0831024,
1492
+ 0.11907972
1493
+ ]
1494
+ },
1495
+ "rw_intact_nll": {
1496
+ "mean": 0.6367851249131584,
1497
+ "ci95": [
1498
+ 0.6062237888087522,
1499
+ 0.6686761082412468
1500
+ ]
1501
+ },
1502
+ "rw_donor_nll": {
1503
+ "mean": 1.7542504882058652,
1504
+ "ci95": [
1505
+ 1.7240342990137505,
1506
+ 1.7826033724751518
1507
+ ]
1508
+ },
1509
+ "rw_delta_nll": {
1510
+ "mean": 1.1174653632927065,
1511
+ "ci95": [
1512
+ 1.0743206871677031,
1513
+ 1.1590762397115524
1514
+ ]
1515
+ },
1516
+ "rw_response_error": {
1517
+ "mean": 0.46508049211239433,
1518
+ "ci95": [
1519
+ 0.45742209826830893,
1520
+ 0.4734375790348134
1521
+ ]
1522
+ },
1523
+ "gru_intact_nll": {
1524
+ "mean": 0.43352238996782294,
1525
+ "ci95": [
1526
+ 0.41481874523566686,
1527
+ 0.45245633458546514
1528
+ ]
1529
+ },
1530
+ "gru_donor_nll": {
1531
+ "mean": 0.8305084236511499,
1532
+ "ci95": [
1533
+ 0.7865084313653908,
1534
+ 0.8736835526655155
1535
+ ]
1536
+ },
1537
+ "gru_delta_nll": {
1538
+ "mean": 0.3969860336833268,
1539
+ "ci95": [
1540
+ 0.3603585324551862,
1541
+ 0.43228300310841344
1542
+ ]
1543
+ },
1544
+ "gru_response_error": {
1545
+ "mean": 0.47436250322923085,
1546
+ "ci95": [
1547
+ 0.45668643212549676,
1548
+ 0.4917381840218523
1549
+ ]
1550
+ },
1551
+ "gru_full_nll": {
1552
+ "mean": 0.43352238996782294,
1553
+ "ci95": [
1554
+ 0.41481874523566686,
1555
+ 0.45245633458546514
1556
+ ]
1557
+ },
1558
+ "gru_choice_nll": {
1559
+ "mean": 0.495903679754697,
1560
+ "ci95": [
1561
+ 0.47317675171886986,
1562
+ 0.5196695544489524
1563
+ ]
1564
+ },
1565
+ "transformer_intact_nll": {
1566
+ "mean": 0.5486615970356825,
1567
+ "ci95": [
1568
+ 0.5204956015832901,
1569
+ 0.5757764860976032
1570
+ ]
1571
+ },
1572
+ "transformer_donor_nll": {
1573
+ "mean": 0.5974257733244831,
1574
+ "ci95": [
1575
+ 0.5677512554799186,
1576
+ 0.6274911532771006
1577
+ ]
1578
+ },
1579
+ "transformer_delta_nll": {
1580
+ "mean": 0.04876417628880067,
1581
+ "ci95": [
1582
+ 0.04379575082706635,
1583
+ 0.05436622825302045
1584
+ ]
1585
+ },
1586
+ "transformer_response_error": {
1587
+ "mean": 0.5682790094067541,
1588
+ "ci95": [
1589
+ 0.5453435994574595,
1590
+ 0.5907230337345212
1591
+ ]
1592
+ },
1593
+ "transformer_full_nll": {
1594
+ "mean": 0.5486615970356825,
1595
+ "ci95": [
1596
+ 0.5204956015832901,
1597
+ 0.5757764860976032
1598
+ ]
1599
+ },
1600
+ "transformer_choice_nll": {
1601
+ "mean": 0.5262255678096864,
1602
+ "ci95": [
1603
+ 0.5000937996184767,
1604
+ 0.5526728572725551
1605
+ ]
1606
+ },
1607
+ "oracle_nll": {
1608
+ "mean": 0.35354679656271903,
1609
+ "ci95": [
1610
+ 0.33708101088106135,
1611
+ 0.36973093064719037
1612
+ ]
1613
+ },
1614
+ "llama_intact_nll": {
1615
+ "mean": 0.3974785488950231,
1616
+ "ci95": [
1617
+ 0.380080654508458,
1618
+ 0.4144363998542966
1619
+ ]
1620
+ },
1621
+ "llama_donor_nll": {
1622
+ "mean": 0.6605001509942868,
1623
+ "ci95": [
1624
+ 0.6271926209379498,
1625
+ 0.6936083935639172
1626
+ ]
1627
+ },
1628
+ "llama_delta_nll": {
1629
+ "mean": 0.2630216020992638,
1630
+ "ci95": [
1631
+ 0.24054951176665107,
1632
+ 0.28584948863636683
1633
+ ]
1634
+ },
1635
+ "llama_response_error": {
1636
+ "mean": 0.4869802405920419,
1637
+ "ci95": [
1638
+ 0.4660788894272456,
1639
+ 0.5081262377202105
1640
+ ]
1641
+ },
1642
+ "llama_full_nll": {
1643
+ "mean": 0.3974785488950231,
1644
+ "ci95": [
1645
+ 0.380080654508458,
1646
+ 0.4144363998542966
1647
+ ]
1648
+ },
1649
+ "llama_choice_nll": {
1650
+ "mean": 0.49387579370840684,
1651
+ "ci95": [
1652
+ 0.4712477815121117,
1653
+ 0.517045396465575
1654
+ ]
1655
+ },
1656
+ "llama_reward_predictive_gain": {
1657
+ "mean": 0.09639724481338369,
1658
+ "ci95": [
1659
+ 0.08522192019083413,
1660
+ 0.10836390890938048
1661
+ ]
1662
+ },
1663
+ "llama_full_token_nll": {
1664
+ "mean": 0.39753857254981995,
1665
+ "ci95": [
1666
+ 0.3801401451230049,
1667
+ 0.4144964061677456
1668
+ ]
1669
+ },
1670
+ "llama_full_valid_mass": {
1671
+ "mean": 0.9999399781227112,
1672
+ "ci95": [
1673
+ 0.9999388456344604,
1674
+ 0.9999411702156067
1675
+ ]
1676
+ },
1677
+ "llama_choice_token_nll": {
1678
+ "mean": 0.4939333498477936,
1679
+ "ci95": [
1680
+ 0.4713045693933964,
1681
+ 0.5171042591333389
1682
+ ]
1683
+ },
1684
+ "llama_choice_valid_mass": {
1685
+ "mean": 0.999942421913147,
1686
+ "ci95": [
1687
+ 0.999940812587738,
1688
+ 0.9999440908432007
1689
+ ]
1690
+ },
1691
+ "llama_intact_nll_minus_oracle": {
1692
+ "mean": 0.043931752332304126,
1693
+ "ci95": [
1694
+ 0.036854597532115695,
1695
+ 0.05173495101456569
1696
+ ]
1697
+ },
1698
+ "llama_response_error_minus_oracle": {
1699
+ "mean": 0.4869802405920419,
1700
+ "ci95": [
1701
+ 0.4660788894272456,
1702
+ 0.5081262377202105
1703
+ ]
1704
+ },
1705
+ "llama_delta_nll_minus_oracle": {
1706
+ "mean": -1.6487283338994438,
1707
+ "ci95": [
1708
+ -1.7249575418072387,
1709
+ -1.577057046715769
1710
+ ]
1711
+ },
1712
+ "llama_intact_nll_minus_fit": {
1713
+ "mean": -0.09588298924528023,
1714
+ "ci95": [
1715
+ -0.15214240188075048,
1716
+ -0.04897140405950008
1717
+ ]
1718
+ },
1719
+ "llama_response_error_minus_fit": {
1720
+ "mean": 0.3866842405920419,
1721
+ "ci95": [
1722
+ 0.3557939978612507,
1723
+ 0.41692314179455275
1724
+ ]
1725
+ },
1726
+ "llama_delta_nll_minus_fit": {
1727
+ "mean": -1.4718944618814307,
1728
+ "ci95": [
1729
+ -1.5635489892432026,
1730
+ -1.3764020795206622
1731
+ ]
1732
+ },
1733
+ "llama_intact_nll_minus_rw": {
1734
+ "mean": -0.2393065760181352,
1735
+ "ci95": [
1736
+ -0.26506185249864594,
1737
+ -0.21611196263014162
1738
+ ]
1739
+ },
1740
+ "llama_response_error_minus_rw": {
1741
+ "mean": 0.021899748479647596,
1742
+ "ci95": [
1743
+ 0.002835994371429257,
1744
+ 0.042526657757830356
1745
+ ]
1746
+ },
1747
+ "llama_delta_nll_minus_rw": {
1748
+ "mean": -0.8544437611934429,
1749
+ "ci95": [
1750
+ -0.9062715653782083,
1751
+ -0.801456764813822
1752
+ ]
1753
+ },
1754
+ "llama_intact_nll_minus_gru": {
1755
+ "mean": -0.03604384107279983,
1756
+ "ci95": [
1757
+ -0.04197193074133295,
1758
+ -0.030143613303549227
1759
+ ]
1760
+ },
1761
+ "llama_response_error_minus_gru": {
1762
+ "mean": 0.012617737362811087,
1763
+ "ci95": [
1764
+ 0.006723725703928434,
1765
+ 0.01854876521114142
1766
+ ]
1767
+ },
1768
+ "llama_delta_nll_minus_gru": {
1769
+ "mean": -0.1339644315840631,
1770
+ "ci95": [
1771
+ -0.15985919765659581,
1772
+ -0.10819204135960189
1773
+ ]
1774
+ },
1775
+ "llama_intact_nll_minus_transformer": {
1776
+ "mean": -0.15118304814065933,
1777
+ "ci95": [
1778
+ -0.16909179573355676,
1779
+ -0.13355382551360453
1780
+ ]
1781
+ },
1782
+ "llama_response_error_minus_transformer": {
1783
+ "mean": -0.08129876881471218,
1784
+ "ci95": [
1785
+ -0.08700684454602214,
1786
+ -0.07563797756565344
1787
+ ]
1788
+ },
1789
+ "llama_delta_nll_minus_transformer": {
1790
+ "mean": 0.2142574258104631,
1791
+ "ci95": [
1792
+ 0.19306042822227676,
1793
+ 0.2363567747733869
1794
+ ]
1795
+ }
1796
+ },
1797
+ "0.9": {
1798
+ "oracle_intact_nll": {
1799
+ "mean": 0.3538356699957305,
1800
+ "ci95": [
1801
+ 0.33765153591125907,
1802
+ 0.37030145567738815
1803
+ ]
1804
+ },
1805
+ "oracle_donor_nll": {
1806
+ "mean": 2.41985846289425,
1807
+ "ci95": [
1808
+ 2.369679703212981,
1809
+ 2.4652137584278093
1810
+ ]
1811
+ },
1812
+ "oracle_delta_nll": {
1813
+ "mean": 2.0660227928985195,
1814
+ "ci95": [
1815
+ 2.0177632360878204,
1816
+ 2.1139530339955845
1817
+ ]
1818
+ },
1819
+ "oracle_response_error": {
1820
+ "mean": 0.0,
1821
+ "ci95": [
1822
+ 0.0,
1823
+ 0.0
1824
+ ]
1825
+ },
1826
+ "fit_intact_nll": {
1827
+ "mean": 0.4725626509634726,
1828
+ "ci95": [
1829
+ 0.4272095219806612,
1830
+ 0.5289001922365478
1831
+ ]
1832
+ },
1833
+ "fit_donor_nll": {
1834
+ "mean": 2.3385839225164538,
1835
+ "ci95": [
1836
+ 2.260696423140718,
1837
+ 2.410050487659926
1838
+ ]
1839
+ },
1840
+ "fit_delta_nll": {
1841
+ "mean": 1.8660212715529811,
1842
+ "ci95": [
1843
+ 1.7797882186975418,
1844
+ 1.9454040520363431
1845
+ ]
1846
+ },
1847
+ "fit_response_error": {
1848
+ "mean": 0.10665,
1849
+ "ci95": [
1850
+ 0.08836884,
1851
+ 0.12607901999999996
1852
+ ]
1853
+ },
1854
+ "rw_intact_nll": {
1855
+ "mean": 0.5646761974613452,
1856
+ "ci95": [
1857
+ 0.537570337446115,
1858
+ 0.5915569053754879
1859
+ ]
1860
+ },
1861
+ "rw_donor_nll": {
1862
+ "mean": 1.8486147261739798,
1863
+ "ci95": [
1864
+ 1.8142479206821613,
1865
+ 1.8822080563514352
1866
+ ]
1867
+ },
1868
+ "rw_delta_nll": {
1869
+ "mean": 1.2839385287126341,
1870
+ "ci95": [
1871
+ 1.241250557216505,
1872
+ 1.3268271689525293
1873
+ ]
1874
+ },
1875
+ "rw_response_error": {
1876
+ "mean": 0.4112662210155444,
1877
+ "ci95": [
1878
+ 0.4035144972266136,
1879
+ 0.4193465953308267
1880
+ ]
1881
+ },
1882
+ "gru_intact_nll": {
1883
+ "mean": 0.42817653542883566,
1884
+ "ci95": [
1885
+ 0.4085331050896236,
1886
+ 0.44769042308671836
1887
+ ]
1888
+ },
1889
+ "gru_donor_nll": {
1890
+ "mean": 0.7773886712318161,
1891
+ "ci95": [
1892
+ 0.7521886647158033,
1893
+ 0.8027420850885961
1894
+ ]
1895
+ },
1896
+ "gru_delta_nll": {
1897
+ "mean": 0.3492121358029805,
1898
+ "ci95": [
1899
+ 0.33413805732716767,
1900
+ 0.36448600874580334
1901
+ ]
1902
+ },
1903
+ "gru_response_error": {
1904
+ "mean": 0.5252097769509073,
1905
+ "ci95": [
1906
+ 0.5132097820888408,
1907
+ 0.5367206557750891
1908
+ ]
1909
+ },
1910
+ "gru_full_nll": {
1911
+ "mean": 0.42817653542883566,
1912
+ "ci95": [
1913
+ 0.4085331050896236,
1914
+ 0.44769042308671836
1915
+ ]
1916
+ },
1917
+ "gru_choice_nll": {
1918
+ "mean": 0.48175588727990154,
1919
+ "ci95": [
1920
+ 0.4600201290380613,
1921
+ 0.5046304512817076
1922
+ ]
1923
+ },
1924
+ "transformer_intact_nll": {
1925
+ "mean": 0.47869842622950065,
1926
+ "ci95": [
1927
+ 0.4561680661028501,
1928
+ 0.5016709454520106
1929
+ ]
1930
+ },
1931
+ "transformer_donor_nll": {
1932
+ "mean": 0.5376308068473252,
1933
+ "ci95": [
1934
+ 0.5113061337210391,
1935
+ 0.5652464509314599
1936
+ ]
1937
+ },
1938
+ "transformer_delta_nll": {
1939
+ "mean": 0.058932380617824506,
1940
+ "ci95": [
1941
+ 0.051641284946157176,
1942
+ 0.06626256840178946
1943
+ ]
1944
+ },
1945
+ "transformer_response_error": {
1946
+ "mean": 0.5924415928248987,
1947
+ "ci95": [
1948
+ 0.5784672039496963,
1949
+ 0.6053087833465967
1950
+ ]
1951
+ },
1952
+ "transformer_full_nll": {
1953
+ "mean": 0.47869842622950065,
1954
+ "ci95": [
1955
+ 0.4561680661028501,
1956
+ 0.5016709454520106
1957
+ ]
1958
+ },
1959
+ "transformer_choice_nll": {
1960
+ "mean": 0.5006007779105845,
1961
+ "ci95": [
1962
+ 0.47637668673713157,
1963
+ 0.5257339563882727
1964
+ ]
1965
+ },
1966
+ "oracle_nll": {
1967
+ "mean": 0.3538356699957305,
1968
+ "ci95": [
1969
+ 0.33765153591125907,
1970
+ 0.37030145567738815
1971
+ ]
1972
+ },
1973
+ "llama_intact_nll": {
1974
+ "mean": 0.39667870658271903,
1975
+ "ci95": [
1976
+ 0.37927451894877817,
1977
+ 0.4139721763957136
1978
+ ]
1979
+ },
1980
+ "llama_donor_nll": {
1981
+ "mean": 0.608545701028622,
1982
+ "ci95": [
1983
+ 0.5791454309951275,
1984
+ 0.6388817952415438
1985
+ ]
1986
+ },
1987
+ "llama_delta_nll": {
1988
+ "mean": 0.21186699444590307,
1989
+ "ci95": [
1990
+ 0.19202949506561026,
1991
+ 0.23230461939409788
1992
+ ]
1993
+ },
1994
+ "llama_response_error": {
1995
+ "mean": 0.5403098921261067,
1996
+ "ci95": [
1997
+ 0.5277650008091874,
1998
+ 0.5517975862179028
1999
+ ]
2000
+ },
2001
+ "llama_full_nll": {
2002
+ "mean": 0.39667870658271903,
2003
+ "ci95": [
2004
+ 0.37927451894877817,
2005
+ 0.4139721763957136
2006
+ ]
2007
+ },
2008
+ "llama_choice_nll": {
2009
+ "mean": 0.48026610224555905,
2010
+ "ci95": [
2011
+ 0.4588816590884829,
2012
+ 0.502987530858407
2013
+ ]
2014
+ },
2015
+ "llama_reward_predictive_gain": {
2016
+ "mean": 0.08358739566284004,
2017
+ "ci95": [
2018
+ 0.07248513776142348,
2019
+ 0.09623014834951416
2020
+ ]
2021
+ },
2022
+ "llama_full_token_nll": {
2023
+ "mean": 0.39672568440437317,
2024
+ "ci95": [
2025
+ 0.37932121828198434,
2026
+ 0.4140197172760963
2027
+ ]
2028
+ },
2029
+ "llama_full_valid_mass": {
2030
+ "mean": 0.9999530911445618,
2031
+ "ci95": [
2032
+ 0.9999517798423767,
2033
+ 0.9999542236328125
2034
+ ]
2035
+ },
2036
+ "llama_choice_token_nll": {
2037
+ "mean": 0.480336457490921,
2038
+ "ci95": [
2039
+ 0.45895086228847504,
2040
+ 0.5030584007501602
2041
+ ]
2042
+ },
2043
+ "llama_choice_valid_mass": {
2044
+ "mean": 0.999929666519165,
2045
+ "ci95": [
2046
+ 0.9999282956123352,
2047
+ 0.9999310374259949
2048
+ ]
2049
+ },
2050
+ "llama_intact_nll_minus_oracle": {
2051
+ "mean": 0.04284303658698849,
2052
+ "ci95": [
2053
+ 0.035765527939378886,
2054
+ 0.05045790424461734
2055
+ ]
2056
+ },
2057
+ "llama_response_error_minus_oracle": {
2058
+ "mean": 0.5403098921261067,
2059
+ "ci95": [
2060
+ 0.5277650008091874,
2061
+ 0.5517975862179028
2062
+ ]
2063
+ },
2064
+ "llama_delta_nll_minus_oracle": {
2065
+ "mean": -1.8541557984526162,
2066
+ "ci95": [
2067
+ -1.8988863953627915,
2068
+ -1.8105206502453879
2069
+ ]
2070
+ },
2071
+ "llama_intact_nll_minus_fit": {
2072
+ "mean": -0.0758839443807536,
2073
+ "ci95": [
2074
+ -0.12871826114420942,
2075
+ -0.03361666673799958
2076
+ ]
2077
+ },
2078
+ "llama_response_error_minus_fit": {
2079
+ "mean": 0.43365989212610656,
2080
+ "ci95": [
2081
+ 0.4074478403434141,
2082
+ 0.45693999930065327
2083
+ ]
2084
+ },
2085
+ "llama_delta_nll_minus_fit": {
2086
+ "mean": -1.654154277107078,
2087
+ "ci95": [
2088
+ -1.7326281130287842,
2089
+ -1.568852129220658
2090
+ ]
2091
+ },
2092
+ "llama_intact_nll_minus_rw": {
2093
+ "mean": -0.16799749087862625,
2094
+ "ci95": [
2095
+ -0.18612315345751024,
2096
+ -0.15047954623914872
2097
+ ]
2098
+ },
2099
+ "llama_response_error_minus_rw": {
2100
+ "mean": 0.12904367111056222,
2101
+ "ci95": [
2102
+ 0.11618892982769927,
2103
+ 0.14227559475049703
2104
+ ]
2105
+ },
2106
+ "llama_delta_nll_minus_rw": {
2107
+ "mean": -1.0720715342667313,
2108
+ "ci95": [
2109
+ -1.1183355629288811,
2110
+ -1.0233181365739292
2111
+ ]
2112
+ },
2113
+ "llama_intact_nll_minus_gru": {
2114
+ "mean": -0.031497828846116654,
2115
+ "ci95": [
2116
+ -0.03815511682656203,
2117
+ -0.024886098531542653
2118
+ ]
2119
+ },
2120
+ "llama_response_error_minus_gru": {
2121
+ "mean": 0.015100115175199246,
2122
+ "ci95": [
2123
+ 0.010715179975850543,
2124
+ 0.01944850838273202
2125
+ ]
2126
+ },
2127
+ "llama_delta_nll_minus_gru": {
2128
+ "mean": -0.13734514135707737,
2129
+ "ci95": [
2130
+ -0.15092070054641715,
2131
+ -0.123119345173893
2132
+ ]
2133
+ },
2134
+ "llama_intact_nll_minus_transformer": {
2135
+ "mean": -0.08201971964678159,
2136
+ "ci95": [
2137
+ -0.09421665066702264,
2138
+ -0.070218062217118
2139
+ ]
2140
+ },
2141
+ "llama_response_error_minus_transformer": {
2142
+ "mean": -0.052131700698792104,
2143
+ "ci95": [
2144
+ -0.05629321360111152,
2145
+ -0.048103697563149236
2146
+ ]
2147
+ },
2148
+ "llama_delta_nll_minus_transformer": {
2149
+ "mean": 0.15293461382807857,
2150
+ "ci95": [
2151
+ 0.1376275486143692,
2152
+ 0.16851263365293667
2153
+ ]
2154
+ }
2155
+ },
2156
+ "1.0": {
2157
+ "oracle_intact_nll": {
2158
+ "mean": 0.3555689105937998,
2159
+ "ci95": [
2160
+ 0.3396736499423399,
2161
+ 0.3711680759764228
2162
+ ]
2163
+ },
2164
+ "oracle_donor_nll": {
2165
+ "mean": 2.422530542149607,
2166
+ "ci95": [
2167
+ 2.3779468997739794,
2168
+ 2.465183065625552
2169
+ ]
2170
+ },
2171
+ "oracle_delta_nll": {
2172
+ "mean": 2.066961631555807,
2173
+ "ci95": [
2174
+ 2.022474039596656,
2175
+ 2.1101843189701586
2176
+ ]
2177
+ },
2178
+ "oracle_response_error": {
2179
+ "mean": 0.0,
2180
+ "ci95": [
2181
+ 0.0,
2182
+ 0.0
2183
+ ]
2184
+ },
2185
+ "fit_intact_nll": {
2186
+ "mean": 0.4893173100781419,
2187
+ "ci95": [
2188
+ 0.4410682249293897,
2189
+ 0.5502696044435763
2190
+ ]
2191
+ },
2192
+ "fit_donor_nll": {
2193
+ "mean": 2.3547897221084013,
2194
+ "ci95": [
2195
+ 2.282675719383192,
2196
+ 2.4228771902692205
2197
+ ]
2198
+ },
2199
+ "fit_delta_nll": {
2200
+ "mean": 1.865472412030259,
2201
+ "ci95": [
2202
+ 1.7785410076505144,
2203
+ 1.9401815814592855
2204
+ ]
2205
+ },
2206
+ "fit_response_error": {
2207
+ "mean": 0.1060848,
2208
+ "ci95": [
2209
+ 0.08908181999999999,
2210
+ 0.12599712
2211
+ ]
2212
+ },
2213
+ "rw_intact_nll": {
2214
+ "mean": 0.5564620276621879,
2215
+ "ci95": [
2216
+ 0.530472277106952,
2217
+ 0.5812907234230831
2218
+ ]
2219
+ },
2220
+ "rw_donor_nll": {
2221
+ "mean": 1.843404323107438,
2222
+ "ci95": [
2223
+ 1.809046268009209,
2224
+ 1.8763265015444153
2225
+ ]
2226
+ },
2227
+ "rw_delta_nll": {
2228
+ "mean": 1.2869422954452503,
2229
+ "ci95": [
2230
+ 1.2449159386403028,
2231
+ 1.3291663409876573
2232
+ ]
2233
+ },
2234
+ "rw_response_error": {
2235
+ "mean": 0.4067648792552328,
2236
+ "ci95": [
2237
+ 0.3990020277362318,
2238
+ 0.4146327477106259
2239
+ ]
2240
+ },
2241
+ "gru_intact_nll": {
2242
+ "mean": 0.4122413658801879,
2243
+ "ci95": [
2244
+ 0.3931783804615061,
2245
+ 0.43102947815657017
2246
+ ]
2247
+ },
2248
+ "gru_donor_nll": {
2249
+ "mean": 0.8702557120266952,
2250
+ "ci95": [
2251
+ 0.8435820908089514,
2252
+ 0.8957992050433226
2253
+ ]
2254
+ },
2255
+ "gru_delta_nll": {
2256
+ "mean": 0.4580143461465074,
2257
+ "ci95": [
2258
+ 0.43902728630998195,
2259
+ 0.47724652441173804
2260
+ ]
2261
+ },
2262
+ "gru_response_error": {
2263
+ "mean": 0.500764796355698,
2264
+ "ci95": [
2265
+ 0.4899591589280651,
2266
+ 0.5111032163516833
2267
+ ]
2268
+ },
2269
+ "gru_full_nll": {
2270
+ "mean": 0.4122413658801879,
2271
+ "ci95": [
2272
+ 0.3931783804615061,
2273
+ 0.43102947815657017
2274
+ ]
2275
+ },
2276
+ "gru_choice_nll": {
2277
+ "mean": 0.47484978552653206,
2278
+ "ci95": [
2279
+ 0.45233506516121796,
2280
+ 0.4971920267155352
2281
+ ]
2282
+ },
2283
+ "transformer_intact_nll": {
2284
+ "mean": 0.4626244695171602,
2285
+ "ci95": [
2286
+ 0.4396565570485781,
2287
+ 0.48442435589321375
2288
+ ]
2289
+ },
2290
+ "transformer_donor_nll": {
2291
+ "mean": 0.5467612604003217,
2292
+ "ci95": [
2293
+ 0.5183106513919654,
2294
+ 0.573915460684448
2295
+ ]
2296
+ },
2297
+ "transformer_delta_nll": {
2298
+ "mean": 0.08413679088316148,
2299
+ "ci95": [
2300
+ 0.07327072215090144,
2301
+ 0.09466927768676733
2302
+ ]
2303
+ },
2304
+ "transformer_response_error": {
2305
+ "mean": 0.5924755479283029,
2306
+ "ci95": [
2307
+ 0.5798533560695258,
2308
+ 0.6044707689413092
2309
+ ]
2310
+ },
2311
+ "transformer_full_nll": {
2312
+ "mean": 0.4626244695171602,
2313
+ "ci95": [
2314
+ 0.4396565570485781,
2315
+ 0.48442435589321375
2316
+ ]
2317
+ },
2318
+ "transformer_choice_nll": {
2319
+ "mean": 0.4868411339728783,
2320
+ "ci95": [
2321
+ 0.46297930507026996,
2322
+ 0.5105435287868588
2323
+ ]
2324
+ },
2325
+ "oracle_nll": {
2326
+ "mean": 0.3555689105937998,
2327
+ "ci95": [
2328
+ 0.3396736499423399,
2329
+ 0.3711680759764228
2330
+ ]
2331
+ },
2332
+ "llama_intact_nll": {
2333
+ "mean": 0.39184374347422907,
2334
+ "ci95": [
2335
+ 0.37459694013477624,
2336
+ 0.4088950564098682
2337
+ ]
2338
+ },
2339
+ "llama_donor_nll": {
2340
+ "mean": 0.5859508390906728,
2341
+ "ci95": [
2342
+ 0.5577570742185399,
2343
+ 0.6126744585042119
2344
+ ]
2345
+ },
2346
+ "llama_delta_nll": {
2347
+ "mean": 0.19410709561644388,
2348
+ "ci95": [
2349
+ 0.17587620969499435,
2350
+ 0.21193392297664818
2351
+ ]
2352
+ },
2353
+ "llama_response_error": {
2354
+ "mean": 0.5390233532237088,
2355
+ "ci95": [
2356
+ 0.5286321838819992,
2357
+ 0.5494029351578702
2358
+ ]
2359
+ },
2360
+ "llama_full_nll": {
2361
+ "mean": 0.39184374347422907,
2362
+ "ci95": [
2363
+ 0.37459694013477624,
2364
+ 0.4088950564098682
2365
+ ]
2366
+ },
2367
+ "llama_choice_nll": {
2368
+ "mean": 0.4751047001399281,
2369
+ "ci95": [
2370
+ 0.4528385337205593,
2371
+ 0.49719918769813887
2372
+ ]
2373
+ },
2374
+ "llama_reward_predictive_gain": {
2375
+ "mean": 0.083260956665699,
2376
+ "ci95": [
2377
+ 0.0715406502095134,
2378
+ 0.09523279384042423
2379
+ ]
2380
+ },
2381
+ "llama_full_token_nll": {
2382
+ "mean": 0.3919104039669037,
2383
+ "ci95": [
2384
+ 0.3746641531586647,
2385
+ 0.4089613363146782
2386
+ ]
2387
+ },
2388
+ "llama_full_valid_mass": {
2389
+ "mean": 0.9999333620071411,
2390
+ "ci95": [
2391
+ 0.9999320492148399,
2392
+ 0.9999346733093262
2393
+ ]
2394
+ },
2395
+ "llama_choice_token_nll": {
2396
+ "mean": 0.4751896262168884,
2397
+ "ci95": [
2398
+ 0.4529243931174278,
2399
+ 0.49728643521666527
2400
+ ]
2401
+ },
2402
+ "llama_choice_valid_mass": {
2403
+ "mean": 0.9999150633811951,
2404
+ "ci95": [
2405
+ 0.9999131560325623,
2406
+ 0.9999168515205383
2407
+ ]
2408
+ },
2409
+ "llama_intact_nll_minus_oracle": {
2410
+ "mean": 0.03627483288042927,
2411
+ "ci95": [
2412
+ 0.029940674450755312,
2413
+ 0.04277786668057771
2414
+ ]
2415
+ },
2416
+ "llama_response_error_minus_oracle": {
2417
+ "mean": 0.5390233532237088,
2418
+ "ci95": [
2419
+ 0.5286321838819992,
2420
+ 0.5494029351578702
2421
+ ]
2422
+ },
2423
+ "llama_delta_nll_minus_oracle": {
2424
+ "mean": -1.872854535939363,
2425
+ "ci95": [
2426
+ -1.9137238574950182,
2427
+ -1.8322875447547307
2428
+ ]
2429
+ },
2430
+ "llama_intact_nll_minus_fit": {
2431
+ "mean": -0.09747356660391279,
2432
+ "ci95": [
2433
+ -0.15633134963038178,
2434
+ -0.05143152074867326
2435
+ ]
2436
+ },
2437
+ "llama_response_error_minus_fit": {
2438
+ "mean": 0.4329385532237088,
2439
+ "ci95": [
2440
+ 0.4090187008293367,
2441
+ 0.4540663865153819
2442
+ ]
2443
+ },
2444
+ "llama_delta_nll_minus_fit": {
2445
+ "mean": -1.6713653164138151,
2446
+ "ci95": [
2447
+ -1.7464913186875348,
2448
+ -1.5851355245108543
2449
+ ]
2450
+ },
2451
+ "llama_intact_nll_minus_rw": {
2452
+ "mean": -0.16461828418795885,
2453
+ "ci95": [
2454
+ -0.18002135087454282,
2455
+ -0.14941361428660996
2456
+ ]
2457
+ },
2458
+ "llama_response_error_minus_rw": {
2459
+ "mean": 0.13225847396847604,
2460
+ "ci95": [
2461
+ 0.12028988578559921,
2462
+ 0.1444736162391249
2463
+ ]
2464
+ },
2465
+ "llama_delta_nll_minus_rw": {
2466
+ "mean": -1.0928351998288062,
2467
+ "ci95": [
2468
+ -1.1400206526131709,
2469
+ -1.0440431147716018
2470
+ ]
2471
+ },
2472
+ "llama_intact_nll_minus_gru": {
2473
+ "mean": -0.02039762240595883,
2474
+ "ci95": [
2475
+ -0.026885908353433494,
2476
+ -0.015150904406027703
2477
+ ]
2478
+ },
2479
+ "llama_response_error_minus_gru": {
2480
+ "mean": 0.03825855686801073,
2481
+ "ci95": [
2482
+ 0.034273034001589354,
2483
+ 0.04249493798013469
2484
+ ]
2485
+ },
2486
+ "llama_delta_nll_minus_gru": {
2487
+ "mean": -0.2639072505300634,
2488
+ "ci95": [
2489
+ -0.278195858213479,
2490
+ -0.2503508002045456
2491
+ ]
2492
+ },
2493
+ "llama_intact_nll_minus_transformer": {
2494
+ "mean": -0.07078072604293115,
2495
+ "ci95": [
2496
+ -0.08061262707565196,
2497
+ -0.06079880706465013
2498
+ ]
2499
+ },
2500
+ "llama_response_error_minus_transformer": {
2501
+ "mean": -0.0534521947045942,
2502
+ "ci95": [
2503
+ -0.0575724754484284,
2504
+ -0.04925699080755236
2505
+ ]
2506
+ },
2507
+ "llama_delta_nll_minus_transformer": {
2508
+ "mean": 0.10997030473328238,
2509
+ "ci95": [
2510
+ 0.0992116317615521,
2511
+ 0.12104390484589093
2512
+ ]
2513
+ }
2514
+ }
2515
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/restless/COMPLETE.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "split": "test",
3
+ "test_used": true,
4
+ "scientific_audit_passed": true,
5
+ "main_complete": true,
6
+ "models": [
7
+ "oracle",
8
+ "fit",
9
+ "rw",
10
+ "gru",
11
+ "transformer",
12
+ "llama"
13
+ ],
14
+ "weights": [
15
+ "000",
16
+ "010",
17
+ "030",
18
+ "050",
19
+ "070",
20
+ "090",
21
+ "100"
22
+ ],
23
+ "donors": 20,
24
+ "participants_per_weight": 250,
25
+ "scoring": "trials151-200",
26
+ "appendix_operations_included": false
27
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/A_alltrials_summary.json ADDED
@@ -0,0 +1,422 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "all": {
4
+ "oracle_nll": {
5
+ "mean": 2.2407917981264505,
6
+ "ci95": [
7
+ 2.163193930535018,
8
+ 2.3164335192815635
9
+ ]
10
+ },
11
+ "rw_nll": {
12
+ "mean": 3.218557253060558,
13
+ "ci95": [
14
+ 3.216544274939429,
15
+ 3.2205419387733265
16
+ ]
17
+ },
18
+ "gru_full_nll": {
19
+ "mean": 2.282470226287842,
20
+ "ci95": [
21
+ 2.2045391619205477,
22
+ 2.3594693958759305
23
+ ]
24
+ },
25
+ "gru_choice_nll": {
26
+ "mean": 2.2800965309143066,
27
+ "ci95": [
28
+ 2.2032397747039796,
29
+ 2.3564850211143495
30
+ ]
31
+ },
32
+ "transformer_full_nll": {
33
+ "mean": 2.336376428604126,
34
+ "ci95": [
35
+ 2.257382333278656,
36
+ 2.414495450258255
37
+ ]
38
+ },
39
+ "transformer_choice_nll": {
40
+ "mean": 2.289947986602783,
41
+ "ci95": [
42
+ 2.210496240854263,
43
+ 2.368713545799255
44
+ ]
45
+ },
46
+ "llama_full_nll": {
47
+ "mean": 2.2788708803286406,
48
+ "ci95": [
49
+ 2.2027046180530485,
50
+ 2.354693792307771
51
+ ]
52
+ },
53
+ "llama_choice_nll": {
54
+ "mean": 2.2800295143043012,
55
+ "ci95": [
56
+ 2.203834352390601,
57
+ 2.3557167702587267
58
+ ]
59
+ }
60
+ }
61
+ },
62
+ "0.1": {
63
+ "all": {
64
+ "oracle_nll": {
65
+ "mean": 2.330756310644509,
66
+ "ci95": [
67
+ 2.260268593453287,
68
+ 2.398793129249143
69
+ ]
70
+ },
71
+ "rw_nll": {
72
+ "mean": 3.1921232598320524,
73
+ "ci95": [
74
+ 3.1761472110491833,
75
+ 3.2069216444786224
76
+ ]
77
+ },
78
+ "gru_full_nll": {
79
+ "mean": 2.3964734077453613,
80
+ "ci95": [
81
+ 2.3256391286849976,
82
+ 2.468572771549225
83
+ ]
84
+ },
85
+ "gru_choice_nll": {
86
+ "mean": 2.3968405723571777,
87
+ "ci95": [
88
+ 2.3259042263031007,
89
+ 2.468537390232086
90
+ ]
91
+ },
92
+ "transformer_full_nll": {
93
+ "mean": 2.463813543319702,
94
+ "ci95": [
95
+ 2.394661957025528,
96
+ 2.5326750755310057
97
+ ]
98
+ },
99
+ "transformer_choice_nll": {
100
+ "mean": 2.4061079025268555,
101
+ "ci95": [
102
+ 2.3339729428291323,
103
+ 2.477258688211441
104
+ ]
105
+ },
106
+ "llama_full_nll": {
107
+ "mean": 2.38630148940712,
108
+ "ci95": [
109
+ 2.3179021679193244,
110
+ 2.4536329552842058
111
+ ]
112
+ },
113
+ "llama_choice_nll": {
114
+ "mean": 2.392797034129502,
115
+ "ci95": [
116
+ 2.3226687547375993,
117
+ 2.4609082533837534
118
+ ]
119
+ }
120
+ }
121
+ },
122
+ "0.3": {
123
+ "all": {
124
+ "oracle_nll": {
125
+ "mean": 2.380004840600781,
126
+ "ci95": [
127
+ 2.3185715596121215,
128
+ 2.438883334185901
129
+ ]
130
+ },
131
+ "rw_nll": {
132
+ "mean": 3.0269783707407356,
133
+ "ci95": [
134
+ 2.989489158780852,
135
+ 3.063557661821549
136
+ ]
137
+ },
138
+ "gru_full_nll": {
139
+ "mean": 2.5049357414245605,
140
+ "ci95": [
141
+ 2.4476216971874236,
142
+ 2.5613901913166046
143
+ ]
144
+ },
145
+ "gru_choice_nll": {
146
+ "mean": 2.5335240364074707,
147
+ "ci95": [
148
+ 2.4753704369068146,
149
+ 2.588935452699661
150
+ ]
151
+ },
152
+ "transformer_full_nll": {
153
+ "mean": 2.6127402782440186,
154
+ "ci95": [
155
+ 2.557130444049835,
156
+ 2.6683213293552397
157
+ ]
158
+ },
159
+ "transformer_choice_nll": {
160
+ "mean": 2.539529323577881,
161
+ "ci95": [
162
+ 2.4802891492843626,
163
+ 2.5979774057865144
164
+ ]
165
+ },
166
+ "llama_full_nll": {
167
+ "mean": 2.4477569201233584,
168
+ "ci95": [
169
+ 2.388281041264481,
170
+ 2.5062818950123322
171
+ ]
172
+ },
173
+ "llama_choice_nll": {
174
+ "mean": 2.5248020348266205,
175
+ "ci95": [
176
+ 2.468299495374132,
177
+ 2.580356526043874
178
+ ]
179
+ }
180
+ }
181
+ },
182
+ "0.5": {
183
+ "all": {
184
+ "oracle_nll": {
185
+ "mean": 2.2971771709347775,
186
+ "ci95": [
187
+ 2.2289068897899256,
188
+ 2.365295172115381
189
+ ]
190
+ },
191
+ "rw_nll": {
192
+ "mean": 2.7893531293244047,
193
+ "ci95": [
194
+ 2.733724785921244,
195
+ 2.8447403324878717
196
+ ]
197
+ },
198
+ "gru_full_nll": {
199
+ "mean": 2.4463932514190674,
200
+ "ci95": [
201
+ 2.381036365032196,
202
+ 2.514950156211853
203
+ ]
204
+ },
205
+ "gru_choice_nll": {
206
+ "mean": 2.5365405082702637,
207
+ "ci95": [
208
+ 2.473630279302597,
209
+ 2.6024184048175814
210
+ ]
211
+ },
212
+ "transformer_full_nll": {
213
+ "mean": 2.6039299964904785,
214
+ "ci95": [
215
+ 2.540676635503769,
216
+ 2.666206884384155
217
+ ]
218
+ },
219
+ "transformer_choice_nll": {
220
+ "mean": 2.5465657711029053,
221
+ "ci95": [
222
+ 2.4822121918201447,
223
+ 2.6125925302505495
224
+ ]
225
+ },
226
+ "llama_full_nll": {
227
+ "mean": 2.3891987121593683,
228
+ "ci95": [
229
+ 2.3208481151822333,
230
+ 2.457557690497194
231
+ ]
232
+ },
233
+ "llama_choice_nll": {
234
+ "mean": 2.5117965114270207,
235
+ "ci95": [
236
+ 2.4475009707087483,
237
+ 2.5776022411478783
238
+ ]
239
+ }
240
+ }
241
+ },
242
+ "0.7": {
243
+ "all": {
244
+ "oracle_nll": {
245
+ "mean": 2.170521562928485,
246
+ "ci95": [
247
+ 2.091652480641617,
248
+ 2.2470193000135064
249
+ ]
250
+ },
251
+ "rw_nll": {
252
+ "mean": 2.5741874694421307,
253
+ "ci95": [
254
+ 2.5040318477379047,
255
+ 2.6442251828075176
256
+ ]
257
+ },
258
+ "gru_full_nll": {
259
+ "mean": 2.3510477542877197,
260
+ "ci95": [
261
+ 2.270873934030533,
262
+ 2.431276684999466
263
+ ]
264
+ },
265
+ "gru_choice_nll": {
266
+ "mean": 2.4570748805999756,
267
+ "ci95": [
268
+ 2.3789141952991484,
269
+ 2.535234421491623
270
+ ]
271
+ },
272
+ "transformer_full_nll": {
273
+ "mean": 2.5160937309265137,
274
+ "ci95": [
275
+ 2.4342929780483247,
276
+ 2.5965543031692504
277
+ ]
278
+ },
279
+ "transformer_choice_nll": {
280
+ "mean": 2.472677230834961,
281
+ "ci95": [
282
+ 2.393271601200104,
283
+ 2.551168769598007
284
+ ]
285
+ },
286
+ "llama_full_nll": {
287
+ "mean": 2.2725123322046357,
288
+ "ci95": [
289
+ 2.196239453642533,
290
+ 2.350747435712886
291
+ ]
292
+ },
293
+ "llama_choice_nll": {
294
+ "mean": 2.441380426501888,
295
+ "ci95": [
296
+ 2.362496420370555,
297
+ 2.520071810296925
298
+ ]
299
+ }
300
+ }
301
+ },
302
+ "0.9": {
303
+ "all": {
304
+ "oracle_nll": {
305
+ "mean": 2.013132844700362,
306
+ "ci95": [
307
+ 1.9232147862168596,
308
+ 2.098767286998288
309
+ ]
310
+ },
311
+ "rw_nll": {
312
+ "mean": 2.3486296728962732,
313
+ "ci95": [
314
+ 2.261578854035819,
315
+ 2.4343474125207134
316
+ ]
317
+ },
318
+ "gru_full_nll": {
319
+ "mean": 2.2091283798217773,
320
+ "ci95": [
321
+ 2.1179920732975006,
322
+ 2.2970135033130648
323
+ ]
324
+ },
325
+ "gru_choice_nll": {
326
+ "mean": 2.364687919616699,
327
+ "ci95": [
328
+ 2.2731286764144896,
329
+ 2.451507943868637
330
+ ]
331
+ },
332
+ "transformer_full_nll": {
333
+ "mean": 2.4155631065368652,
334
+ "ci95": [
335
+ 2.325894796848297,
336
+ 2.5016388833522796
337
+ ]
338
+ },
339
+ "transformer_choice_nll": {
340
+ "mean": 2.379507064819336,
341
+ "ci95": [
342
+ 2.289050740003586,
343
+ 2.4666596949100494
344
+ ]
345
+ },
346
+ "llama_full_nll": {
347
+ "mean": 2.1173633001839693,
348
+ "ci95": [
349
+ 2.0277182092303234,
350
+ 2.202218445814759
351
+ ]
352
+ },
353
+ "llama_choice_nll": {
354
+ "mean": 2.3389426888290936,
355
+ "ci95": [
356
+ 2.2494154959976362,
357
+ 2.4250590988043634
358
+ ]
359
+ }
360
+ }
361
+ },
362
+ "1.0": {
363
+ "all": {
364
+ "oracle_nll": {
365
+ "mean": 1.8731762803574148,
366
+ "ci95": [
367
+ 1.777315062035161,
368
+ 1.970925304203498
369
+ ]
370
+ },
371
+ "rw_nll": {
372
+ "mean": 2.209035875549125,
373
+ "ci95": [
374
+ 2.110485823811036,
375
+ 2.312573675779551
376
+ ]
377
+ },
378
+ "gru_full_nll": {
379
+ "mean": 2.0877323150634766,
380
+ "ci95": [
381
+ 1.986038053035736,
382
+ 2.1904436349868774
383
+ ]
384
+ },
385
+ "gru_choice_nll": {
386
+ "mean": 2.258831262588501,
387
+ "ci95": [
388
+ 2.1564736545085905,
389
+ 2.3602626383304592
390
+ ]
391
+ },
392
+ "transformer_full_nll": {
393
+ "mean": 2.300532341003418,
394
+ "ci95": [
395
+ 2.198591870069504,
396
+ 2.4023356974124908
397
+ ]
398
+ },
399
+ "transformer_choice_nll": {
400
+ "mean": 2.280344009399414,
401
+ "ci95": [
402
+ 2.1749097287654875,
403
+ 2.3857450127601623
404
+ ]
405
+ },
406
+ "llama_full_nll": {
407
+ "mean": 1.982778057523228,
408
+ "ci95": [
409
+ 1.8845374558567696,
410
+ 2.0811499026519296
411
+ ]
412
+ },
413
+ "llama_choice_nll": {
414
+ "mean": 2.2229474962998244,
415
+ "ci95": [
416
+ 2.119441581737968,
417
+ 2.3262832757753276
418
+ ]
419
+ }
420
+ }
421
+ }
422
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/A_summary.json ADDED
@@ -0,0 +1,1402 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "all": {
4
+ "oracle_intact_nll": {
5
+ "mean": 2.256899081853402,
6
+ "ci95": [
7
+ 2.1539276579886697,
8
+ 2.3536397088713543
9
+ ]
10
+ },
11
+ "oracle_donor_nll": {
12
+ "mean": 2.256899081853402,
13
+ "ci95": [
14
+ 2.1539276579886697,
15
+ 2.3536397088713543
16
+ ]
17
+ },
18
+ "oracle_delta_nll": {
19
+ "mean": 3.552713678800501e-17,
20
+ "ci95": [
21
+ -1.6431300764452317e-16,
22
+ 2.353672812205332e-16
23
+ ]
24
+ },
25
+ "oracle_response_error": {
26
+ "mean": 0.0,
27
+ "ci95": [
28
+ 0.0,
29
+ 0.0
30
+ ]
31
+ },
32
+ "fit_intact_nll": {
33
+ "mean": 2.2665208854580676,
34
+ "ci95": [
35
+ 2.165162571793024,
36
+ 2.3635493247560464
37
+ ]
38
+ },
39
+ "fit_donor_nll": {
40
+ "mean": 2.266188960502596,
41
+ "ci95": [
42
+ 2.1646592445093957,
43
+ 2.361894910786617
44
+ ]
45
+ },
46
+ "fit_delta_nll": {
47
+ "mean": -0.00033192495547172294,
48
+ "ci95": [
49
+ -0.0029861019709051927,
50
+ 0.0025254340726127834
51
+ ]
52
+ },
53
+ "fit_response_error": {
54
+ "mean": 0.010783533619515567,
55
+ "ci95": [
56
+ 0.006974173524976919,
57
+ 0.015135500302760889
58
+ ]
59
+ },
60
+ "rw_intact_nll": {
61
+ "mean": 3.2175559090104016,
62
+ "ci95": [
63
+ 3.213504313467025,
64
+ 3.22121403483305
65
+ ]
66
+ },
67
+ "rw_donor_nll": {
68
+ "mean": 3.2189942581169397,
69
+ "ci95": [
70
+ 3.2181489329875665,
71
+ 3.2197627748496
72
+ ]
73
+ },
74
+ "rw_delta_nll": {
75
+ "mean": 0.0014383491065387587,
76
+ "ci95": [
77
+ -0.0019054543112809552,
78
+ 0.005397353097600356
79
+ ]
80
+ },
81
+ "rw_response_error": {
82
+ "mean": 0.0065546631610965255,
83
+ "ci95": [
84
+ 0.006253397299093739,
85
+ 0.006874478198303794
86
+ ]
87
+ },
88
+ "gru_intact_nll": {
89
+ "mean": 2.3003609185091234,
90
+ "ci95": [
91
+ 2.198413620263349,
92
+ 2.400978339217044
93
+ ]
94
+ },
95
+ "gru_donor_nll": {
96
+ "mean": 2.299973465268989,
97
+ "ci95": [
98
+ 2.1976383500693397,
99
+ 2.400265694486215
100
+ ]
101
+ },
102
+ "gru_delta_nll": {
103
+ "mean": -0.0003874532401340769,
104
+ "ci95": [
105
+ -0.0014255719779839957,
106
+ 0.0007252131943553924
107
+ ]
108
+ },
109
+ "gru_response_error": {
110
+ "mean": 0.018504102779924144,
111
+ "ci95": [
112
+ 0.018051273358047692,
113
+ 0.018952429187911405
114
+ ]
115
+ },
116
+ "gru_full_nll": {
117
+ "mean": 2.3003609185091234,
118
+ "ci95": [
119
+ 2.198413620263349,
120
+ 2.400978339217044
121
+ ]
122
+ },
123
+ "gru_choice_nll": {
124
+ "mean": 2.2997417918175467,
125
+ "ci95": [
126
+ 2.198184169213947,
127
+ 2.396640228449533
128
+ ]
129
+ },
130
+ "transformer_intact_nll": {
131
+ "mean": 2.360867912326785,
132
+ "ci95": [
133
+ 2.259512517521976,
134
+ 2.4570797715249877
135
+ ]
136
+ },
137
+ "transformer_donor_nll": {
138
+ "mean": 2.3535000802925157,
139
+ "ci95": [
140
+ 2.252967515338882,
141
+ 2.449982839537024
142
+ ]
143
+ },
144
+ "transformer_delta_nll": {
145
+ "mean": -0.007367832034269306,
146
+ "ci95": [
147
+ -0.02186180228286032,
148
+ 0.00516436291414688
149
+ ]
150
+ },
151
+ "transformer_response_error": {
152
+ "mean": 0.1170145991357271,
153
+ "ci95": [
154
+ 0.11260599481564193,
155
+ 0.1216407268660003
156
+ ]
157
+ },
158
+ "transformer_full_nll": {
159
+ "mean": 2.360867912326785,
160
+ "ci95": [
161
+ 2.259512517521976,
162
+ 2.4570797715249877
163
+ ]
164
+ },
165
+ "transformer_choice_nll": {
166
+ "mean": 2.3071931246818678,
167
+ "ci95": [
168
+ 2.201677337007935,
169
+ 2.4067951834649857
170
+ ]
171
+ },
172
+ "oracle_nll": {
173
+ "mean": 2.256899081853402,
174
+ "ci95": [
175
+ 2.1539276579886697,
176
+ 2.3536397088713543
177
+ ]
178
+ },
179
+ "llama_full_nll": {
180
+ "mean": 2.2982575594719052,
181
+ "ci95": [
182
+ 2.195976598373851,
183
+ 2.3952368262662347
184
+ ]
185
+ },
186
+ "llama_choice_nll": {
187
+ "mean": 2.3024848628881505,
188
+ "ci95": [
189
+ 2.2006664714056825,
190
+ 2.398696368876586
191
+ ]
192
+ },
193
+ "llama_reward_predictive_gain": {
194
+ "mean": 0.004227303416245162,
195
+ "ci95": [
196
+ -0.001178904395716497,
197
+ 0.009791267599562867
198
+ ]
199
+ }
200
+ }
201
+ },
202
+ "0.1": {
203
+ "all": {
204
+ "oracle_intact_nll": {
205
+ "mean": 2.358300456778261,
206
+ "ci95": [
207
+ 2.2629998556727573,
208
+ 2.4481495877709705
209
+ ]
210
+ },
211
+ "oracle_donor_nll": {
212
+ "mean": 2.4054874510008455,
213
+ "ci95": [
214
+ 2.3087123638900544,
215
+ 2.5022874313472685
216
+ ]
217
+ },
218
+ "oracle_delta_nll": {
219
+ "mean": 0.04718699422258422,
220
+ "ci95": [
221
+ 0.03436657494213905,
222
+ 0.05970594641691559
223
+ ]
224
+ },
225
+ "oracle_response_error": {
226
+ "mean": 0.0,
227
+ "ci95": [
228
+ 0.0,
229
+ 0.0
230
+ ]
231
+ },
232
+ "fit_intact_nll": {
233
+ "mean": 2.3795351204281063,
234
+ "ci95": [
235
+ 2.2851009583330377,
236
+ 2.4709300500665647
237
+ ]
238
+ },
239
+ "fit_donor_nll": {
240
+ "mean": 2.4210424329847204,
241
+ "ci95": [
242
+ 2.3238395816470176,
243
+ 2.517420854643946
244
+ ]
245
+ },
246
+ "fit_delta_nll": {
247
+ "mean": 0.04150731255661429,
248
+ "ci95": [
249
+ 0.026934889627137568,
250
+ 0.0564772909759607
251
+ ]
252
+ },
253
+ "fit_response_error": {
254
+ "mean": 0.052102911681065664,
255
+ "ci95": [
256
+ 0.0450136975626096,
257
+ 0.06001172458386209
258
+ ]
259
+ },
260
+ "rw_intact_nll": {
261
+ "mean": 3.189361391632845,
262
+ "ci95": [
263
+ 3.153220666502596,
264
+ 3.2250492036221883
265
+ ]
266
+ },
267
+ "rw_donor_nll": {
268
+ "mean": 3.2160267405427345,
269
+ "ci95": [
270
+ 3.208121107134707,
271
+ 3.2233383354985814
272
+ ]
273
+ },
274
+ "rw_delta_nll": {
275
+ "mean": 0.02666534890988978,
276
+ "ci95": [
277
+ -0.005695483999127539,
278
+ 0.0591333235912339
279
+ ]
280
+ },
281
+ "rw_response_error": {
282
+ "mean": 0.08648257221904393,
283
+ "ci95": [
284
+ 0.08208808266072022,
285
+ 0.09081407695981439
286
+ ]
287
+ },
288
+ "gru_intact_nll": {
289
+ "mean": 2.4251186469737713,
290
+ "ci95": [
291
+ 2.3284747965202945,
292
+ 2.524579672921996
293
+ ]
294
+ },
295
+ "gru_donor_nll": {
296
+ "mean": 2.426071139402867,
297
+ "ci95": [
298
+ 2.329496817277385,
299
+ 2.5249042540418163
300
+ ]
301
+ },
302
+ "gru_delta_nll": {
303
+ "mean": 0.0009524924290957977,
304
+ "ci95": [
305
+ -0.0007303317010866242,
306
+ 0.002569258607637665
307
+ ]
308
+ },
309
+ "gru_response_error": {
310
+ "mean": 0.07014626407047886,
311
+ "ci95": [
312
+ 0.0678662056582069,
313
+ 0.07250281904867092
314
+ ]
315
+ },
316
+ "gru_full_nll": {
317
+ "mean": 2.4251186469737713,
318
+ "ci95": [
319
+ 2.3284747965202945,
320
+ 2.524579672921996
321
+ ]
322
+ },
323
+ "gru_choice_nll": {
324
+ "mean": 2.423530669573678,
325
+ "ci95": [
326
+ 2.3258626021057913,
327
+ 2.5209163679798374
328
+ ]
329
+ },
330
+ "transformer_intact_nll": {
331
+ "mean": 2.5021050715922293,
332
+ "ci95": [
333
+ 2.4071664429399253,
334
+ 2.595780298187066
335
+ ]
336
+ },
337
+ "transformer_donor_nll": {
338
+ "mean": 2.4970589588915537,
339
+ "ci95": [
340
+ 2.4001181301544525,
341
+ 2.593694670124522
342
+ ]
343
+ },
344
+ "transformer_delta_nll": {
345
+ "mean": -0.0050461127006755715,
346
+ "ci95": [
347
+ -0.022723500215269178,
348
+ 0.009568835278212057
349
+ ]
350
+ },
351
+ "transformer_response_error": {
352
+ "mean": 0.15332741715503675,
353
+ "ci95": [
354
+ 0.14787185484306695,
355
+ 0.15920991345423632
356
+ ]
357
+ },
358
+ "transformer_full_nll": {
359
+ "mean": 2.5021050715922293,
360
+ "ci95": [
361
+ 2.4071664429399253,
362
+ 2.595780298187066
363
+ ]
364
+ },
365
+ "transformer_choice_nll": {
366
+ "mean": 2.4344931105129946,
367
+ "ci95": [
368
+ 2.3364925924441855,
369
+ 2.533731411874208
370
+ ]
371
+ },
372
+ "oracle_nll": {
373
+ "mean": 2.358300456778261,
374
+ "ci95": [
375
+ 2.2629998556727573,
376
+ 2.4481495877709705
377
+ ]
378
+ },
379
+ "llama_full_nll": {
380
+ "mean": 2.4158071960705993,
381
+ "ci95": [
382
+ 2.3239026143343584,
383
+ 2.5053612041975044
384
+ ]
385
+ },
386
+ "llama_choice_nll": {
387
+ "mean": 2.422510292332979,
388
+ "ci95": [
389
+ 2.32652689505674,
390
+ 2.5194355990993675
391
+ ]
392
+ },
393
+ "llama_reward_predictive_gain": {
394
+ "mean": 0.006703096262379908,
395
+ "ci95": [
396
+ -0.006473624100292907,
397
+ 0.02141005324122265
398
+ ]
399
+ }
400
+ }
401
+ },
402
+ "0.3": {
403
+ "all": {
404
+ "oracle_intact_nll": {
405
+ "mean": 2.4031890757538656,
406
+ "ci95": [
407
+ 2.3080410147812622,
408
+ 2.493060873165434
409
+ ]
410
+ },
411
+ "oracle_donor_nll": {
412
+ "mean": 2.618646521081867,
413
+ "ci95": [
414
+ 2.527502915350205,
415
+ 2.700554180347221
416
+ ]
417
+ },
418
+ "oracle_delta_nll": {
419
+ "mean": 0.21545744532800146,
420
+ "ci95": [
421
+ 0.18986788030576535,
422
+ 0.24149767842861766
423
+ ]
424
+ },
425
+ "oracle_response_error": {
426
+ "mean": 0.0,
427
+ "ci95": [
428
+ 0.0,
429
+ 0.0
430
+ ]
431
+ },
432
+ "fit_intact_nll": {
433
+ "mean": 2.41717893155207,
434
+ "ci95": [
435
+ 2.322061230563267,
436
+ 2.505455425612881
437
+ ]
438
+ },
439
+ "fit_donor_nll": {
440
+ "mean": 2.6242176974420066,
441
+ "ci95": [
442
+ 2.5317101501606807,
443
+ 2.7095291421819456
444
+ ]
445
+ },
446
+ "fit_delta_nll": {
447
+ "mean": 0.20703876588993686,
448
+ "ci95": [
449
+ 0.17986128071058735,
450
+ 0.23594040419722156
451
+ ]
452
+ },
453
+ "fit_response_error": {
454
+ "mean": 0.0691367955015857,
455
+ "ci95": [
456
+ 0.059267532699285544,
457
+ 0.07887737741055921
458
+ ]
459
+ },
460
+ "rw_intact_nll": {
461
+ "mean": 3.0243127705433683,
462
+ "ci95": [
463
+ 2.9625603901177726,
464
+ 3.083535840993125
465
+ ]
466
+ },
467
+ "rw_donor_nll": {
468
+ "mean": 3.203483069337998,
469
+ "ci95": [
470
+ 3.1824372245788966,
471
+ 3.2239390213336447
472
+ ]
473
+ },
474
+ "rw_delta_nll": {
475
+ "mean": 0.17917029879463012,
476
+ "ci95": [
477
+ 0.13433822314957689,
478
+ 0.22693241171930228
479
+ ]
480
+ },
481
+ "rw_response_error": {
482
+ "mean": 0.21432052268495294,
483
+ "ci95": [
484
+ 0.20350847432596209,
485
+ 0.22558303850567366
486
+ ]
487
+ },
488
+ "gru_intact_nll": {
489
+ "mean": 2.524001883635677,
490
+ "ci95": [
491
+ 2.4321140389831455,
492
+ 2.6099826043107632
493
+ ]
494
+ },
495
+ "gru_donor_nll": {
496
+ "mean": 2.571780270978965,
497
+ "ci95": [
498
+ 2.4773468057793844,
499
+ 2.6560668792585798
500
+ ]
501
+ },
502
+ "gru_delta_nll": {
503
+ "mean": 0.0477783873432883,
504
+ "ci95": [
505
+ 0.03760635238477913,
506
+ 0.05711140657155545
507
+ ]
508
+ },
509
+ "gru_response_error": {
510
+ "mean": 0.20187495556609986,
511
+ "ci95": [
512
+ 0.19484867770261477,
513
+ 0.2084915060931258
514
+ ]
515
+ },
516
+ "gru_full_nll": {
517
+ "mean": 2.524001883635677,
518
+ "ci95": [
519
+ 2.4321140389831455,
520
+ 2.6099826043107632
521
+ ]
522
+ },
523
+ "gru_choice_nll": {
524
+ "mean": 2.550332199199586,
525
+ "ci95": [
526
+ 2.458054609892982,
527
+ 2.632661163300756
528
+ ]
529
+ },
530
+ "transformer_intact_nll": {
531
+ "mean": 2.6372881952925993,
532
+ "ci95": [
533
+ 2.546320653589055,
534
+ 2.722487010986659
535
+ ]
536
+ },
537
+ "transformer_donor_nll": {
538
+ "mean": 2.6450593082988045,
539
+ "ci95": [
540
+ 2.5556576753880242,
541
+ 2.7302018826026617
542
+ ]
543
+ },
544
+ "transformer_delta_nll": {
545
+ "mean": 0.007771113006204906,
546
+ "ci95": [
547
+ -0.006287134574327571,
548
+ 0.020174173096946673
549
+ ]
550
+ },
551
+ "transformer_response_error": {
552
+ "mean": 0.2640845588027949,
553
+ "ci95": [
554
+ 0.2570148489025373,
555
+ 0.27070137321903603
556
+ ]
557
+ },
558
+ "transformer_full_nll": {
559
+ "mean": 2.6372881952925993,
560
+ "ci95": [
561
+ 2.546320653589055,
562
+ 2.722487010986659
563
+ ]
564
+ },
565
+ "transformer_choice_nll": {
566
+ "mean": 2.554989409750647,
567
+ "ci95": [
568
+ 2.461253575379825,
569
+ 2.6405940931788856
570
+ ]
571
+ },
572
+ "oracle_nll": {
573
+ "mean": 2.4031890757538656,
574
+ "ci95": [
575
+ 2.3080410147812622,
576
+ 2.493060873165434
577
+ ]
578
+ },
579
+ "llama_full_nll": {
580
+ "mean": 2.466419687330082,
581
+ "ci95": [
582
+ 2.372019483451541,
583
+ 2.554812335193939
584
+ ]
585
+ },
586
+ "llama_choice_nll": {
587
+ "mean": 2.534494785695525,
588
+ "ci95": [
589
+ 2.4440944376267026,
590
+ 2.620180488364762
591
+ ]
592
+ },
593
+ "llama_reward_predictive_gain": {
594
+ "mean": 0.06807509836544341,
595
+ "ci95": [
596
+ 0.04992199862100402,
597
+ 0.08532290762912904
598
+ ]
599
+ }
600
+ }
601
+ },
602
+ "0.5": {
603
+ "all": {
604
+ "oracle_intact_nll": {
605
+ "mean": 2.319283828723959,
606
+ "ci95": [
607
+ 2.215970742786679,
608
+ 2.4251377502112526
609
+ ]
610
+ },
611
+ "oracle_donor_nll": {
612
+ "mean": 2.7559770217009416,
613
+ "ci95": [
614
+ 2.6745745656011968,
615
+ 2.836698169145955
616
+ ]
617
+ },
618
+ "oracle_delta_nll": {
619
+ "mean": 0.43669319297698217,
620
+ "ci95": [
621
+ 0.3936526854652242,
622
+ 0.4783219587769138
623
+ ]
624
+ },
625
+ "oracle_response_error": {
626
+ "mean": 0.0,
627
+ "ci95": [
628
+ 0.0,
629
+ 0.0
630
+ ]
631
+ },
632
+ "fit_intact_nll": {
633
+ "mean": 2.342043599243521,
634
+ "ci95": [
635
+ 2.2380130079444327,
636
+ 2.4455075486523197
637
+ ]
638
+ },
639
+ "fit_donor_nll": {
640
+ "mean": 2.7809330530121543,
641
+ "ci95": [
642
+ 2.697637756052261,
643
+ 2.8650316810601226
644
+ ]
645
+ },
646
+ "fit_delta_nll": {
647
+ "mean": 0.43888945376863364,
648
+ "ci95": [
649
+ 0.3955619119800974,
650
+ 0.4823279484351039
651
+ ]
652
+ },
653
+ "fit_response_error": {
654
+ "mean": 0.07216533259262915,
655
+ "ci95": [
656
+ 0.061967845980080045,
657
+ 0.08186041789541766
658
+ ]
659
+ },
660
+ "rw_intact_nll": {
661
+ "mean": 2.805631640622673,
662
+ "ci95": [
663
+ 2.717358818971326,
664
+ 2.8872822320788982
665
+ ]
666
+ },
667
+ "rw_donor_nll": {
668
+ "mean": 3.1864433123704146,
669
+ "ci95": [
670
+ 3.148402000229726,
671
+ 3.2230480775807617
672
+ ]
673
+ },
674
+ "rw_delta_nll": {
675
+ "mean": 0.3808116717477414,
676
+ "ci95": [
677
+ 0.3250388310644342,
678
+ 0.4398623790768384
679
+ ]
680
+ },
681
+ "rw_response_error": {
682
+ "mean": 0.31536258468027717,
683
+ "ci95": [
684
+ 0.2983742329804894,
685
+ 0.33353337030248736
686
+ ]
687
+ },
688
+ "gru_intact_nll": {
689
+ "mean": 2.4698075510185014,
690
+ "ci95": [
691
+ 2.369894730087947,
692
+ 2.5732815028055542
693
+ ]
694
+ },
695
+ "gru_donor_nll": {
696
+ "mean": 2.6250467260884376,
697
+ "ci95": [
698
+ 2.5314117240209546,
699
+ 2.721516455584175
700
+ ]
701
+ },
702
+ "gru_delta_nll": {
703
+ "mean": 0.15523917506993587,
704
+ "ci95": [
705
+ 0.1397050799581931,
706
+ 0.17083378776948413
707
+ ]
708
+ },
709
+ "gru_response_error": {
710
+ "mean": 0.2876848040377506,
711
+ "ci95": [
712
+ 0.27845440241345654,
713
+ 0.2974706192101464
714
+ ]
715
+ },
716
+ "gru_full_nll": {
717
+ "mean": 2.4698075510185014,
718
+ "ci95": [
719
+ 2.369894730087947,
720
+ 2.5732815028055542
721
+ ]
722
+ },
723
+ "gru_choice_nll": {
724
+ "mean": 2.560591917239732,
725
+ "ci95": [
726
+ 2.4642544195651266,
727
+ 2.6602050281172898
728
+ ]
729
+ },
730
+ "transformer_intact_nll": {
731
+ "mean": 2.6154398084941217,
732
+ "ci95": [
733
+ 2.5168079947367183,
734
+ 2.7138013142448343
735
+ ]
736
+ },
737
+ "transformer_donor_nll": {
738
+ "mean": 2.6315340423144757,
739
+ "ci95": [
740
+ 2.5365447755142534,
741
+ 2.729152883552159
742
+ ]
743
+ },
744
+ "transformer_delta_nll": {
745
+ "mean": 0.016094233820354173,
746
+ "ci95": [
747
+ 0.0035590184174254503,
748
+ 0.028568504439261653
749
+ ]
750
+ },
751
+ "transformer_response_error": {
752
+ "mean": 0.35910503126230187,
753
+ "ci95": [
754
+ 0.3508898449212565,
755
+ 0.36753351884435065
756
+ ]
757
+ },
758
+ "transformer_full_nll": {
759
+ "mean": 2.6154398084941217,
760
+ "ci95": [
761
+ 2.5168079947367183,
762
+ 2.7138013142448343
763
+ ]
764
+ },
765
+ "transformer_choice_nll": {
766
+ "mean": 2.5741605505376075,
767
+ "ci95": [
768
+ 2.474338639043075,
769
+ 2.6771158429884063
770
+ ]
771
+ },
772
+ "oracle_nll": {
773
+ "mean": 2.319283828723959,
774
+ "ci95": [
775
+ 2.215970742786679,
776
+ 2.4251377502112526
777
+ ]
778
+ },
779
+ "llama_full_nll": {
780
+ "mean": 2.414704988275222,
781
+ "ci95": [
782
+ 2.3157662530219927,
783
+ 2.5174059327438485
784
+ ]
785
+ },
786
+ "llama_choice_nll": {
787
+ "mean": 2.544273042956066,
788
+ "ci95": [
789
+ 2.4475386055396084,
790
+ 2.643421831369369
791
+ ]
792
+ },
793
+ "llama_reward_predictive_gain": {
794
+ "mean": 0.12956805468084437,
795
+ "ci95": [
796
+ 0.11431891063622074,
797
+ 0.14454876293835775
798
+ ]
799
+ }
800
+ }
801
+ },
802
+ "0.7": {
803
+ "all": {
804
+ "oracle_intact_nll": {
805
+ "mean": 2.1435564477701416,
806
+ "ci95": [
807
+ 2.0159541343855927,
808
+ 2.2656410205098103
809
+ ]
810
+ },
811
+ "oracle_donor_nll": {
812
+ "mean": 2.871062122680968,
813
+ "ci95": [
814
+ 2.771050651290626,
815
+ 2.9651100152575345
816
+ ]
817
+ },
818
+ "oracle_delta_nll": {
819
+ "mean": 0.7275056749108267,
820
+ "ci95": [
821
+ 0.6755428709723962,
822
+ 0.7837150770637379
823
+ ]
824
+ },
825
+ "oracle_response_error": {
826
+ "mean": 0.0,
827
+ "ci95": [
828
+ 0.0,
829
+ 0.0
830
+ ]
831
+ },
832
+ "fit_intact_nll": {
833
+ "mean": 2.1702145896796368,
834
+ "ci95": [
835
+ 2.044535120248899,
836
+ 2.2913780175881757
837
+ ]
838
+ },
839
+ "fit_donor_nll": {
840
+ "mean": 2.8718458628829295,
841
+ "ci95": [
842
+ 2.767901620071093,
843
+ 2.972574590194788
844
+ ]
845
+ },
846
+ "fit_delta_nll": {
847
+ "mean": 0.7016312732032931,
848
+ "ci95": [
849
+ 0.6527425935612794,
850
+ 0.757010351344706
851
+ ]
852
+ },
853
+ "fit_response_error": {
854
+ "mean": 0.09135141120397736,
855
+ "ci95": [
856
+ 0.07908178328124472,
857
+ 0.10371831445253293
858
+ ]
859
+ },
860
+ "rw_intact_nll": {
861
+ "mean": 2.55631368629174,
862
+ "ci95": [
863
+ 2.4418039685845145,
864
+ 2.6609694359300526
865
+ ]
866
+ },
867
+ "rw_donor_nll": {
868
+ "mean": 3.1023155506208098,
869
+ "ci95": [
870
+ 3.0352899066684635,
871
+ 3.1677688725162674
872
+ ]
873
+ },
874
+ "rw_delta_nll": {
875
+ "mean": 0.5460018643290695,
876
+ "ci95": [
877
+ 0.4891281777276762,
878
+ 0.6077525759115737
879
+ ]
880
+ },
881
+ "rw_response_error": {
882
+ "mean": 0.4015211379717528,
883
+ "ci95": [
884
+ 0.38037476934737585,
885
+ 0.4232246683185908
886
+ ]
887
+ },
888
+ "gru_intact_nll": {
889
+ "mean": 2.3373696046564776,
890
+ "ci95": [
891
+ 2.1997238039068523,
892
+ 2.470009003797405
893
+ ]
894
+ },
895
+ "gru_donor_nll": {
896
+ "mean": 2.575359011909752,
897
+ "ci95": [
898
+ 2.4479691908112295,
899
+ 2.694542565639238
900
+ ]
901
+ },
902
+ "gru_delta_nll": {
903
+ "mean": 0.23798940725327455,
904
+ "ci95": [
905
+ 0.21504852421153092,
906
+ 0.2625664992208941
907
+ ]
908
+ },
909
+ "gru_response_error": {
910
+ "mean": 0.3764652173827853,
911
+ "ci95": [
912
+ 0.3651758577581391,
913
+ 0.38848614637524614
914
+ ]
915
+ },
916
+ "gru_full_nll": {
917
+ "mean": 2.3373696046564776,
918
+ "ci95": [
919
+ 2.1997238039068523,
920
+ 2.470009003797405
921
+ ]
922
+ },
923
+ "gru_choice_nll": {
924
+ "mean": 2.4435253696905175,
925
+ "ci95": [
926
+ 2.312274576152947,
927
+ 2.571105610966659
928
+ ]
929
+ },
930
+ "transformer_intact_nll": {
931
+ "mean": 2.4937030065468395,
932
+ "ci95": [
933
+ 2.360721876964319,
934
+ 2.6284748750254256
935
+ ]
936
+ },
937
+ "transformer_donor_nll": {
938
+ "mean": 2.5077245637863577,
939
+ "ci95": [
940
+ 2.377005762955793,
941
+ 2.6373347475548568
942
+ ]
943
+ },
944
+ "transformer_delta_nll": {
945
+ "mean": 0.014021557239517557,
946
+ "ci95": [
947
+ 0.0025055985462562992,
948
+ 0.024780594294718662
949
+ ]
950
+ },
951
+ "transformer_response_error": {
952
+ "mean": 0.4662228174648522,
953
+ "ci95": [
954
+ 0.4558828834752213,
955
+ 0.47718799263244
956
+ ]
957
+ },
958
+ "transformer_full_nll": {
959
+ "mean": 2.4937030065468395,
960
+ "ci95": [
961
+ 2.360721876964319,
962
+ 2.6284748750254256
963
+ ]
964
+ },
965
+ "transformer_choice_nll": {
966
+ "mean": 2.452637400242134,
967
+ "ci95": [
968
+ 2.323421033681237,
969
+ 2.581084595169812
970
+ ]
971
+ },
972
+ "oracle_nll": {
973
+ "mean": 2.1435564477701416,
974
+ "ci95": [
975
+ 2.0159541343855927,
976
+ 2.2656410205098103
977
+ ]
978
+ },
979
+ "llama_full_nll": {
980
+ "mean": 2.2510137311827587,
981
+ "ci95": [
982
+ 2.122494384717716,
983
+ 2.3720210707871052
984
+ ]
985
+ },
986
+ "llama_choice_nll": {
987
+ "mean": 2.4189901027790803,
988
+ "ci95": [
989
+ 2.2905766582241265,
990
+ 2.5415705206814843
991
+ ]
992
+ },
993
+ "llama_reward_predictive_gain": {
994
+ "mean": 0.1679763715963224,
995
+ "ci95": [
996
+ 0.14920931286576464,
997
+ 0.1865610506148221
998
+ ]
999
+ }
1000
+ }
1001
+ },
1002
+ "0.9": {
1003
+ "all": {
1004
+ "oracle_intact_nll": {
1005
+ "mean": 1.965550659603369,
1006
+ "ci95": [
1007
+ 1.8382024060606479,
1008
+ 2.0891000379175035
1009
+ ]
1010
+ },
1011
+ "oracle_donor_nll": {
1012
+ "mean": 3.077029769914495,
1013
+ "ci95": [
1014
+ 2.977664020324591,
1015
+ 3.1650890310729984
1016
+ ]
1017
+ },
1018
+ "oracle_delta_nll": {
1019
+ "mean": 1.1114791103111255,
1020
+ "ci95": [
1021
+ 1.0353625793274794,
1022
+ 1.189223306976189
1023
+ ]
1024
+ },
1025
+ "oracle_response_error": {
1026
+ "mean": 0.0,
1027
+ "ci95": [
1028
+ 0.0,
1029
+ 0.0
1030
+ ]
1031
+ },
1032
+ "fit_intact_nll": {
1033
+ "mean": 1.989387447624577,
1034
+ "ci95": [
1035
+ 1.8592236560998066,
1036
+ 2.1119541075664086
1037
+ ]
1038
+ },
1039
+ "fit_donor_nll": {
1040
+ "mean": 3.105151471359101,
1041
+ "ci95": [
1042
+ 2.9953974567649815,
1043
+ 3.2069870618685328
1044
+ ]
1045
+ },
1046
+ "fit_delta_nll": {
1047
+ "mean": 1.1157640237345232,
1048
+ "ci95": [
1049
+ 1.0295745180294773,
1050
+ 1.2004888669786236
1051
+ ]
1052
+ },
1053
+ "fit_response_error": {
1054
+ "mean": 0.08758912483241174,
1055
+ "ci95": [
1056
+ 0.07303037041559458,
1057
+ 0.1043730674785598
1058
+ ]
1059
+ },
1060
+ "rw_intact_nll": {
1061
+ "mean": 2.2988755929499636,
1062
+ "ci95": [
1063
+ 2.1731721590488853,
1064
+ 2.42485811990197
1065
+ ]
1066
+ },
1067
+ "rw_donor_nll": {
1068
+ "mean": 3.001340023906025,
1069
+ "ci95": [
1070
+ 2.915388949156131,
1071
+ 3.0838527253928802
1072
+ ]
1073
+ },
1074
+ "rw_delta_nll": {
1075
+ "mean": 0.7024644309560609,
1076
+ "ci95": [
1077
+ 0.6391258291040296,
1078
+ 0.7659256076498202
1079
+ ]
1080
+ },
1081
+ "rw_response_error": {
1082
+ "mean": 0.46638960447755695,
1083
+ "ci95": [
1084
+ 0.43663822206139463,
1085
+ 0.4972329494845311
1086
+ ]
1087
+ },
1088
+ "gru_intact_nll": {
1089
+ "mean": 2.1560211276969277,
1090
+ "ci95": [
1091
+ 2.0233599017969452,
1092
+ 2.285698249896349
1093
+ ]
1094
+ },
1095
+ "gru_donor_nll": {
1096
+ "mean": 2.5309783087694346,
1097
+ "ci95": [
1098
+ 2.4003759190366365,
1099
+ 2.6544608216931116
1100
+ ]
1101
+ },
1102
+ "gru_delta_nll": {
1103
+ "mean": 0.3749571810725067,
1104
+ "ci95": [
1105
+ 0.34535636338354586,
1106
+ 0.40565186318236596
1107
+ ]
1108
+ },
1109
+ "gru_response_error": {
1110
+ "mean": 0.43442578582427266,
1111
+ "ci95": [
1112
+ 0.4170868498347895,
1113
+ 0.45154745158138637
1114
+ ]
1115
+ },
1116
+ "gru_full_nll": {
1117
+ "mean": 2.1560211276969277,
1118
+ "ci95": [
1119
+ 2.0233599017969452,
1120
+ 2.285698249896349
1121
+ ]
1122
+ },
1123
+ "gru_choice_nll": {
1124
+ "mean": 2.3215741339591087,
1125
+ "ci95": [
1126
+ 2.1851232503997142,
1127
+ 2.4508245885274915
1128
+ ]
1129
+ },
1130
+ "transformer_intact_nll": {
1131
+ "mean": 2.3759375327354832,
1132
+ "ci95": [
1133
+ 2.237223435366887,
1134
+ 2.507407954043828
1135
+ ]
1136
+ },
1137
+ "transformer_donor_nll": {
1138
+ "mean": 2.3867767376463767,
1139
+ "ci95": [
1140
+ 2.2494086097691603,
1141
+ 2.517031520655869
1142
+ ]
1143
+ },
1144
+ "transformer_delta_nll": {
1145
+ "mean": 0.010839204910892887,
1146
+ "ci95": [
1147
+ -0.0050139926210633476,
1148
+ 0.024369903512472868
1149
+ ]
1150
+ },
1151
+ "transformer_response_error": {
1152
+ "mean": 0.5608704789459782,
1153
+ "ci95": [
1154
+ 0.5463619754331537,
1155
+ 0.5754792414155258
1156
+ ]
1157
+ },
1158
+ "transformer_full_nll": {
1159
+ "mean": 2.3759375327354832,
1160
+ "ci95": [
1161
+ 2.237223435366887,
1162
+ 2.507407954043828
1163
+ ]
1164
+ },
1165
+ "transformer_choice_nll": {
1166
+ "mean": 2.3362569108904085,
1167
+ "ci95": [
1168
+ 2.202116897656126,
1169
+ 2.4649861695215285
1170
+ ]
1171
+ },
1172
+ "oracle_nll": {
1173
+ "mean": 1.965550659603369,
1174
+ "ci95": [
1175
+ 1.8382024060606479,
1176
+ 2.0891000379175035
1177
+ ]
1178
+ },
1179
+ "llama_full_nll": {
1180
+ "mean": 2.0700606486648856,
1181
+ "ci95": [
1182
+ 1.9352210499508247,
1183
+ 2.200384264629768
1184
+ ]
1185
+ },
1186
+ "llama_choice_nll": {
1187
+ "mean": 2.2991542415836577,
1188
+ "ci95": [
1189
+ 2.1670168032072885,
1190
+ 2.4258120437667134
1191
+ ]
1192
+ },
1193
+ "llama_reward_predictive_gain": {
1194
+ "mean": 0.22909359291877163,
1195
+ "ci95": [
1196
+ 0.20556610874214784,
1197
+ 0.2515510994488256
1198
+ ]
1199
+ }
1200
+ }
1201
+ },
1202
+ "1.0": {
1203
+ "all": {
1204
+ "oracle_intact_nll": {
1205
+ "mean": 1.8409537880893716,
1206
+ "ci95": [
1207
+ 1.69340417839034,
1208
+ 1.9846836798027525
1209
+ ]
1210
+ },
1211
+ "oracle_donor_nll": {
1212
+ "mean": 3.344660315048634,
1213
+ "ci95": [
1214
+ 3.24565980727262,
1215
+ 3.438878124534093
1216
+ ]
1217
+ },
1218
+ "oracle_delta_nll": {
1219
+ "mean": 1.503706526959262,
1220
+ "ci95": [
1221
+ 1.3943931517691155,
1222
+ 1.6241764870671251
1223
+ ]
1224
+ },
1225
+ "oracle_response_error": {
1226
+ "mean": 0.0,
1227
+ "ci95": [
1228
+ 0.0,
1229
+ 0.0
1230
+ ]
1231
+ },
1232
+ "fit_intact_nll": {
1233
+ "mean": 1.85205902407303,
1234
+ "ci95": [
1235
+ 1.7073344746824741,
1236
+ 1.9946343361928758
1237
+ ]
1238
+ },
1239
+ "fit_donor_nll": {
1240
+ "mean": 3.2959482457335247,
1241
+ "ci95": [
1242
+ 3.1894709207424197,
1243
+ 3.399085471724626
1244
+ ]
1245
+ },
1246
+ "fit_delta_nll": {
1247
+ "mean": 1.4438892216604955,
1248
+ "ci95": [
1249
+ 1.3266872473142837,
1250
+ 1.5718226756755076
1251
+ ]
1252
+ },
1253
+ "fit_response_error": {
1254
+ "mean": 0.08293078847487799,
1255
+ "ci95": [
1256
+ 0.07136590429310782,
1257
+ 0.09527586474564628
1258
+ ]
1259
+ },
1260
+ "rw_intact_nll": {
1261
+ "mean": 2.182240777872362,
1262
+ "ci95": [
1263
+ 2.037978379296234,
1264
+ 2.3253016925257044
1265
+ ]
1266
+ },
1267
+ "rw_donor_nll": {
1268
+ "mean": 2.9628095374735746,
1269
+ "ci95": [
1270
+ 2.854472799920614,
1271
+ 3.0668661621173405
1272
+ ]
1273
+ },
1274
+ "rw_delta_nll": {
1275
+ "mean": 0.7805687596012121,
1276
+ "ci95": [
1277
+ 0.7223996285194172,
1278
+ 0.8395591263590166
1279
+ ]
1280
+ },
1281
+ "rw_response_error": {
1282
+ "mean": 0.48694770469193427,
1283
+ "ci95": [
1284
+ 0.4586505485735272,
1285
+ 0.5145272442178125
1286
+ ]
1287
+ },
1288
+ "gru_intact_nll": {
1289
+ "mean": 2.0615749176865354,
1290
+ "ci95": [
1291
+ 1.9050403284820128,
1292
+ 2.2182691723729158
1293
+ ]
1294
+ },
1295
+ "gru_donor_nll": {
1296
+ "mean": 2.490458816793799,
1297
+ "ci95": [
1298
+ 2.336341070814708,
1299
+ 2.6367704886574392
1300
+ ]
1301
+ },
1302
+ "gru_delta_nll": {
1303
+ "mean": 0.42888389910726343,
1304
+ "ci95": [
1305
+ 0.38640727494611526,
1306
+ 0.4694190823899969
1307
+ ]
1308
+ },
1309
+ "gru_response_error": {
1310
+ "mean": 0.47156827265382206,
1311
+ "ci95": [
1312
+ 0.45450496680639135,
1313
+ 0.4882113786669313
1314
+ ]
1315
+ },
1316
+ "gru_full_nll": {
1317
+ "mean": 2.0615749176865354,
1318
+ "ci95": [
1319
+ 1.9050403284820128,
1320
+ 2.2182691723729158
1321
+ ]
1322
+ },
1323
+ "gru_choice_nll": {
1324
+ "mean": 2.2329980957421025,
1325
+ "ci95": [
1326
+ 2.076134840078994,
1327
+ 2.385221468320119
1328
+ ]
1329
+ },
1330
+ "transformer_intact_nll": {
1331
+ "mean": 2.283559003037639,
1332
+ "ci95": [
1333
+ 2.122726693768307,
1334
+ 2.4448122117462296
1335
+ ]
1336
+ },
1337
+ "transformer_donor_nll": {
1338
+ "mean": 2.298194320301616,
1339
+ "ci95": [
1340
+ 2.139506217368428,
1341
+ 2.453585895814056
1342
+ ]
1343
+ },
1344
+ "transformer_delta_nll": {
1345
+ "mean": 0.014635317263977026,
1346
+ "ci95": [
1347
+ 0.0031516652533479417,
1348
+ 0.025456896991478142
1349
+ ]
1350
+ },
1351
+ "transformer_response_error": {
1352
+ "mean": 0.6047550646863844,
1353
+ "ci95": [
1354
+ 0.5903452006943551,
1355
+ 0.6197256267873892
1356
+ ]
1357
+ },
1358
+ "transformer_full_nll": {
1359
+ "mean": 2.283559003037639,
1360
+ "ci95": [
1361
+ 2.122726693768307,
1362
+ 2.4448122117462296
1363
+ ]
1364
+ },
1365
+ "transformer_choice_nll": {
1366
+ "mean": 2.262184366134564,
1367
+ "ci95": [
1368
+ 2.0978098308708124,
1369
+ 2.424228047042848
1370
+ ]
1371
+ },
1372
+ "oracle_nll": {
1373
+ "mean": 1.8409537880893716,
1374
+ "ci95": [
1375
+ 1.69340417839034,
1376
+ 1.9846836798027525
1377
+ ]
1378
+ },
1379
+ "llama_full_nll": {
1380
+ "mean": 1.9458607498465463,
1381
+ "ci95": [
1382
+ 1.7952800593747194,
1383
+ 2.0909579866150034
1384
+ ]
1385
+ },
1386
+ "llama_choice_nll": {
1387
+ "mean": 2.19591801244102,
1388
+ "ci95": [
1389
+ 2.0370182757496655,
1390
+ 2.3519438757927493
1391
+ ]
1392
+ },
1393
+ "llama_reward_predictive_gain": {
1394
+ "mean": 0.2500572625944739,
1395
+ "ci95": [
1396
+ 0.2258001188455806,
1397
+ 0.2753454848194946
1398
+ ]
1399
+ }
1400
+ }
1401
+ }
1402
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/B_summary.json ADDED
@@ -0,0 +1,2515 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "0.0": {
3
+ "oracle_intact_nll": {
4
+ "mean": 2.256899081853402,
5
+ "ci95": [
6
+ 2.1539276579886697,
7
+ 2.3536397088713543
8
+ ]
9
+ },
10
+ "oracle_donor_nll": {
11
+ "mean": 2.256899081853402,
12
+ "ci95": [
13
+ 2.1539276579886697,
14
+ 2.3536397088713543
15
+ ]
16
+ },
17
+ "oracle_delta_nll": {
18
+ "mean": 3.552713678800501e-17,
19
+ "ci95": [
20
+ -1.6431300764452317e-16,
21
+ 2.353672812205332e-16
22
+ ]
23
+ },
24
+ "oracle_response_error": {
25
+ "mean": 0.0,
26
+ "ci95": [
27
+ 0.0,
28
+ 0.0
29
+ ]
30
+ },
31
+ "fit_intact_nll": {
32
+ "mean": 2.2665208854580676,
33
+ "ci95": [
34
+ 2.165162571793024,
35
+ 2.3635493247560464
36
+ ]
37
+ },
38
+ "fit_donor_nll": {
39
+ "mean": 2.266188960502596,
40
+ "ci95": [
41
+ 2.1646592445093957,
42
+ 2.361894910786617
43
+ ]
44
+ },
45
+ "fit_delta_nll": {
46
+ "mean": -0.00033192495547172294,
47
+ "ci95": [
48
+ -0.0029861019709051927,
49
+ 0.0025254340726127834
50
+ ]
51
+ },
52
+ "fit_response_error": {
53
+ "mean": 0.010783533619515567,
54
+ "ci95": [
55
+ 0.006974173524976919,
56
+ 0.015135500302760889
57
+ ]
58
+ },
59
+ "rw_intact_nll": {
60
+ "mean": 3.2175559090104016,
61
+ "ci95": [
62
+ 3.213504313467025,
63
+ 3.22121403483305
64
+ ]
65
+ },
66
+ "rw_donor_nll": {
67
+ "mean": 3.2189942581169397,
68
+ "ci95": [
69
+ 3.2181489329875665,
70
+ 3.2197627748496
71
+ ]
72
+ },
73
+ "rw_delta_nll": {
74
+ "mean": 0.0014383491065387587,
75
+ "ci95": [
76
+ -0.0019054543112809552,
77
+ 0.005397353097600356
78
+ ]
79
+ },
80
+ "rw_response_error": {
81
+ "mean": 0.0065546631610965255,
82
+ "ci95": [
83
+ 0.006253397299093739,
84
+ 0.006874478198303794
85
+ ]
86
+ },
87
+ "gru_intact_nll": {
88
+ "mean": 2.3003609185091234,
89
+ "ci95": [
90
+ 2.198413620263349,
91
+ 2.400978339217044
92
+ ]
93
+ },
94
+ "gru_donor_nll": {
95
+ "mean": 2.299973465268989,
96
+ "ci95": [
97
+ 2.1976383500693397,
98
+ 2.400265694486215
99
+ ]
100
+ },
101
+ "gru_delta_nll": {
102
+ "mean": -0.0003874532401340769,
103
+ "ci95": [
104
+ -0.0014255719779839957,
105
+ 0.0007252131943553924
106
+ ]
107
+ },
108
+ "gru_response_error": {
109
+ "mean": 0.018504102779924144,
110
+ "ci95": [
111
+ 0.018051273358047692,
112
+ 0.018952429187911405
113
+ ]
114
+ },
115
+ "gru_full_nll": {
116
+ "mean": 2.3003609185091234,
117
+ "ci95": [
118
+ 2.198413620263349,
119
+ 2.400978339217044
120
+ ]
121
+ },
122
+ "gru_choice_nll": {
123
+ "mean": 2.2997417918175467,
124
+ "ci95": [
125
+ 2.198184169213947,
126
+ 2.396640228449533
127
+ ]
128
+ },
129
+ "transformer_intact_nll": {
130
+ "mean": 2.360867912326785,
131
+ "ci95": [
132
+ 2.259512517521976,
133
+ 2.4570797715249877
134
+ ]
135
+ },
136
+ "transformer_donor_nll": {
137
+ "mean": 2.3535000802925157,
138
+ "ci95": [
139
+ 2.252967515338882,
140
+ 2.449982839537024
141
+ ]
142
+ },
143
+ "transformer_delta_nll": {
144
+ "mean": -0.007367832034269306,
145
+ "ci95": [
146
+ -0.02186180228286032,
147
+ 0.00516436291414688
148
+ ]
149
+ },
150
+ "transformer_response_error": {
151
+ "mean": 0.1170145991357271,
152
+ "ci95": [
153
+ 0.11260599481564193,
154
+ 0.1216407268660003
155
+ ]
156
+ },
157
+ "transformer_full_nll": {
158
+ "mean": 2.360867912326785,
159
+ "ci95": [
160
+ 2.259512517521976,
161
+ 2.4570797715249877
162
+ ]
163
+ },
164
+ "transformer_choice_nll": {
165
+ "mean": 2.3071931246818678,
166
+ "ci95": [
167
+ 2.201677337007935,
168
+ 2.4067951834649857
169
+ ]
170
+ },
171
+ "oracle_nll": {
172
+ "mean": 2.256899081853402,
173
+ "ci95": [
174
+ 2.1539276579886697,
175
+ 2.3536397088713543
176
+ ]
177
+ },
178
+ "llama_intact_nll": {
179
+ "mean": 2.2982575594719052,
180
+ "ci95": [
181
+ 2.195976598373851,
182
+ 2.3952368262662347
183
+ ]
184
+ },
185
+ "llama_donor_nll": {
186
+ "mean": 2.3000127776288415,
187
+ "ci95": [
188
+ 2.1988087420927815,
189
+ 2.3957475699919732
190
+ ]
191
+ },
192
+ "llama_delta_nll": {
193
+ "mean": 0.0017552181569360625,
194
+ "ci95": [
195
+ -0.0010138906491887032,
196
+ 0.004542731493251509
197
+ ]
198
+ },
199
+ "llama_response_error": {
200
+ "mean": 0.034685900451633486,
201
+ "ci95": [
202
+ 0.03421444779553409,
203
+ 0.03513343994067801
204
+ ]
205
+ },
206
+ "llama_full_nll": {
207
+ "mean": 2.2982575594719052,
208
+ "ci95": [
209
+ 2.195976598373851,
210
+ 2.3952368262662347
211
+ ]
212
+ },
213
+ "llama_choice_nll": {
214
+ "mean": 2.3024848628881505,
215
+ "ci95": [
216
+ 2.2006664714056825,
217
+ 2.398696368876586
218
+ ]
219
+ },
220
+ "llama_reward_predictive_gain": {
221
+ "mean": 0.004227303416245162,
222
+ "ci95": [
223
+ -0.001178904395716497,
224
+ 0.009791267599562867
225
+ ]
226
+ },
227
+ "llama_full_token_nll": {
228
+ "mean": 2.298715114593506,
229
+ "ci95": [
230
+ 2.1964227616786958,
231
+ 2.3956938862800596
232
+ ]
233
+ },
234
+ "llama_full_valid_mass": {
235
+ "mean": 0.9995424151420593,
236
+ "ci95": [
237
+ 0.9995232224464417,
238
+ 0.9995617896318436
239
+ ]
240
+ },
241
+ "llama_choice_token_nll": {
242
+ "mean": 2.302917718887329,
243
+ "ci95": [
244
+ 2.201080638170242,
245
+ 2.399127560853958
246
+ ]
247
+ },
248
+ "llama_choice_valid_mass": {
249
+ "mean": 0.9995672106742859,
250
+ "ci95": [
251
+ 0.9995524883270264,
252
+ 0.9995830714702606
253
+ ]
254
+ },
255
+ "llama_intact_nll_minus_oracle": {
256
+ "mean": 0.041358477618503135,
257
+ "ci95": [
258
+ 0.03095186344855763,
259
+ 0.05128248010646278
260
+ ]
261
+ },
262
+ "llama_response_error_minus_oracle": {
263
+ "mean": 0.034685900451633486,
264
+ "ci95": [
265
+ 0.03421444779553409,
266
+ 0.03513343994067801
267
+ ]
268
+ },
269
+ "llama_delta_nll_minus_oracle": {
270
+ "mean": 0.001755218156936027,
271
+ "ci95": [
272
+ -0.0010138906491885939,
273
+ 0.00454273149325147
274
+ ]
275
+ },
276
+ "llama_intact_nll_minus_fit": {
277
+ "mean": 0.03173667401383725,
278
+ "ci95": [
279
+ 0.0193147562294747,
280
+ 0.04378725646787503
281
+ ]
282
+ },
283
+ "llama_response_error_minus_fit": {
284
+ "mean": 0.023902366832117918,
285
+ "ci95": [
286
+ 0.01951889712972559,
287
+ 0.02783099522185491
288
+ ]
289
+ },
290
+ "llama_delta_nll_minus_fit": {
291
+ "mean": 0.0020871431124077854,
292
+ "ci95": [
293
+ -0.0019720818277989076,
294
+ 0.0060312025073745315
295
+ ]
296
+ },
297
+ "llama_intact_nll_minus_rw": {
298
+ "mean": -0.9192983495384958,
299
+ "ci95": [
300
+ -1.0199171741658368,
301
+ -0.8239544387468823
302
+ ]
303
+ },
304
+ "llama_response_error_minus_rw": {
305
+ "mean": 0.028131237290536962,
306
+ "ci95": [
307
+ 0.0275573780809003,
308
+ 0.028686345016509837
309
+ ]
310
+ },
311
+ "llama_delta_nll_minus_rw": {
312
+ "mean": 0.00031686905039730374,
313
+ "ci95": [
314
+ -0.004378245981014182,
315
+ 0.004808541140668664
316
+ ]
317
+ },
318
+ "llama_intact_nll_minus_gru": {
319
+ "mean": -0.0021033590372179487,
320
+ "ci95": [
321
+ -0.01310353827380741,
322
+ 0.008212192878320687
323
+ ]
324
+ },
325
+ "llama_response_error_minus_gru": {
326
+ "mean": 0.016181797671709342,
327
+ "ci95": [
328
+ 0.015576578416638565,
329
+ 0.0167434160641622
330
+ ]
331
+ },
332
+ "llama_delta_nll_minus_gru": {
333
+ "mean": 0.002142671397070139,
334
+ "ci95": [
335
+ -0.000864028760935866,
336
+ 0.005082591075951415
337
+ ]
338
+ },
339
+ "llama_intact_nll_minus_transformer": {
340
+ "mean": -0.06261035285487998,
341
+ "ci95": [
342
+ -0.08177649670689986,
343
+ -0.043755821991460886
344
+ ]
345
+ },
346
+ "llama_response_error_minus_transformer": {
347
+ "mean": -0.08232869868409358,
348
+ "ci95": [
349
+ -0.08696662343721409,
350
+ -0.07810447119034557
351
+ ]
352
+ },
353
+ "llama_delta_nll_minus_transformer": {
354
+ "mean": 0.009123050191205368,
355
+ "ci95": [
356
+ -0.0038930127112542117,
357
+ 0.023560970689973773
358
+ ]
359
+ }
360
+ },
361
+ "0.1": {
362
+ "oracle_intact_nll": {
363
+ "mean": 2.358300456778261,
364
+ "ci95": [
365
+ 2.2629998556727573,
366
+ 2.4481495877709705
367
+ ]
368
+ },
369
+ "oracle_donor_nll": {
370
+ "mean": 2.4054874510008455,
371
+ "ci95": [
372
+ 2.3087123638900544,
373
+ 2.5022874313472685
374
+ ]
375
+ },
376
+ "oracle_delta_nll": {
377
+ "mean": 0.04718699422258422,
378
+ "ci95": [
379
+ 0.03436657494213905,
380
+ 0.05970594641691559
381
+ ]
382
+ },
383
+ "oracle_response_error": {
384
+ "mean": 0.0,
385
+ "ci95": [
386
+ 0.0,
387
+ 0.0
388
+ ]
389
+ },
390
+ "fit_intact_nll": {
391
+ "mean": 2.3795351204281063,
392
+ "ci95": [
393
+ 2.2851009583330377,
394
+ 2.4709300500665647
395
+ ]
396
+ },
397
+ "fit_donor_nll": {
398
+ "mean": 2.4210424329847204,
399
+ "ci95": [
400
+ 2.3238395816470176,
401
+ 2.517420854643946
402
+ ]
403
+ },
404
+ "fit_delta_nll": {
405
+ "mean": 0.04150731255661429,
406
+ "ci95": [
407
+ 0.026934889627137568,
408
+ 0.0564772909759607
409
+ ]
410
+ },
411
+ "fit_response_error": {
412
+ "mean": 0.052102911681065664,
413
+ "ci95": [
414
+ 0.0450136975626096,
415
+ 0.06001172458386209
416
+ ]
417
+ },
418
+ "rw_intact_nll": {
419
+ "mean": 3.189361391632845,
420
+ "ci95": [
421
+ 3.153220666502596,
422
+ 3.2250492036221883
423
+ ]
424
+ },
425
+ "rw_donor_nll": {
426
+ "mean": 3.2160267405427345,
427
+ "ci95": [
428
+ 3.208121107134707,
429
+ 3.2233383354985814
430
+ ]
431
+ },
432
+ "rw_delta_nll": {
433
+ "mean": 0.02666534890988978,
434
+ "ci95": [
435
+ -0.005695483999127539,
436
+ 0.0591333235912339
437
+ ]
438
+ },
439
+ "rw_response_error": {
440
+ "mean": 0.08648257221904393,
441
+ "ci95": [
442
+ 0.08208808266072022,
443
+ 0.09081407695981439
444
+ ]
445
+ },
446
+ "gru_intact_nll": {
447
+ "mean": 2.4251186469737713,
448
+ "ci95": [
449
+ 2.3284747965202945,
450
+ 2.524579672921996
451
+ ]
452
+ },
453
+ "gru_donor_nll": {
454
+ "mean": 2.426071139402867,
455
+ "ci95": [
456
+ 2.329496817277385,
457
+ 2.5249042540418163
458
+ ]
459
+ },
460
+ "gru_delta_nll": {
461
+ "mean": 0.0009524924290957977,
462
+ "ci95": [
463
+ -0.0007303317010866242,
464
+ 0.002569258607637665
465
+ ]
466
+ },
467
+ "gru_response_error": {
468
+ "mean": 0.07014626407047886,
469
+ "ci95": [
470
+ 0.0678662056582069,
471
+ 0.07250281904867092
472
+ ]
473
+ },
474
+ "gru_full_nll": {
475
+ "mean": 2.4251186469737713,
476
+ "ci95": [
477
+ 2.3284747965202945,
478
+ 2.524579672921996
479
+ ]
480
+ },
481
+ "gru_choice_nll": {
482
+ "mean": 2.423530669573678,
483
+ "ci95": [
484
+ 2.3258626021057913,
485
+ 2.5209163679798374
486
+ ]
487
+ },
488
+ "transformer_intact_nll": {
489
+ "mean": 2.5021050715922293,
490
+ "ci95": [
491
+ 2.4071664429399253,
492
+ 2.595780298187066
493
+ ]
494
+ },
495
+ "transformer_donor_nll": {
496
+ "mean": 2.4970589588915537,
497
+ "ci95": [
498
+ 2.4001181301544525,
499
+ 2.593694670124522
500
+ ]
501
+ },
502
+ "transformer_delta_nll": {
503
+ "mean": -0.0050461127006755715,
504
+ "ci95": [
505
+ -0.022723500215269178,
506
+ 0.009568835278212057
507
+ ]
508
+ },
509
+ "transformer_response_error": {
510
+ "mean": 0.15332741715503675,
511
+ "ci95": [
512
+ 0.14787185484306695,
513
+ 0.15920991345423632
514
+ ]
515
+ },
516
+ "transformer_full_nll": {
517
+ "mean": 2.5021050715922293,
518
+ "ci95": [
519
+ 2.4071664429399253,
520
+ 2.595780298187066
521
+ ]
522
+ },
523
+ "transformer_choice_nll": {
524
+ "mean": 2.4344931105129946,
525
+ "ci95": [
526
+ 2.3364925924441855,
527
+ 2.533731411874208
528
+ ]
529
+ },
530
+ "oracle_nll": {
531
+ "mean": 2.358300456778261,
532
+ "ci95": [
533
+ 2.2629998556727573,
534
+ 2.4481495877709705
535
+ ]
536
+ },
537
+ "llama_intact_nll": {
538
+ "mean": 2.4158071960705993,
539
+ "ci95": [
540
+ 2.3239026143343584,
541
+ 2.5053612041975044
542
+ ]
543
+ },
544
+ "llama_donor_nll": {
545
+ "mean": 2.434293540329535,
546
+ "ci95": [
547
+ 2.344002872475072,
548
+ 2.5261841661128814
549
+ ]
550
+ },
551
+ "llama_delta_nll": {
552
+ "mean": 0.018486344258935836,
553
+ "ci95": [
554
+ 0.009299995427665807,
555
+ 0.027368167381460463
556
+ ]
557
+ },
558
+ "llama_response_error": {
559
+ "mean": 0.07818114175191496,
560
+ "ci95": [
561
+ 0.07550256899838828,
562
+ 0.08089869862399571
563
+ ]
564
+ },
565
+ "llama_full_nll": {
566
+ "mean": 2.4158071960705993,
567
+ "ci95": [
568
+ 2.3239026143343584,
569
+ 2.5053612041975044
570
+ ]
571
+ },
572
+ "llama_choice_nll": {
573
+ "mean": 2.422510292332979,
574
+ "ci95": [
575
+ 2.32652689505674,
576
+ 2.5194355990993675
577
+ ]
578
+ },
579
+ "llama_reward_predictive_gain": {
580
+ "mean": 0.006703096262379908,
581
+ "ci95": [
582
+ -0.006473624100292907,
583
+ 0.02141005324122265
584
+ ]
585
+ },
586
+ "llama_full_token_nll": {
587
+ "mean": 2.4161577224731445,
588
+ "ci95": [
589
+ 2.324247473478317,
590
+ 2.5057081043720246
591
+ ]
592
+ },
593
+ "llama_full_valid_mass": {
594
+ "mean": 0.9996495842933655,
595
+ "ci95": [
596
+ 0.9996389120817184,
597
+ 0.9996605545282364
598
+ ]
599
+ },
600
+ "llama_choice_token_nll": {
601
+ "mean": 2.423147439956665,
602
+ "ci95": [
603
+ 2.3271656334400177,
604
+ 2.5200860798358917
605
+ ]
606
+ },
607
+ "llama_choice_valid_mass": {
608
+ "mean": 0.9993631839752197,
609
+ "ci95": [
610
+ 0.9993451803922653,
611
+ 0.999381422996521
612
+ ]
613
+ },
614
+ "llama_intact_nll_minus_oracle": {
615
+ "mean": 0.05750673929233845,
616
+ "ci95": [
617
+ 0.04292334428607893,
618
+ 0.07019658542915569
619
+ ]
620
+ },
621
+ "llama_response_error_minus_oracle": {
622
+ "mean": 0.07818114175191496,
623
+ "ci95": [
624
+ 0.07550256899838828,
625
+ 0.08089869862399571
626
+ ]
627
+ },
628
+ "llama_delta_nll_minus_oracle": {
629
+ "mean": -0.028700649963648385,
630
+ "ci95": [
631
+ -0.036707294279320936,
632
+ -0.01969705151890679
633
+ ]
634
+ },
635
+ "llama_intact_nll_minus_fit": {
636
+ "mean": 0.0362720756424935,
637
+ "ci95": [
638
+ 0.021664311878601725,
639
+ 0.0511443213202533
640
+ ]
641
+ },
642
+ "llama_response_error_minus_fit": {
643
+ "mean": 0.026078230070849293,
644
+ "ci95": [
645
+ 0.017834663097351152,
646
+ 0.03332474298947259
647
+ ]
648
+ },
649
+ "llama_delta_nll_minus_fit": {
650
+ "mean": -0.023020968297678454,
651
+ "ci95": [
652
+ -0.03400584707003782,
653
+ -0.011808941166238223
654
+ ]
655
+ },
656
+ "llama_intact_nll_minus_rw": {
657
+ "mean": -0.773554195562245,
658
+ "ci95": [
659
+ -0.8716405519212695,
660
+ -0.6773840328091378
661
+ ]
662
+ },
663
+ "llama_response_error_minus_rw": {
664
+ "mean": -0.00830143046712901,
665
+ "ci95": [
666
+ -0.011020810677948035,
667
+ -0.00574724807161236
668
+ ]
669
+ },
670
+ "llama_delta_nll_minus_rw": {
671
+ "mean": -0.008179004650953945,
672
+ "ci95": [
673
+ -0.03756042314387396,
674
+ 0.021674808687014126
675
+ ]
676
+ },
677
+ "llama_intact_nll_minus_gru": {
678
+ "mean": -0.009311450903171821,
679
+ "ci95": [
680
+ -0.024680799057298985,
681
+ 0.0045575068508944385
682
+ ]
683
+ },
684
+ "llama_response_error_minus_gru": {
685
+ "mean": 0.008034877681436084,
686
+ "ci95": [
687
+ 0.006347595306773485,
688
+ 0.009596415369328242
689
+ ]
690
+ },
691
+ "llama_delta_nll_minus_gru": {
692
+ "mean": 0.017533851829840037,
693
+ "ci95": [
694
+ 0.00826569936125985,
695
+ 0.026435680330504654
696
+ ]
697
+ },
698
+ "llama_intact_nll_minus_transformer": {
699
+ "mean": -0.08629787552162987,
700
+ "ci95": [
701
+ -0.10827617311922638,
702
+ -0.0653981838387274
703
+ ]
704
+ },
705
+ "llama_response_error_minus_transformer": {
706
+ "mean": -0.07514627540312178,
707
+ "ci95": [
708
+ -0.08021481322192267,
709
+ -0.07020987904179572
710
+ ]
711
+ },
712
+ "llama_delta_nll_minus_transformer": {
713
+ "mean": 0.02353245695961141,
714
+ "ci95": [
715
+ 0.008080263943979756,
716
+ 0.039239434247823335
717
+ ]
718
+ }
719
+ },
720
+ "0.3": {
721
+ "oracle_intact_nll": {
722
+ "mean": 2.4031890757538656,
723
+ "ci95": [
724
+ 2.3080410147812622,
725
+ 2.493060873165434
726
+ ]
727
+ },
728
+ "oracle_donor_nll": {
729
+ "mean": 2.618646521081867,
730
+ "ci95": [
731
+ 2.527502915350205,
732
+ 2.700554180347221
733
+ ]
734
+ },
735
+ "oracle_delta_nll": {
736
+ "mean": 0.21545744532800146,
737
+ "ci95": [
738
+ 0.18986788030576535,
739
+ 0.24149767842861766
740
+ ]
741
+ },
742
+ "oracle_response_error": {
743
+ "mean": 0.0,
744
+ "ci95": [
745
+ 0.0,
746
+ 0.0
747
+ ]
748
+ },
749
+ "fit_intact_nll": {
750
+ "mean": 2.41717893155207,
751
+ "ci95": [
752
+ 2.322061230563267,
753
+ 2.505455425612881
754
+ ]
755
+ },
756
+ "fit_donor_nll": {
757
+ "mean": 2.6242176974420066,
758
+ "ci95": [
759
+ 2.5317101501606807,
760
+ 2.7095291421819456
761
+ ]
762
+ },
763
+ "fit_delta_nll": {
764
+ "mean": 0.20703876588993686,
765
+ "ci95": [
766
+ 0.17986128071058735,
767
+ 0.23594040419722156
768
+ ]
769
+ },
770
+ "fit_response_error": {
771
+ "mean": 0.0691367955015857,
772
+ "ci95": [
773
+ 0.059267532699285544,
774
+ 0.07887737741055921
775
+ ]
776
+ },
777
+ "rw_intact_nll": {
778
+ "mean": 3.0243127705433683,
779
+ "ci95": [
780
+ 2.9625603901177726,
781
+ 3.083535840993125
782
+ ]
783
+ },
784
+ "rw_donor_nll": {
785
+ "mean": 3.203483069337998,
786
+ "ci95": [
787
+ 3.1824372245788966,
788
+ 3.2239390213336447
789
+ ]
790
+ },
791
+ "rw_delta_nll": {
792
+ "mean": 0.17917029879463012,
793
+ "ci95": [
794
+ 0.13433822314957689,
795
+ 0.22693241171930228
796
+ ]
797
+ },
798
+ "rw_response_error": {
799
+ "mean": 0.21432052268495294,
800
+ "ci95": [
801
+ 0.20350847432596209,
802
+ 0.22558303850567366
803
+ ]
804
+ },
805
+ "gru_intact_nll": {
806
+ "mean": 2.524001883635677,
807
+ "ci95": [
808
+ 2.4321140389831455,
809
+ 2.6099826043107632
810
+ ]
811
+ },
812
+ "gru_donor_nll": {
813
+ "mean": 2.571780270978965,
814
+ "ci95": [
815
+ 2.4773468057793844,
816
+ 2.6560668792585798
817
+ ]
818
+ },
819
+ "gru_delta_nll": {
820
+ "mean": 0.0477783873432883,
821
+ "ci95": [
822
+ 0.03760635238477913,
823
+ 0.05711140657155545
824
+ ]
825
+ },
826
+ "gru_response_error": {
827
+ "mean": 0.20187495556609986,
828
+ "ci95": [
829
+ 0.19484867770261477,
830
+ 0.2084915060931258
831
+ ]
832
+ },
833
+ "gru_full_nll": {
834
+ "mean": 2.524001883635677,
835
+ "ci95": [
836
+ 2.4321140389831455,
837
+ 2.6099826043107632
838
+ ]
839
+ },
840
+ "gru_choice_nll": {
841
+ "mean": 2.550332199199586,
842
+ "ci95": [
843
+ 2.458054609892982,
844
+ 2.632661163300756
845
+ ]
846
+ },
847
+ "transformer_intact_nll": {
848
+ "mean": 2.6372881952925993,
849
+ "ci95": [
850
+ 2.546320653589055,
851
+ 2.722487010986659
852
+ ]
853
+ },
854
+ "transformer_donor_nll": {
855
+ "mean": 2.6450593082988045,
856
+ "ci95": [
857
+ 2.5556576753880242,
858
+ 2.7302018826026617
859
+ ]
860
+ },
861
+ "transformer_delta_nll": {
862
+ "mean": 0.007771113006204906,
863
+ "ci95": [
864
+ -0.006287134574327571,
865
+ 0.020174173096946673
866
+ ]
867
+ },
868
+ "transformer_response_error": {
869
+ "mean": 0.2640845588027949,
870
+ "ci95": [
871
+ 0.2570148489025373,
872
+ 0.27070137321903603
873
+ ]
874
+ },
875
+ "transformer_full_nll": {
876
+ "mean": 2.6372881952925993,
877
+ "ci95": [
878
+ 2.546320653589055,
879
+ 2.722487010986659
880
+ ]
881
+ },
882
+ "transformer_choice_nll": {
883
+ "mean": 2.554989409750647,
884
+ "ci95": [
885
+ 2.461253575379825,
886
+ 2.6405940931788856
887
+ ]
888
+ },
889
+ "oracle_nll": {
890
+ "mean": 2.4031890757538656,
891
+ "ci95": [
892
+ 2.3080410147812622,
893
+ 2.493060873165434
894
+ ]
895
+ },
896
+ "llama_intact_nll": {
897
+ "mean": 2.466419687330082,
898
+ "ci95": [
899
+ 2.372019483451541,
900
+ 2.554812335193939
901
+ ]
902
+ },
903
+ "llama_donor_nll": {
904
+ "mean": 2.5944409420649324,
905
+ "ci95": [
906
+ 2.506159909308146,
907
+ 2.6744012508148973
908
+ ]
909
+ },
910
+ "llama_delta_nll": {
911
+ "mean": 0.12802125473485015,
912
+ "ci95": [
913
+ 0.10804493840766487,
914
+ 0.14787100178474072
915
+ ]
916
+ },
917
+ "llama_response_error": {
918
+ "mean": 0.16380456286752706,
919
+ "ci95": [
920
+ 0.15512916684352704,
921
+ 0.17229401935397826
922
+ ]
923
+ },
924
+ "llama_full_nll": {
925
+ "mean": 2.466419687330082,
926
+ "ci95": [
927
+ 2.372019483451541,
928
+ 2.554812335193939
929
+ ]
930
+ },
931
+ "llama_choice_nll": {
932
+ "mean": 2.534494785695525,
933
+ "ci95": [
934
+ 2.4440944376267026,
935
+ 2.620180488364762
936
+ ]
937
+ },
938
+ "llama_reward_predictive_gain": {
939
+ "mean": 0.06807509836544341,
940
+ "ci95": [
941
+ 0.04992199862100402,
942
+ 0.08532290762912904
943
+ ]
944
+ },
945
+ "llama_full_token_nll": {
946
+ "mean": 2.4668521881103516,
947
+ "ci95": [
948
+ 2.372445547580719,
949
+ 2.5552481949329375
950
+ ]
951
+ },
952
+ "llama_full_valid_mass": {
953
+ "mean": 0.9995673894882202,
954
+ "ci95": [
955
+ 0.9995462849736214,
956
+ 0.9995886132121086
957
+ ]
958
+ },
959
+ "llama_choice_token_nll": {
960
+ "mean": 2.534930944442749,
961
+ "ci95": [
962
+ 2.444511431455612,
963
+ 2.6206152617931364
964
+ ]
965
+ },
966
+ "llama_choice_valid_mass": {
967
+ "mean": 0.9995638132095337,
968
+ "ci95": [
969
+ 0.9995467618107796,
970
+ 0.9995796710252762
971
+ ]
972
+ },
973
+ "llama_intact_nll_minus_oracle": {
974
+ "mean": 0.0632306115762161,
975
+ "ci95": [
976
+ 0.04828396671227799,
977
+ 0.07702036290514606
978
+ ]
979
+ },
980
+ "llama_response_error_minus_oracle": {
981
+ "mean": 0.16380456286752706,
982
+ "ci95": [
983
+ 0.15512916684352704,
984
+ 0.17229401935397826
985
+ ]
986
+ },
987
+ "llama_delta_nll_minus_oracle": {
988
+ "mean": -0.08743619059315129,
989
+ "ci95": [
990
+ -0.10591114264761337,
991
+ -0.07185737349326406
992
+ ]
993
+ },
994
+ "llama_intact_nll_minus_fit": {
995
+ "mean": 0.04924075577801213,
996
+ "ci95": [
997
+ 0.03340692313141176,
998
+ 0.06390979128814406
999
+ ]
1000
+ },
1001
+ "llama_response_error_minus_fit": {
1002
+ "mean": 0.0946677673659414,
1003
+ "ci95": [
1004
+ 0.08144499942123704,
1005
+ 0.10836554314672622
1006
+ ]
1007
+ },
1008
+ "llama_delta_nll_minus_fit": {
1009
+ "mean": -0.07901751115508669,
1010
+ "ci95": [
1011
+ -0.0999981205320528,
1012
+ -0.06107686516716186
1013
+ ]
1014
+ },
1015
+ "llama_intact_nll_minus_rw": {
1016
+ "mean": -0.557893083213286,
1017
+ "ci95": [
1018
+ -0.6240226952174234,
1019
+ -0.4944246296601903
1020
+ ]
1021
+ },
1022
+ "llama_response_error_minus_rw": {
1023
+ "mean": -0.05051595981742585,
1024
+ "ci95": [
1025
+ -0.05684643747492986,
1026
+ -0.044387654540953
1027
+ ]
1028
+ },
1029
+ "llama_delta_nll_minus_rw": {
1030
+ "mean": -0.05114904405977996,
1031
+ "ci95": [
1032
+ -0.08998265338327036,
1033
+ -0.014652606919730107
1034
+ ]
1035
+ },
1036
+ "llama_intact_nll_minus_gru": {
1037
+ "mean": -0.05758219630559473,
1038
+ "ci95": [
1039
+ -0.07592445144321852,
1040
+ -0.039859448219858325
1041
+ ]
1042
+ },
1043
+ "llama_response_error_minus_gru": {
1044
+ "mean": -0.03807039269857281,
1045
+ "ci95": [
1046
+ -0.043414670550790194,
1047
+ -0.032823984042472214
1048
+ ]
1049
+ },
1050
+ "llama_delta_nll_minus_gru": {
1051
+ "mean": 0.08024286739156185,
1052
+ "ci95": [
1053
+ 0.06442035536756117,
1054
+ 0.09536200118751151
1055
+ ]
1056
+ },
1057
+ "llama_intact_nll_minus_transformer": {
1058
+ "mean": -0.1708685079625176,
1059
+ "ci95": [
1060
+ -0.19632667009352667,
1061
+ -0.1459546403778737
1062
+ ]
1063
+ },
1064
+ "llama_response_error_minus_transformer": {
1065
+ "mean": -0.1002799959352678,
1066
+ "ci95": [
1067
+ -0.10568980079929212,
1068
+ -0.09490147661143745
1069
+ ]
1070
+ },
1071
+ "llama_delta_nll_minus_transformer": {
1072
+ "mean": 0.12025014172864525,
1073
+ "ci95": [
1074
+ 0.098721993073337,
1075
+ 0.1425654247023786
1076
+ ]
1077
+ }
1078
+ },
1079
+ "0.5": {
1080
+ "oracle_intact_nll": {
1081
+ "mean": 2.319283828723959,
1082
+ "ci95": [
1083
+ 2.215970742786679,
1084
+ 2.4251377502112526
1085
+ ]
1086
+ },
1087
+ "oracle_donor_nll": {
1088
+ "mean": 2.7559770217009416,
1089
+ "ci95": [
1090
+ 2.6745745656011968,
1091
+ 2.836698169145955
1092
+ ]
1093
+ },
1094
+ "oracle_delta_nll": {
1095
+ "mean": 0.43669319297698217,
1096
+ "ci95": [
1097
+ 0.3936526854652242,
1098
+ 0.4783219587769138
1099
+ ]
1100
+ },
1101
+ "oracle_response_error": {
1102
+ "mean": 0.0,
1103
+ "ci95": [
1104
+ 0.0,
1105
+ 0.0
1106
+ ]
1107
+ },
1108
+ "fit_intact_nll": {
1109
+ "mean": 2.342043599243521,
1110
+ "ci95": [
1111
+ 2.2380130079444327,
1112
+ 2.4455075486523197
1113
+ ]
1114
+ },
1115
+ "fit_donor_nll": {
1116
+ "mean": 2.7809330530121543,
1117
+ "ci95": [
1118
+ 2.697637756052261,
1119
+ 2.8650316810601226
1120
+ ]
1121
+ },
1122
+ "fit_delta_nll": {
1123
+ "mean": 0.43888945376863364,
1124
+ "ci95": [
1125
+ 0.3955619119800974,
1126
+ 0.4823279484351039
1127
+ ]
1128
+ },
1129
+ "fit_response_error": {
1130
+ "mean": 0.07216533259262915,
1131
+ "ci95": [
1132
+ 0.061967845980080045,
1133
+ 0.08186041789541766
1134
+ ]
1135
+ },
1136
+ "rw_intact_nll": {
1137
+ "mean": 2.805631640622673,
1138
+ "ci95": [
1139
+ 2.717358818971326,
1140
+ 2.8872822320788982
1141
+ ]
1142
+ },
1143
+ "rw_donor_nll": {
1144
+ "mean": 3.1864433123704146,
1145
+ "ci95": [
1146
+ 3.148402000229726,
1147
+ 3.2230480775807617
1148
+ ]
1149
+ },
1150
+ "rw_delta_nll": {
1151
+ "mean": 0.3808116717477414,
1152
+ "ci95": [
1153
+ 0.3250388310644342,
1154
+ 0.4398623790768384
1155
+ ]
1156
+ },
1157
+ "rw_response_error": {
1158
+ "mean": 0.31536258468027717,
1159
+ "ci95": [
1160
+ 0.2983742329804894,
1161
+ 0.33353337030248736
1162
+ ]
1163
+ },
1164
+ "gru_intact_nll": {
1165
+ "mean": 2.4698075510185014,
1166
+ "ci95": [
1167
+ 2.369894730087947,
1168
+ 2.5732815028055542
1169
+ ]
1170
+ },
1171
+ "gru_donor_nll": {
1172
+ "mean": 2.6250467260884376,
1173
+ "ci95": [
1174
+ 2.5314117240209546,
1175
+ 2.721516455584175
1176
+ ]
1177
+ },
1178
+ "gru_delta_nll": {
1179
+ "mean": 0.15523917506993587,
1180
+ "ci95": [
1181
+ 0.1397050799581931,
1182
+ 0.17083378776948413
1183
+ ]
1184
+ },
1185
+ "gru_response_error": {
1186
+ "mean": 0.2876848040377506,
1187
+ "ci95": [
1188
+ 0.27845440241345654,
1189
+ 0.2974706192101464
1190
+ ]
1191
+ },
1192
+ "gru_full_nll": {
1193
+ "mean": 2.4698075510185014,
1194
+ "ci95": [
1195
+ 2.369894730087947,
1196
+ 2.5732815028055542
1197
+ ]
1198
+ },
1199
+ "gru_choice_nll": {
1200
+ "mean": 2.560591917239732,
1201
+ "ci95": [
1202
+ 2.4642544195651266,
1203
+ 2.6602050281172898
1204
+ ]
1205
+ },
1206
+ "transformer_intact_nll": {
1207
+ "mean": 2.6154398084941217,
1208
+ "ci95": [
1209
+ 2.5168079947367183,
1210
+ 2.7138013142448343
1211
+ ]
1212
+ },
1213
+ "transformer_donor_nll": {
1214
+ "mean": 2.6315340423144757,
1215
+ "ci95": [
1216
+ 2.5365447755142534,
1217
+ 2.729152883552159
1218
+ ]
1219
+ },
1220
+ "transformer_delta_nll": {
1221
+ "mean": 0.016094233820354173,
1222
+ "ci95": [
1223
+ 0.0035590184174254503,
1224
+ 0.028568504439261653
1225
+ ]
1226
+ },
1227
+ "transformer_response_error": {
1228
+ "mean": 0.35910503126230187,
1229
+ "ci95": [
1230
+ 0.3508898449212565,
1231
+ 0.36753351884435065
1232
+ ]
1233
+ },
1234
+ "transformer_full_nll": {
1235
+ "mean": 2.6154398084941217,
1236
+ "ci95": [
1237
+ 2.5168079947367183,
1238
+ 2.7138013142448343
1239
+ ]
1240
+ },
1241
+ "transformer_choice_nll": {
1242
+ "mean": 2.5741605505376075,
1243
+ "ci95": [
1244
+ 2.474338639043075,
1245
+ 2.6771158429884063
1246
+ ]
1247
+ },
1248
+ "oracle_nll": {
1249
+ "mean": 2.319283828723959,
1250
+ "ci95": [
1251
+ 2.215970742786679,
1252
+ 2.4251377502112526
1253
+ ]
1254
+ },
1255
+ "llama_intact_nll": {
1256
+ "mean": 2.414704988275222,
1257
+ "ci95": [
1258
+ 2.3157662530219927,
1259
+ 2.5174059327438485
1260
+ ]
1261
+ },
1262
+ "llama_donor_nll": {
1263
+ "mean": 2.657213805221217,
1264
+ "ci95": [
1265
+ 2.568709764931531,
1266
+ 2.747369293289025
1267
+ ]
1268
+ },
1269
+ "llama_delta_nll": {
1270
+ "mean": 0.24250881694599516,
1271
+ "ci95": [
1272
+ 0.22129692063629822,
1273
+ 0.26491437588123484
1274
+ ]
1275
+ },
1276
+ "llama_response_error": {
1277
+ "mean": 0.2262315034968623,
1278
+ "ci95": [
1279
+ 0.21497340519820998,
1280
+ 0.23866869816804126
1281
+ ]
1282
+ },
1283
+ "llama_full_nll": {
1284
+ "mean": 2.414704988275222,
1285
+ "ci95": [
1286
+ 2.3157662530219927,
1287
+ 2.5174059327438485
1288
+ ]
1289
+ },
1290
+ "llama_choice_nll": {
1291
+ "mean": 2.544273042956066,
1292
+ "ci95": [
1293
+ 2.4475386055396084,
1294
+ 2.643421831369369
1295
+ ]
1296
+ },
1297
+ "llama_reward_predictive_gain": {
1298
+ "mean": 0.12956805468084437,
1299
+ "ci95": [
1300
+ 0.11431891063622074,
1301
+ 0.14454876293835775
1302
+ ]
1303
+ },
1304
+ "llama_full_token_nll": {
1305
+ "mean": 2.414879560470581,
1306
+ "ci95": [
1307
+ 2.315944504737854,
1308
+ 2.5175853967666626
1309
+ ]
1310
+ },
1311
+ "llama_full_valid_mass": {
1312
+ "mean": 0.999825656414032,
1313
+ "ci95": [
1314
+ 0.9998164147138595,
1315
+ 0.9998342990875244
1316
+ ]
1317
+ },
1318
+ "llama_choice_token_nll": {
1319
+ "mean": 2.5446856021881104,
1320
+ "ci95": [
1321
+ 2.447963297367096,
1322
+ 2.643859750032425
1323
+ ]
1324
+ },
1325
+ "llama_choice_valid_mass": {
1326
+ "mean": 0.9995875358581543,
1327
+ "ci95": [
1328
+ 0.9995646446943283,
1329
+ 0.9996093541383744
1330
+ ]
1331
+ },
1332
+ "llama_intact_nll_minus_oracle": {
1333
+ "mean": 0.09542115955126278,
1334
+ "ci95": [
1335
+ 0.07567326640737465,
1336
+ 0.11458879435465752
1337
+ ]
1338
+ },
1339
+ "llama_response_error_minus_oracle": {
1340
+ "mean": 0.2262315034968623,
1341
+ "ci95": [
1342
+ 0.21497340519820998,
1343
+ 0.23866869816804126
1344
+ ]
1345
+ },
1346
+ "llama_delta_nll_minus_oracle": {
1347
+ "mean": -0.1941843760309871,
1348
+ "ci95": [
1349
+ -0.21973662863400262,
1350
+ -0.16626581625772996
1351
+ ]
1352
+ },
1353
+ "llama_intact_nll_minus_fit": {
1354
+ "mean": 0.07266138903170108,
1355
+ "ci95": [
1356
+ 0.05135998663730955,
1357
+ 0.09365757429620654
1358
+ ]
1359
+ },
1360
+ "llama_response_error_minus_fit": {
1361
+ "mean": 0.15406617090423314,
1362
+ "ci95": [
1363
+ 0.14115603067303967,
1364
+ 0.16786221173196644
1365
+ ]
1366
+ },
1367
+ "llama_delta_nll_minus_fit": {
1368
+ "mean": -0.1963806368226384,
1369
+ "ci95": [
1370
+ -0.22750941528031482,
1371
+ -0.1651794552337908
1372
+ ]
1373
+ },
1374
+ "llama_intact_nll_minus_rw": {
1375
+ "mean": -0.390926652347451,
1376
+ "ci95": [
1377
+ -0.44424724841477786,
1378
+ -0.33294266902490477
1379
+ ]
1380
+ },
1381
+ "llama_response_error_minus_rw": {
1382
+ "mean": -0.08913108118341487,
1383
+ "ci95": [
1384
+ -0.0988288087304757,
1385
+ -0.07950768658234207
1386
+ ]
1387
+ },
1388
+ "llama_delta_nll_minus_rw": {
1389
+ "mean": -0.13830285480174623,
1390
+ "ci95": [
1391
+ -0.18747104214323165,
1392
+ -0.09076236171664488
1393
+ ]
1394
+ },
1395
+ "llama_intact_nll_minus_gru": {
1396
+ "mean": -0.055102562743279675,
1397
+ "ci95": [
1398
+ -0.07281628564151954,
1399
+ -0.03869224345952685
1400
+ ]
1401
+ },
1402
+ "llama_response_error_minus_gru": {
1403
+ "mean": -0.06145330054088832,
1404
+ "ci95": [
1405
+ -0.06573972554135765,
1406
+ -0.05708974318052376
1407
+ ]
1408
+ },
1409
+ "llama_delta_nll_minus_gru": {
1410
+ "mean": 0.08726964187605932,
1411
+ "ci95": [
1412
+ 0.07139073902749318,
1413
+ 0.10330323880821522
1414
+ ]
1415
+ },
1416
+ "llama_intact_nll_minus_transformer": {
1417
+ "mean": -0.20073482021889932,
1418
+ "ci95": [
1419
+ -0.2193266922977887,
1420
+ -0.18085479359690132
1421
+ ]
1422
+ },
1423
+ "llama_response_error_minus_transformer": {
1424
+ "mean": -0.13287352776543954,
1425
+ "ci95": [
1426
+ -0.139932559895035,
1427
+ -0.12608337920528573
1428
+ ]
1429
+ },
1430
+ "llama_delta_nll_minus_transformer": {
1431
+ "mean": 0.22641458312564106,
1432
+ "ci95": [
1433
+ 0.20280454342138285,
1434
+ 0.25026246089310517
1435
+ ]
1436
+ }
1437
+ },
1438
+ "0.7": {
1439
+ "oracle_intact_nll": {
1440
+ "mean": 2.1435564477701416,
1441
+ "ci95": [
1442
+ 2.0159541343855927,
1443
+ 2.2656410205098103
1444
+ ]
1445
+ },
1446
+ "oracle_donor_nll": {
1447
+ "mean": 2.871062122680968,
1448
+ "ci95": [
1449
+ 2.771050651290626,
1450
+ 2.9651100152575345
1451
+ ]
1452
+ },
1453
+ "oracle_delta_nll": {
1454
+ "mean": 0.7275056749108267,
1455
+ "ci95": [
1456
+ 0.6755428709723962,
1457
+ 0.7837150770637379
1458
+ ]
1459
+ },
1460
+ "oracle_response_error": {
1461
+ "mean": 0.0,
1462
+ "ci95": [
1463
+ 0.0,
1464
+ 0.0
1465
+ ]
1466
+ },
1467
+ "fit_intact_nll": {
1468
+ "mean": 2.1702145896796368,
1469
+ "ci95": [
1470
+ 2.044535120248899,
1471
+ 2.2913780175881757
1472
+ ]
1473
+ },
1474
+ "fit_donor_nll": {
1475
+ "mean": 2.8718458628829295,
1476
+ "ci95": [
1477
+ 2.767901620071093,
1478
+ 2.972574590194788
1479
+ ]
1480
+ },
1481
+ "fit_delta_nll": {
1482
+ "mean": 0.7016312732032931,
1483
+ "ci95": [
1484
+ 0.6527425935612794,
1485
+ 0.757010351344706
1486
+ ]
1487
+ },
1488
+ "fit_response_error": {
1489
+ "mean": 0.09135141120397736,
1490
+ "ci95": [
1491
+ 0.07908178328124472,
1492
+ 0.10371831445253293
1493
+ ]
1494
+ },
1495
+ "rw_intact_nll": {
1496
+ "mean": 2.55631368629174,
1497
+ "ci95": [
1498
+ 2.4418039685845145,
1499
+ 2.6609694359300526
1500
+ ]
1501
+ },
1502
+ "rw_donor_nll": {
1503
+ "mean": 3.1023155506208098,
1504
+ "ci95": [
1505
+ 3.0352899066684635,
1506
+ 3.1677688725162674
1507
+ ]
1508
+ },
1509
+ "rw_delta_nll": {
1510
+ "mean": 0.5460018643290695,
1511
+ "ci95": [
1512
+ 0.4891281777276762,
1513
+ 0.6077525759115737
1514
+ ]
1515
+ },
1516
+ "rw_response_error": {
1517
+ "mean": 0.4015211379717528,
1518
+ "ci95": [
1519
+ 0.38037476934737585,
1520
+ 0.4232246683185908
1521
+ ]
1522
+ },
1523
+ "gru_intact_nll": {
1524
+ "mean": 2.3373696046564776,
1525
+ "ci95": [
1526
+ 2.1997238039068523,
1527
+ 2.470009003797405
1528
+ ]
1529
+ },
1530
+ "gru_donor_nll": {
1531
+ "mean": 2.575359011909752,
1532
+ "ci95": [
1533
+ 2.4479691908112295,
1534
+ 2.694542565639238
1535
+ ]
1536
+ },
1537
+ "gru_delta_nll": {
1538
+ "mean": 0.23798940725327455,
1539
+ "ci95": [
1540
+ 0.21504852421153092,
1541
+ 0.2625664992208941
1542
+ ]
1543
+ },
1544
+ "gru_response_error": {
1545
+ "mean": 0.3764652173827853,
1546
+ "ci95": [
1547
+ 0.3651758577581391,
1548
+ 0.38848614637524614
1549
+ ]
1550
+ },
1551
+ "gru_full_nll": {
1552
+ "mean": 2.3373696046564776,
1553
+ "ci95": [
1554
+ 2.1997238039068523,
1555
+ 2.470009003797405
1556
+ ]
1557
+ },
1558
+ "gru_choice_nll": {
1559
+ "mean": 2.4435253696905175,
1560
+ "ci95": [
1561
+ 2.312274576152947,
1562
+ 2.571105610966659
1563
+ ]
1564
+ },
1565
+ "transformer_intact_nll": {
1566
+ "mean": 2.4937030065468395,
1567
+ "ci95": [
1568
+ 2.360721876964319,
1569
+ 2.6284748750254256
1570
+ ]
1571
+ },
1572
+ "transformer_donor_nll": {
1573
+ "mean": 2.5077245637863577,
1574
+ "ci95": [
1575
+ 2.377005762955793,
1576
+ 2.6373347475548568
1577
+ ]
1578
+ },
1579
+ "transformer_delta_nll": {
1580
+ "mean": 0.014021557239517557,
1581
+ "ci95": [
1582
+ 0.0025055985462562992,
1583
+ 0.024780594294718662
1584
+ ]
1585
+ },
1586
+ "transformer_response_error": {
1587
+ "mean": 0.4662228174648522,
1588
+ "ci95": [
1589
+ 0.4558828834752213,
1590
+ 0.47718799263244
1591
+ ]
1592
+ },
1593
+ "transformer_full_nll": {
1594
+ "mean": 2.4937030065468395,
1595
+ "ci95": [
1596
+ 2.360721876964319,
1597
+ 2.6284748750254256
1598
+ ]
1599
+ },
1600
+ "transformer_choice_nll": {
1601
+ "mean": 2.452637400242134,
1602
+ "ci95": [
1603
+ 2.323421033681237,
1604
+ 2.581084595169812
1605
+ ]
1606
+ },
1607
+ "oracle_nll": {
1608
+ "mean": 2.1435564477701416,
1609
+ "ci95": [
1610
+ 2.0159541343855927,
1611
+ 2.2656410205098103
1612
+ ]
1613
+ },
1614
+ "llama_intact_nll": {
1615
+ "mean": 2.2510137311827587,
1616
+ "ci95": [
1617
+ 2.122494384717716,
1618
+ 2.3720210707871052
1619
+ ]
1620
+ },
1621
+ "llama_donor_nll": {
1622
+ "mean": 2.666239711469099,
1623
+ "ci95": [
1624
+ 2.550409443472277,
1625
+ 2.770915753081823
1626
+ ]
1627
+ },
1628
+ "llama_delta_nll": {
1629
+ "mean": 0.4152259802863405,
1630
+ "ci95": [
1631
+ 0.38473049580230745,
1632
+ 0.44627916012573327
1633
+ ]
1634
+ },
1635
+ "llama_response_error": {
1636
+ "mean": 0.2842295694569437,
1637
+ "ci95": [
1638
+ 0.2701170254472702,
1639
+ 0.2997595852815924
1640
+ ]
1641
+ },
1642
+ "llama_full_nll": {
1643
+ "mean": 2.2510137311827587,
1644
+ "ci95": [
1645
+ 2.122494384717716,
1646
+ 2.3720210707871052
1647
+ ]
1648
+ },
1649
+ "llama_choice_nll": {
1650
+ "mean": 2.4189901027790803,
1651
+ "ci95": [
1652
+ 2.2905766582241265,
1653
+ 2.5415705206814843
1654
+ ]
1655
+ },
1656
+ "llama_reward_predictive_gain": {
1657
+ "mean": 0.1679763715963224,
1658
+ "ci95": [
1659
+ 0.14920931286576464,
1660
+ 0.1865610506148221
1661
+ ]
1662
+ },
1663
+ "llama_full_token_nll": {
1664
+ "mean": 2.251458168029785,
1665
+ "ci95": [
1666
+ 2.122929209470749,
1667
+ 2.3725016117095947
1668
+ ]
1669
+ },
1670
+ "llama_full_valid_mass": {
1671
+ "mean": 0.9995555877685547,
1672
+ "ci95": [
1673
+ 0.9995307296514511,
1674
+ 0.9995803847908974
1675
+ ]
1676
+ },
1677
+ "llama_choice_token_nll": {
1678
+ "mean": 2.419452428817749,
1679
+ "ci95": [
1680
+ 2.291048389673233,
1681
+ 2.5420177400112154
1682
+ ]
1683
+ },
1684
+ "llama_choice_valid_mass": {
1685
+ "mean": 0.9995377063751221,
1686
+ "ci95": [
1687
+ 0.9995033740997314,
1688
+ 0.9995700135827065
1689
+ ]
1690
+ },
1691
+ "llama_intact_nll_minus_oracle": {
1692
+ "mean": 0.10745728341261711,
1693
+ "ci95": [
1694
+ 0.08875702781453025,
1695
+ 0.12541283812661302
1696
+ ]
1697
+ },
1698
+ "llama_response_error_minus_oracle": {
1699
+ "mean": 0.2842295694569437,
1700
+ "ci95": [
1701
+ 0.2701170254472702,
1702
+ 0.2997595852815924
1703
+ ]
1704
+ },
1705
+ "llama_delta_nll_minus_oracle": {
1706
+ "mean": -0.3122796946244862,
1707
+ "ci95": [
1708
+ -0.34997939821619184,
1709
+ -0.278614807806386
1710
+ ]
1711
+ },
1712
+ "llama_intact_nll_minus_fit": {
1713
+ "mean": 0.08079914150312163,
1714
+ "ci95": [
1715
+ 0.05929212665409138,
1716
+ 0.10110494143613911
1717
+ ]
1718
+ },
1719
+ "llama_response_error_minus_fit": {
1720
+ "mean": 0.19287815825296645,
1721
+ "ci95": [
1722
+ 0.17552208993525636,
1723
+ 0.21173690422630123
1724
+ ]
1725
+ },
1726
+ "llama_delta_nll_minus_fit": {
1727
+ "mean": -0.28640529291695266,
1728
+ "ci95": [
1729
+ -0.3232877803531259,
1730
+ -0.25148821446963937
1731
+ ]
1732
+ },
1733
+ "llama_intact_nll_minus_rw": {
1734
+ "mean": -0.30529995510898206,
1735
+ "ci95": [
1736
+ -0.3499914736710208,
1737
+ -0.2612940707383876
1738
+ ]
1739
+ },
1740
+ "llama_response_error_minus_rw": {
1741
+ "mean": -0.11729156851480908,
1742
+ "ci95": [
1743
+ -0.12918184950665285,
1744
+ -0.10579851892724626
1745
+ ]
1746
+ },
1747
+ "llama_delta_nll_minus_rw": {
1748
+ "mean": -0.13077588404272888,
1749
+ "ci95": [
1750
+ -0.17706114071421877,
1751
+ -0.08572322823053656
1752
+ ]
1753
+ },
1754
+ "llama_intact_nll_minus_gru": {
1755
+ "mean": -0.0863558734737192,
1756
+ "ci95": [
1757
+ -0.10918040534591868,
1758
+ -0.06461610620449017
1759
+ ]
1760
+ },
1761
+ "llama_response_error_minus_gru": {
1762
+ "mean": -0.0922356479258415,
1763
+ "ci95": [
1764
+ -0.09780486176442141,
1765
+ -0.08620316105968338
1766
+ ]
1767
+ },
1768
+ "llama_delta_nll_minus_gru": {
1769
+ "mean": 0.17723657303306592,
1770
+ "ci95": [
1771
+ 0.16066894096562157,
1772
+ 0.19484357402957062
1773
+ ]
1774
+ },
1775
+ "llama_intact_nll_minus_transformer": {
1776
+ "mean": -0.24268927536408147,
1777
+ "ci95": [
1778
+ -0.2639374583219686,
1779
+ -0.22244619893916637
1780
+ ]
1781
+ },
1782
+ "llama_response_error_minus_transformer": {
1783
+ "mean": -0.18199324800790845,
1784
+ "ci95": [
1785
+ -0.19161802504201728,
1786
+ -0.17256474456825935
1787
+ ]
1788
+ },
1789
+ "llama_delta_nll_minus_transformer": {
1790
+ "mean": 0.4012044230468229,
1791
+ "ci95": [
1792
+ 0.3705955893707403,
1793
+ 0.4319986397815301
1794
+ ]
1795
+ }
1796
+ },
1797
+ "0.9": {
1798
+ "oracle_intact_nll": {
1799
+ "mean": 1.965550659603369,
1800
+ "ci95": [
1801
+ 1.8382024060606479,
1802
+ 2.0891000379175035
1803
+ ]
1804
+ },
1805
+ "oracle_donor_nll": {
1806
+ "mean": 3.077029769914495,
1807
+ "ci95": [
1808
+ 2.977664020324591,
1809
+ 3.1650890310729984
1810
+ ]
1811
+ },
1812
+ "oracle_delta_nll": {
1813
+ "mean": 1.1114791103111255,
1814
+ "ci95": [
1815
+ 1.0353625793274794,
1816
+ 1.189223306976189
1817
+ ]
1818
+ },
1819
+ "oracle_response_error": {
1820
+ "mean": 0.0,
1821
+ "ci95": [
1822
+ 0.0,
1823
+ 0.0
1824
+ ]
1825
+ },
1826
+ "fit_intact_nll": {
1827
+ "mean": 1.989387447624577,
1828
+ "ci95": [
1829
+ 1.8592236560998066,
1830
+ 2.1119541075664086
1831
+ ]
1832
+ },
1833
+ "fit_donor_nll": {
1834
+ "mean": 3.105151471359101,
1835
+ "ci95": [
1836
+ 2.9953974567649815,
1837
+ 3.2069870618685328
1838
+ ]
1839
+ },
1840
+ "fit_delta_nll": {
1841
+ "mean": 1.1157640237345232,
1842
+ "ci95": [
1843
+ 1.0295745180294773,
1844
+ 1.2004888669786236
1845
+ ]
1846
+ },
1847
+ "fit_response_error": {
1848
+ "mean": 0.08758912483241174,
1849
+ "ci95": [
1850
+ 0.07303037041559458,
1851
+ 0.1043730674785598
1852
+ ]
1853
+ },
1854
+ "rw_intact_nll": {
1855
+ "mean": 2.2988755929499636,
1856
+ "ci95": [
1857
+ 2.1731721590488853,
1858
+ 2.42485811990197
1859
+ ]
1860
+ },
1861
+ "rw_donor_nll": {
1862
+ "mean": 3.001340023906025,
1863
+ "ci95": [
1864
+ 2.915388949156131,
1865
+ 3.0838527253928802
1866
+ ]
1867
+ },
1868
+ "rw_delta_nll": {
1869
+ "mean": 0.7024644309560609,
1870
+ "ci95": [
1871
+ 0.6391258291040296,
1872
+ 0.7659256076498202
1873
+ ]
1874
+ },
1875
+ "rw_response_error": {
1876
+ "mean": 0.46638960447755695,
1877
+ "ci95": [
1878
+ 0.43663822206139463,
1879
+ 0.4972329494845311
1880
+ ]
1881
+ },
1882
+ "gru_intact_nll": {
1883
+ "mean": 2.1560211276969277,
1884
+ "ci95": [
1885
+ 2.0233599017969452,
1886
+ 2.285698249896349
1887
+ ]
1888
+ },
1889
+ "gru_donor_nll": {
1890
+ "mean": 2.5309783087694346,
1891
+ "ci95": [
1892
+ 2.4003759190366365,
1893
+ 2.6544608216931116
1894
+ ]
1895
+ },
1896
+ "gru_delta_nll": {
1897
+ "mean": 0.3749571810725067,
1898
+ "ci95": [
1899
+ 0.34535636338354586,
1900
+ 0.40565186318236596
1901
+ ]
1902
+ },
1903
+ "gru_response_error": {
1904
+ "mean": 0.43442578582427266,
1905
+ "ci95": [
1906
+ 0.4170868498347895,
1907
+ 0.45154745158138637
1908
+ ]
1909
+ },
1910
+ "gru_full_nll": {
1911
+ "mean": 2.1560211276969277,
1912
+ "ci95": [
1913
+ 2.0233599017969452,
1914
+ 2.285698249896349
1915
+ ]
1916
+ },
1917
+ "gru_choice_nll": {
1918
+ "mean": 2.3215741339591087,
1919
+ "ci95": [
1920
+ 2.1851232503997142,
1921
+ 2.4508245885274915
1922
+ ]
1923
+ },
1924
+ "transformer_intact_nll": {
1925
+ "mean": 2.3759375327354832,
1926
+ "ci95": [
1927
+ 2.237223435366887,
1928
+ 2.507407954043828
1929
+ ]
1930
+ },
1931
+ "transformer_donor_nll": {
1932
+ "mean": 2.3867767376463767,
1933
+ "ci95": [
1934
+ 2.2494086097691603,
1935
+ 2.517031520655869
1936
+ ]
1937
+ },
1938
+ "transformer_delta_nll": {
1939
+ "mean": 0.010839204910892887,
1940
+ "ci95": [
1941
+ -0.0050139926210633476,
1942
+ 0.024369903512472868
1943
+ ]
1944
+ },
1945
+ "transformer_response_error": {
1946
+ "mean": 0.5608704789459782,
1947
+ "ci95": [
1948
+ 0.5463619754331537,
1949
+ 0.5754792414155258
1950
+ ]
1951
+ },
1952
+ "transformer_full_nll": {
1953
+ "mean": 2.3759375327354832,
1954
+ "ci95": [
1955
+ 2.237223435366887,
1956
+ 2.507407954043828
1957
+ ]
1958
+ },
1959
+ "transformer_choice_nll": {
1960
+ "mean": 2.3362569108904085,
1961
+ "ci95": [
1962
+ 2.202116897656126,
1963
+ 2.4649861695215285
1964
+ ]
1965
+ },
1966
+ "oracle_nll": {
1967
+ "mean": 1.965550659603369,
1968
+ "ci95": [
1969
+ 1.8382024060606479,
1970
+ 2.0891000379175035
1971
+ ]
1972
+ },
1973
+ "llama_intact_nll": {
1974
+ "mean": 2.0700606486648856,
1975
+ "ci95": [
1976
+ 1.9352210499508247,
1977
+ 2.200384264629768
1978
+ ]
1979
+ },
1980
+ "llama_donor_nll": {
1981
+ "mean": 2.673521485182517,
1982
+ "ci95": [
1983
+ 2.5493013459511196,
1984
+ 2.792119728410036
1985
+ ]
1986
+ },
1987
+ "llama_delta_nll": {
1988
+ "mean": 0.6034608365176316,
1989
+ "ci95": [
1990
+ 0.5636719203067142,
1991
+ 0.6404049859452534
1992
+ ]
1993
+ },
1994
+ "llama_response_error": {
1995
+ "mean": 0.3320756851926578,
1996
+ "ci95": [
1997
+ 0.31305580851969456,
1998
+ 0.3513696180223555
1999
+ ]
2000
+ },
2001
+ "llama_full_nll": {
2002
+ "mean": 2.0700606486648856,
2003
+ "ci95": [
2004
+ 1.9352210499508247,
2005
+ 2.200384264629768
2006
+ ]
2007
+ },
2008
+ "llama_choice_nll": {
2009
+ "mean": 2.2991542415836577,
2010
+ "ci95": [
2011
+ 2.1670168032072885,
2012
+ 2.4258120437667134
2013
+ ]
2014
+ },
2015
+ "llama_reward_predictive_gain": {
2016
+ "mean": 0.22909359291877163,
2017
+ "ci95": [
2018
+ 0.20556610874214784,
2019
+ 0.2515510994488256
2020
+ ]
2021
+ },
2022
+ "llama_full_token_nll": {
2023
+ "mean": 2.070363998413086,
2024
+ "ci95": [
2025
+ 1.9355262130498887,
2026
+ 2.200701171159744
2027
+ ]
2028
+ },
2029
+ "llama_full_valid_mass": {
2030
+ "mean": 0.9996969699859619,
2031
+ "ci95": [
2032
+ 0.9996790289878845,
2033
+ 0.9997138991951943
2034
+ ]
2035
+ },
2036
+ "llama_choice_token_nll": {
2037
+ "mean": 2.2994298934936523,
2038
+ "ci95": [
2039
+ 2.167295277118683,
2040
+ 2.426094841957092
2041
+ ]
2042
+ },
2043
+ "llama_choice_valid_mass": {
2044
+ "mean": 0.9997243285179138,
2045
+ "ci95": [
2046
+ 0.9997108578681946,
2047
+ 0.9997375622391701
2048
+ ]
2049
+ },
2050
+ "llama_intact_nll_minus_oracle": {
2051
+ "mean": 0.10450998906151679,
2052
+ "ci95": [
2053
+ 0.0809611925494224,
2054
+ 0.12954730573825635
2055
+ ]
2056
+ },
2057
+ "llama_response_error_minus_oracle": {
2058
+ "mean": 0.3320756851926578,
2059
+ "ci95": [
2060
+ 0.31305580851969456,
2061
+ 0.3513696180223555
2062
+ ]
2063
+ },
2064
+ "llama_delta_nll_minus_oracle": {
2065
+ "mean": -0.5080182737934937,
2066
+ "ci95": [
2067
+ -0.5760189994742417,
2068
+ -0.44207797635673984
2069
+ ]
2070
+ },
2071
+ "llama_intact_nll_minus_fit": {
2072
+ "mean": 0.08067320104030888,
2073
+ "ci95": [
2074
+ 0.052152533063324076,
2075
+ 0.10989678874340426
2076
+ ]
2077
+ },
2078
+ "llama_response_error_minus_fit": {
2079
+ "mean": 0.24448656036024605,
2080
+ "ci95": [
2081
+ 0.21805078182682192,
2082
+ 0.27046270764419966
2083
+ ]
2084
+ },
2085
+ "llama_delta_nll_minus_fit": {
2086
+ "mean": -0.5123031872168917,
2087
+ "ci95": [
2088
+ -0.5859744835670385,
2089
+ -0.4400441577618669
2090
+ ]
2091
+ },
2092
+ "llama_intact_nll_minus_rw": {
2093
+ "mean": -0.2288149442850777,
2094
+ "ci95": [
2095
+ -0.272373098423258,
2096
+ -0.1867310801008471
2097
+ ]
2098
+ },
2099
+ "llama_response_error_minus_rw": {
2100
+ "mean": -0.13431391928489916,
2101
+ "ci95": [
2102
+ -0.1517097050402328,
2103
+ -0.11730550586194048
2104
+ ]
2105
+ },
2106
+ "llama_delta_nll_minus_rw": {
2107
+ "mean": -0.09900359443842936,
2108
+ "ci95": [
2109
+ -0.1525474119237978,
2110
+ -0.04307130914138562
2111
+ ]
2112
+ },
2113
+ "llama_intact_nll_minus_gru": {
2114
+ "mean": -0.08596047903204203,
2115
+ "ci95": [
2116
+ -0.10461677697889198,
2117
+ -0.06655069571476564
2118
+ ]
2119
+ },
2120
+ "llama_response_error_minus_gru": {
2121
+ "mean": -0.10235010063161493,
2122
+ "ci95": [
2123
+ -0.10989639593653257,
2124
+ -0.09512189174197762
2125
+ ]
2126
+ },
2127
+ "llama_delta_nll_minus_gru": {
2128
+ "mean": 0.22850365544512488,
2129
+ "ci95": [
2130
+ 0.2020664330308164,
2131
+ 0.25635792085827747
2132
+ ]
2133
+ },
2134
+ "llama_intact_nll_minus_transformer": {
2135
+ "mean": -0.3058768840705979,
2136
+ "ci95": [
2137
+ -0.33678422624306165,
2138
+ -0.27359995238125756
2139
+ ]
2140
+ },
2141
+ "llama_response_error_minus_transformer": {
2142
+ "mean": -0.22879479375332035,
2143
+ "ci95": [
2144
+ -0.2404475575132055,
2145
+ -0.21721059945006885
2146
+ ]
2147
+ },
2148
+ "llama_delta_nll_minus_transformer": {
2149
+ "mean": 0.5926216316067386,
2150
+ "ci95": [
2151
+ 0.5477566940406424,
2152
+ 0.6365088853061666
2153
+ ]
2154
+ }
2155
+ },
2156
+ "1.0": {
2157
+ "oracle_intact_nll": {
2158
+ "mean": 1.8409537880893716,
2159
+ "ci95": [
2160
+ 1.69340417839034,
2161
+ 1.9846836798027525
2162
+ ]
2163
+ },
2164
+ "oracle_donor_nll": {
2165
+ "mean": 3.344660315048634,
2166
+ "ci95": [
2167
+ 3.24565980727262,
2168
+ 3.438878124534093
2169
+ ]
2170
+ },
2171
+ "oracle_delta_nll": {
2172
+ "mean": 1.503706526959262,
2173
+ "ci95": [
2174
+ 1.3943931517691155,
2175
+ 1.6241764870671251
2176
+ ]
2177
+ },
2178
+ "oracle_response_error": {
2179
+ "mean": 0.0,
2180
+ "ci95": [
2181
+ 0.0,
2182
+ 0.0
2183
+ ]
2184
+ },
2185
+ "fit_intact_nll": {
2186
+ "mean": 1.85205902407303,
2187
+ "ci95": [
2188
+ 1.7073344746824741,
2189
+ 1.9946343361928758
2190
+ ]
2191
+ },
2192
+ "fit_donor_nll": {
2193
+ "mean": 3.2959482457335247,
2194
+ "ci95": [
2195
+ 3.1894709207424197,
2196
+ 3.399085471724626
2197
+ ]
2198
+ },
2199
+ "fit_delta_nll": {
2200
+ "mean": 1.4438892216604955,
2201
+ "ci95": [
2202
+ 1.3266872473142837,
2203
+ 1.5718226756755076
2204
+ ]
2205
+ },
2206
+ "fit_response_error": {
2207
+ "mean": 0.08293078847487799,
2208
+ "ci95": [
2209
+ 0.07136590429310782,
2210
+ 0.09527586474564628
2211
+ ]
2212
+ },
2213
+ "rw_intact_nll": {
2214
+ "mean": 2.182240777872362,
2215
+ "ci95": [
2216
+ 2.037978379296234,
2217
+ 2.3253016925257044
2218
+ ]
2219
+ },
2220
+ "rw_donor_nll": {
2221
+ "mean": 2.9628095374735746,
2222
+ "ci95": [
2223
+ 2.854472799920614,
2224
+ 3.0668661621173405
2225
+ ]
2226
+ },
2227
+ "rw_delta_nll": {
2228
+ "mean": 0.7805687596012121,
2229
+ "ci95": [
2230
+ 0.7223996285194172,
2231
+ 0.8395591263590166
2232
+ ]
2233
+ },
2234
+ "rw_response_error": {
2235
+ "mean": 0.48694770469193427,
2236
+ "ci95": [
2237
+ 0.4586505485735272,
2238
+ 0.5145272442178125
2239
+ ]
2240
+ },
2241
+ "gru_intact_nll": {
2242
+ "mean": 2.0615749176865354,
2243
+ "ci95": [
2244
+ 1.9050403284820128,
2245
+ 2.2182691723729158
2246
+ ]
2247
+ },
2248
+ "gru_donor_nll": {
2249
+ "mean": 2.490458816793799,
2250
+ "ci95": [
2251
+ 2.336341070814708,
2252
+ 2.6367704886574392
2253
+ ]
2254
+ },
2255
+ "gru_delta_nll": {
2256
+ "mean": 0.42888389910726343,
2257
+ "ci95": [
2258
+ 0.38640727494611526,
2259
+ 0.4694190823899969
2260
+ ]
2261
+ },
2262
+ "gru_response_error": {
2263
+ "mean": 0.47156827265382206,
2264
+ "ci95": [
2265
+ 0.45450496680639135,
2266
+ 0.4882113786669313
2267
+ ]
2268
+ },
2269
+ "gru_full_nll": {
2270
+ "mean": 2.0615749176865354,
2271
+ "ci95": [
2272
+ 1.9050403284820128,
2273
+ 2.2182691723729158
2274
+ ]
2275
+ },
2276
+ "gru_choice_nll": {
2277
+ "mean": 2.2329980957421025,
2278
+ "ci95": [
2279
+ 2.076134840078994,
2280
+ 2.385221468320119
2281
+ ]
2282
+ },
2283
+ "transformer_intact_nll": {
2284
+ "mean": 2.283559003037639,
2285
+ "ci95": [
2286
+ 2.122726693768307,
2287
+ 2.4448122117462296
2288
+ ]
2289
+ },
2290
+ "transformer_donor_nll": {
2291
+ "mean": 2.298194320301616,
2292
+ "ci95": [
2293
+ 2.139506217368428,
2294
+ 2.453585895814056
2295
+ ]
2296
+ },
2297
+ "transformer_delta_nll": {
2298
+ "mean": 0.014635317263977026,
2299
+ "ci95": [
2300
+ 0.0031516652533479417,
2301
+ 0.025456896991478142
2302
+ ]
2303
+ },
2304
+ "transformer_response_error": {
2305
+ "mean": 0.6047550646863844,
2306
+ "ci95": [
2307
+ 0.5903452006943551,
2308
+ 0.6197256267873892
2309
+ ]
2310
+ },
2311
+ "transformer_full_nll": {
2312
+ "mean": 2.283559003037639,
2313
+ "ci95": [
2314
+ 2.122726693768307,
2315
+ 2.4448122117462296
2316
+ ]
2317
+ },
2318
+ "transformer_choice_nll": {
2319
+ "mean": 2.262184366134564,
2320
+ "ci95": [
2321
+ 2.0978098308708124,
2322
+ 2.424228047042848
2323
+ ]
2324
+ },
2325
+ "oracle_nll": {
2326
+ "mean": 1.8409537880893716,
2327
+ "ci95": [
2328
+ 1.69340417839034,
2329
+ 1.9846836798027525
2330
+ ]
2331
+ },
2332
+ "llama_intact_nll": {
2333
+ "mean": 1.9458607498465463,
2334
+ "ci95": [
2335
+ 1.7952800593747194,
2336
+ 2.0909579866150034
2337
+ ]
2338
+ },
2339
+ "llama_donor_nll": {
2340
+ "mean": 2.660933692350797,
2341
+ "ci95": [
2342
+ 2.5101252182769254,
2343
+ 2.808147252593895
2344
+ ]
2345
+ },
2346
+ "llama_delta_nll": {
2347
+ "mean": 0.7150729425042506,
2348
+ "ci95": [
2349
+ 0.6769010239322747,
2350
+ 0.7548949238328967
2351
+ ]
2352
+ },
2353
+ "llama_response_error": {
2354
+ "mean": 0.3679081763676275,
2355
+ "ci95": [
2356
+ 0.34972665858767976,
2357
+ 0.38717806002863675
2358
+ ]
2359
+ },
2360
+ "llama_full_nll": {
2361
+ "mean": 1.9458607498465463,
2362
+ "ci95": [
2363
+ 1.7952800593747194,
2364
+ 2.0909579866150034
2365
+ ]
2366
+ },
2367
+ "llama_choice_nll": {
2368
+ "mean": 2.19591801244102,
2369
+ "ci95": [
2370
+ 2.0370182757496655,
2371
+ 2.3519438757927493
2372
+ ]
2373
+ },
2374
+ "llama_reward_predictive_gain": {
2375
+ "mean": 0.2500572625944739,
2376
+ "ci95": [
2377
+ 0.2258001188455806,
2378
+ 0.2753454848194946
2379
+ ]
2380
+ },
2381
+ "llama_full_token_nll": {
2382
+ "mean": 1.9462753534317017,
2383
+ "ci95": [
2384
+ 1.7957097440958023,
2385
+ 2.0913975954055783
2386
+ ]
2387
+ },
2388
+ "llama_full_valid_mass": {
2389
+ "mean": 0.9995853900909424,
2390
+ "ci95": [
2391
+ 0.9995518311858177,
2392
+ 0.9996174588799477
2393
+ ]
2394
+ },
2395
+ "llama_choice_token_nll": {
2396
+ "mean": 2.196246385574341,
2397
+ "ci95": [
2398
+ 2.0373442828655244,
2399
+ 2.3522961676120757
2400
+ ]
2401
+ },
2402
+ "llama_choice_valid_mass": {
2403
+ "mean": 0.9996713995933533,
2404
+ "ci95": [
2405
+ 0.9996465399861336,
2406
+ 0.9996958985924721
2407
+ ]
2408
+ },
2409
+ "llama_intact_nll_minus_oracle": {
2410
+ "mean": 0.10490696175717443,
2411
+ "ci95": [
2412
+ 0.08401307134906591,
2413
+ 0.12739640481193612
2414
+ ]
2415
+ },
2416
+ "llama_response_error_minus_oracle": {
2417
+ "mean": 0.3679081763676275,
2418
+ "ci95": [
2419
+ 0.34972665858767976,
2420
+ 0.38717806002863675
2421
+ ]
2422
+ },
2423
+ "llama_delta_nll_minus_oracle": {
2424
+ "mean": -0.7886335844550112,
2425
+ "ci95": [
2426
+ -0.9008320751658063,
2427
+ -0.6917021489646574
2428
+ ]
2429
+ },
2430
+ "llama_intact_nll_minus_fit": {
2431
+ "mean": 0.09380172577351648,
2432
+ "ci95": [
2433
+ 0.0731085779301203,
2434
+ 0.11564362128458723
2435
+ ]
2436
+ },
2437
+ "llama_response_error_minus_fit": {
2438
+ "mean": 0.28497738789274957,
2439
+ "ci95": [
2440
+ 0.26688398058268525,
2441
+ 0.30285068958172606
2442
+ ]
2443
+ },
2444
+ "llama_delta_nll_minus_fit": {
2445
+ "mean": -0.7288162791562448,
2446
+ "ci95": [
2447
+ -0.8436742776907835,
2448
+ -0.6211136479172106
2449
+ ]
2450
+ },
2451
+ "llama_intact_nll_minus_rw": {
2452
+ "mean": -0.23638002802581545,
2453
+ "ci95": [
2454
+ -0.27932886976427523,
2455
+ -0.1923659436755912
2456
+ ]
2457
+ },
2458
+ "llama_response_error_minus_rw": {
2459
+ "mean": -0.11903952832430667,
2460
+ "ci95": [
2461
+ -0.1373848141627428,
2462
+ -0.10031392600180658
2463
+ ]
2464
+ },
2465
+ "llama_delta_nll_minus_rw": {
2466
+ "mean": -0.06549581709696141,
2467
+ "ci95": [
2468
+ -0.12372847732869362,
2469
+ -0.005930444561727195
2470
+ ]
2471
+ },
2472
+ "llama_intact_nll_minus_gru": {
2473
+ "mean": -0.11571416783998918,
2474
+ "ci95": [
2475
+ -0.14567882734268867,
2476
+ -0.08909519671889116
2477
+ ]
2478
+ },
2479
+ "llama_response_error_minus_gru": {
2480
+ "mean": -0.10366009628619453,
2481
+ "ci95": [
2482
+ -0.11277802741928496,
2483
+ -0.09487410677529073
2484
+ ]
2485
+ },
2486
+ "llama_delta_nll_minus_gru": {
2487
+ "mean": 0.2861890433969872,
2488
+ "ci95": [
2489
+ 0.25409441377324327,
2490
+ 0.32246790246822643
2491
+ ]
2492
+ },
2493
+ "llama_intact_nll_minus_transformer": {
2494
+ "mean": -0.3376982531910924,
2495
+ "ci95": [
2496
+ -0.36986510977408193,
2497
+ -0.3081927333893907
2498
+ ]
2499
+ },
2500
+ "llama_response_error_minus_transformer": {
2501
+ "mean": -0.2368468883187569,
2502
+ "ci95": [
2503
+ -0.25023625307798236,
2504
+ -0.2242495499513507
2505
+ ]
2506
+ },
2507
+ "llama_delta_nll_minus_transformer": {
2508
+ "mean": 0.7004376252402738,
2509
+ "ci95": [
2510
+ 0.6598371778716663,
2511
+ 0.7414071522506882
2512
+ ]
2513
+ }
2514
+ }
2515
+ }
review_artifacts/archive/outputs/test_main_integrated_20260925/spatial/COMPLETE.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "split": "test",
3
+ "test_used": true,
4
+ "scientific_audit_passed": true,
5
+ "main_complete": true,
6
+ "models": [
7
+ "oracle",
8
+ "fit",
9
+ "rw",
10
+ "gru",
11
+ "transformer",
12
+ "llama"
13
+ ],
14
+ "weights": [
15
+ "000",
16
+ "010",
17
+ "030",
18
+ "050",
19
+ "070",
20
+ "090",
21
+ "100"
22
+ ],
23
+ "donors": 20,
24
+ "participants_per_weight": 50,
25
+ "scoring": "maps7-8",
26
+ "appendix_operations_included": false
27
+ }
review_artifacts/data/llama_test_20260925/MANIFEST.json ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "split": "test",
3
+ "weights": [
4
+ "000",
5
+ "010",
6
+ "030",
7
+ "050",
8
+ "070",
9
+ "090",
10
+ "100"
11
+ ],
12
+ "donor_seeds": {
13
+ "task_a": 20260917,
14
+ "task_b": 81073
15
+ },
16
+ "donors": 20,
17
+ "scoring": {
18
+ "task_a": "trials151-200",
19
+ "task_b": "maps7-8"
20
+ },
21
+ "source_hashes": {
22
+ "outputs/weight_curve_a_20260915/reward_w000/test_000.npz": "842a673ee7866225b9fa30360c99a96446f572c6b7473763e960c2a5467dea49",
23
+ "outputs/spatial_pilot_v1/reward_w000/test_observable.jsonl": "8fa9d408561de77947e2795cb3d29e9d2935e8a583895bedcfd67cfd23f53d51",
24
+ "outputs/weight_curve_a_20260915/reward_w010/test_000.npz": "3f8611eaaaddd42eb4e106f26e63116113b1e6ab45ba7d63aa68a075629f0003",
25
+ "outputs/spatial_pilot_v1/reward_w010/test_observable.jsonl": "626c3ae66d49a40d1a66aaf6f294ebd069acf0e6f6fcdc73f3f21ba7ac4028d8",
26
+ "outputs/weight_curve_a_20260915/reward_w030/test_000.npz": "efd97f57fe9e19d157dfca3716b55d47f04cf5df9d01cb71526018dc603b7397",
27
+ "outputs/spatial_pilot_v1/reward_w030/test_observable.jsonl": "8bcde34d86db093c287f80c9edb39ad6ae0e14a363821992d1ff781d526f9433",
28
+ "outputs/weight_curve_a_20260915/reward_w050/test_000.npz": "73fa290c00fc1d3fdcea4608690ad0cc5c67f9d4b1f15cbccb829e26858fc5bb",
29
+ "outputs/spatial_pilot_v1/reward_w050/test_observable.jsonl": "941399f63007d976cd58da50ba4bc2c68ef9c6a62dcb9675d72f49d4f95fd9bc",
30
+ "outputs/weight_curve_a_20260915/reward_w070/test_000.npz": "1d9dde0bfe62a01d0a724ac8004225b4ed984718505194d029a9bd2c06acd342",
31
+ "outputs/spatial_pilot_v1/reward_w070/test_observable.jsonl": "4147d610e789a31a682c0cd0bfaaa59964fc6dbaa43ecd758808e9674f214f9a",
32
+ "outputs/weight_curve_a_20260915/reward_w090/test_000.npz": "386e7436c517c5c924506901ca0bc19012862f66a0590a1f3164d33da20fbab9",
33
+ "outputs/spatial_pilot_v1/reward_w090/test_observable.jsonl": "8dc7882c99f888342a0ca2a0eb5b2f5c1f7d41c43621890859bf75d1ad23a366",
34
+ "outputs/weight_curve_a_20260915/reward_w100/test_000.npz": "829fbf1333f2bb610c0b90e42af2a196e846b159bb38455d608a4b57f0344d93",
35
+ "outputs/spatial_pilot_v1/reward_w100/test_observable.jsonl": "924f3e41d285c2b9d87325121dba7b7562f1cd00f496a1c7ab6be07d2c135baf"
36
+ },
37
+ "files": {
38
+ "task_a/choice/reward_w000.jsonl": "4402e83dec482c2edbc8c9693615bc7b9fe186781688967fac494608a2e7f0db",
39
+ "task_a/choice/reward_w010.jsonl": "939908d86980cd06601c4d4a9c764e7d4fde1acb7c85efde56cde063f931907b",
40
+ "task_a/choice/reward_w030.jsonl": "95e80223444113b760a732b83f3a2874b9b53480169708e933c98e173f7d4ee0",
41
+ "task_a/choice/reward_w050.jsonl": "7052d13728a39c6a9f948fe91c0df48a9fedcc7ec207d667a33761e48b4d9371",
42
+ "task_a/choice/reward_w070.jsonl": "175ee2d0c9b9fd1a4f037acbd47b582681586cb06ee7923e04dd5b3c2190081c",
43
+ "task_a/choice/reward_w090.jsonl": "c7a1aa1ed314bf523cebdef0d3fcf927895adfba56aa1724281486dcd6d4566c",
44
+ "task_a/choice/reward_w100.jsonl": "868c23c3992089af44fc4a3655867d069af38a0b5f688016b576fa4ca646943c",
45
+ "task_a/full/reward_w000.jsonl": "e37a850a26e7ac1cda1a8c0a263fd0102dc7b73df8465a89352257ce7e0c5b62",
46
+ "task_a/full/reward_w010.jsonl": "31642ee0ab02fa711ae7930a19d35adc50be089f589e4e2bef844cb43c177170",
47
+ "task_a/full/reward_w030.jsonl": "0fe68b204eca551f3df93d03118d60d0a565324ad33a3d08e77b825b2ed341ee",
48
+ "task_a/full/reward_w050.jsonl": "b5bec0a6d3f01d69ebbf1993db798896d09ab0a1c0b9e630a06b1a6787759df2",
49
+ "task_a/full/reward_w070.jsonl": "2f9493d6e16f91d87902cab46b190e868f545564041d9c6caaddd6ac06af6be8",
50
+ "task_a/full/reward_w090.jsonl": "cec47857066d4a2a1ce1cfc3ab98d0f2d4607793162d2606c11148bffdf31920",
51
+ "task_a/full/reward_w100.jsonl": "63d6609a519edb2ce47ac627732090ac5eca9727cfc2deeb6f4755a218df41d1",
52
+ "task_a/reward_w000_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
53
+ "task_a/reward_w010_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
54
+ "task_a/reward_w030_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
55
+ "task_a/reward_w050_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
56
+ "task_a/reward_w070_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
57
+ "task_a/reward_w090_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
58
+ "task_a/reward_w100_maps.npz": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
59
+ "task_b/reward_w000.json": "e37ca6c194ef10a51098658d74459d2a8d926df7c1de0924308959826cd37720",
60
+ "task_b/reward_w010.json": "61e882d059647e50ea27e6505f89c1d7aa62dd8bcb68ba3a6986b84865356c61",
61
+ "task_b/reward_w030.json": "2734207bbdcaebaefe1be0995bfdf3ee26d44741eb64fd8aee2f9817788d6106",
62
+ "task_b/reward_w050.json": "807ebc46c178b83d63f02bcae174577da386bc4529e39a39386773c9d5a618a4",
63
+ "task_b/reward_w070.json": "97c4998acea85435d310dcf8d3b035d5413bdc6c7b4d28b5b89648e03a2398cd",
64
+ "task_b/reward_w090.json": "736ffc27d8451e7c60cec2d4a13acb753687a46c2788a7902bc895534ec32c6a",
65
+ "task_b/reward_w100.json": "32e65a7185896c1759aa3a8c27b36b1e9fb73367a5215ef7fd6d2e9f457ee6ed"
66
+ },
67
+ "no_test_model_selection": true,
68
+ "checkpoints": "existing validation-selected best checkpoints; no retraining"
69
+ }
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w000.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w010.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w030.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w050.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w070.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w090.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/choice/reward_w100.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w000.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w010.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w030.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w050.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w070.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w090.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/full/reward_w100.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_a/reward_w000_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w010_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w030_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w050_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w070_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w090_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_a/reward_w100_maps.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07
3
+ size 9601
review_artifacts/data/llama_test_20260925/task_b/reward_w000.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w010.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w030.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w050.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w070.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w090.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/data/llama_test_20260925/task_b/reward_w100.json ADDED
The diff for this file is too large to render. See raw diff
 
review_artifacts/outputs/llama_test_20260925/FROZEN_INPUTS.json ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "split": "test",
3
+ "adapters": {
4
+ "outputs/weight_llama_sft_20260916/reward_w000_seed100": {
5
+ "adapter_model.safetensors": "531a2bfe02e6fc5caeddcbaa8b257c7f5dd1c923fc41b665c162b0e7abfa38ee",
6
+ "adapter_config.json": "15ff5e03f5f8fe5801e89a34a58e99bc11604a03d80b39d92067c41f6720fcd1"
7
+ },
8
+ "outputs/weight_choice_only_sft_20260916/reward_w000_seed100": {
9
+ "adapter_model.safetensors": "f8c1ea126d2112dbdff2ba66323e3a23b1e4230f153d2d36335d06838227f3cf",
10
+ "adapter_config.json": "29a2c66b32e445531e7f1a2c602e66d1ccf434fbc0a95b424b59bfd326d0766a"
11
+ },
12
+ "outputs/spatial_llama_v1/reward_w000_full_seed100": {
13
+ "adapter_model.safetensors": "14a1918243367c732d04ca6679bee781bc53fda0ce498fa9981d6095038fffc6",
14
+ "adapter_config.json": "b1f9a4127c5f6b6992746076af6e8e54d1f08d0fc4ae1f6542a7ca107c310bc0"
15
+ },
16
+ "outputs/spatial_llama_v1/reward_w000_choice_only_seed100": {
17
+ "adapter_model.safetensors": "2a0143adb31d53d0d77711df8de62bb1860876cdbc5dcbe0a8fe7ff1d77dc3c9",
18
+ "adapter_config.json": "855a988b4408d435fa5c3f83cb0f650c565d7d4400fd4c4c57f889cdc0f181ca"
19
+ },
20
+ "outputs/weight_llama_sft_20260916/reward_w010_seed100": {
21
+ "adapter_model.safetensors": "1e7e2d11ede57f10200ef603792923451e78d057b41919c89388f30b4972bbe7",
22
+ "adapter_config.json": "3d686a02f29619f4a91d0b89cdf6dfd01cd2d5b4b3baae2032183075e13ea9cb"
23
+ },
24
+ "outputs/weight_choice_only_sft_20260916/reward_w010_seed100": {
25
+ "adapter_model.safetensors": "ecb36c6fd92eb0c35181be921f70f5c8e3d6f86d0eb9c86164a771e32b1d02e3",
26
+ "adapter_config.json": "cb45744e264469c3b8c67f993be0f81abaf4142727a3a9dacec84b9af776c251"
27
+ },
28
+ "outputs/spatial_llama_v1/reward_w010_full_seed100": {
29
+ "adapter_model.safetensors": "7c6adeee09b48c0ae7ed1d0b25f181a9d144623ba5a6eff2721a94c775d48a6a",
30
+ "adapter_config.json": "736344002b90dad1e8f33c93e2b8596641d0c4a075de28e1c43f110137932c50"
31
+ },
32
+ "outputs/spatial_llama_v1/reward_w010_choice_only_seed100": {
33
+ "adapter_model.safetensors": "9f4809ac15d83b8f39ec7a3a7069bce6c14e48d2776ffd503bdf03cc1c0c8a13",
34
+ "adapter_config.json": "85cec8494d3de93f81303a01738b237a60e6934aeefbc55eb3c18d4d3ce82fa7"
35
+ },
36
+ "outputs/weight_llama_sft_20260916/reward_w030_seed100": {
37
+ "adapter_model.safetensors": "8c37c5644750c27ea1e96400100434bcc565096160ccf76fd0a25462c7dc239d",
38
+ "adapter_config.json": "597993278812142d10fde6a51f47f89e2c9e7299391fa70736d9f2818667534c"
39
+ },
40
+ "outputs/weight_choice_only_sft_20260916/reward_w030_seed100": {
41
+ "adapter_model.safetensors": "2dcbc41934e97f4268817670fe8118c49ce16ecc56ce7f826041bfef2e796297",
42
+ "adapter_config.json": "7a638d74eafeed06bbe07783c71c3c6cf7720e74c33cdff3fb043defc72afaa7"
43
+ },
44
+ "outputs/spatial_llama_v1/reward_w030_full_seed100": {
45
+ "adapter_model.safetensors": "f0ade20974d2c2cf9041247db097e693f93d8876ab1684f34fa72e0ddb493a0f",
46
+ "adapter_config.json": "8071aa7a4ab9f575c3008d43d65b566ab43d732ca8769285ec9ae1e931c10c1e"
47
+ },
48
+ "outputs/spatial_llama_v1/reward_w030_choice_only_seed100": {
49
+ "adapter_model.safetensors": "adc73956e6435d7332f8b966d05124fccc204b43cda0465c933ef2eced462fed",
50
+ "adapter_config.json": "b380d5c50d0eeda3be6775391c5eb3a30192a36d6baabf25e7fb337a2efa8699"
51
+ },
52
+ "outputs/weight_llama_sft_20260916/reward_w050_seed100": {
53
+ "adapter_model.safetensors": "5db62ae701c9f005cfe34879c44d589ecaeb3949819b44191365dfc144281664",
54
+ "adapter_config.json": "37e9f76744549607322f52110835a9ba4d8049f44650ae8aeee62a5d40b7607a"
55
+ },
56
+ "outputs/weight_choice_only_sft_20260916/reward_w050_seed100": {
57
+ "adapter_model.safetensors": "af0e59a25f9d0f3b7fd4dd4f1b0f0cbac88a29c54d99186087783d0501f8dae8",
58
+ "adapter_config.json": "35b1d3d25e77d0d54d64741b54d353b72395847072a3ef0cbc7241f01aa27f23"
59
+ },
60
+ "outputs/spatial_llama_v1/reward_w050_full_seed100": {
61
+ "adapter_model.safetensors": "475b78ff13de9afeefdb46172dac652028b976993172b95754505333f4808850",
62
+ "adapter_config.json": "0315cb6ec9f36e0c6a5d455a8ee2eccc23e7703ae732004ea4c33401a01dc156"
63
+ },
64
+ "outputs/spatial_llama_v1/reward_w050_choice_only_seed100": {
65
+ "adapter_model.safetensors": "b441bf01552dbac169a6401af17438e672b821dd4d572aa29540b5dc180eaefe",
66
+ "adapter_config.json": "a658b4b64a6e5d1eeb83b2d093234abcd9933755c6515d0b39e19e62e51e3486"
67
+ },
68
+ "outputs/weight_llama_sft_20260916/reward_w070_seed100": {
69
+ "adapter_model.safetensors": "8432907916ea262e2270929e27333954b23a99eddbf439de698557b5751ba3eb",
70
+ "adapter_config.json": "c89444c089b7e30c056b819ab965176e5761fa9819a570afca2da8f1ba6e4992"
71
+ },
72
+ "outputs/weight_choice_only_sft_20260916/reward_w070_seed100": {
73
+ "adapter_model.safetensors": "966b654adea1ac47d0149cc3bf160a20b04363f8e2cf0015e073a838a04c1198",
74
+ "adapter_config.json": "f707f108f19438ac234ed10ae257d674e26ef3327063b4a5dcbec7118d6552e5"
75
+ },
76
+ "outputs/spatial_llama_v1/reward_w070_full_seed100": {
77
+ "adapter_model.safetensors": "38355837cb1c4e343089f5c7f0bff58b0b908584e5bc83ab80a6dd7ce8490a6f",
78
+ "adapter_config.json": "2cc08832838ebf57b8dd44f4c344e117db84fdcac1ec95431dc806498c593c32"
79
+ },
80
+ "outputs/spatial_llama_v1/reward_w070_choice_only_seed100": {
81
+ "adapter_model.safetensors": "bbf7a74af812ee3daf97b741947c66afc444956764ba8ebf41176c00da9f0355",
82
+ "adapter_config.json": "0c6d08fa317871c05e939a85517a1702040ea39c2601e842815da0d40f49fe7a"
83
+ },
84
+ "outputs/weight_llama_sft_20260916/reward_w090_seed100": {
85
+ "adapter_model.safetensors": "74714048c2dc2feef680c8773132fd6d67248a25e8589abd7ca0f3ab37ba8201",
86
+ "adapter_config.json": "bdea86efcc68227b01b722bc4bf1af45b4df1e1f2453d294d1e338a7da137e81"
87
+ },
88
+ "outputs/weight_choice_only_sft_20260916/reward_w090_seed100": {
89
+ "adapter_model.safetensors": "eb35bcfd888bc36aa7f39138e8229dffe5a05bb173c87210598c44791f48c03e",
90
+ "adapter_config.json": "6dd90a8298d57a38d2cf213b2cf084c0ffe09f6609e2af78c831d2f98294b292"
91
+ },
92
+ "outputs/spatial_llama_v1/reward_w090_full_seed100": {
93
+ "adapter_model.safetensors": "f94333175aa80f80cee3527657e668210e8741a128e2a9ce98d5d2429c76337c",
94
+ "adapter_config.json": "51166be58a8b5e4f2ac56441c422a463db8f538997f968ee5785d3573a4064c8"
95
+ },
96
+ "outputs/spatial_llama_v1/reward_w090_choice_only_seed100": {
97
+ "adapter_model.safetensors": "893e6583924d2654453bfd0348c1a2533ef9f02e673bd85682ff228abdf54eb2",
98
+ "adapter_config.json": "2e6989f0356c2f90a7af1aa6a9db32eb85796a76ae0fabb8e4b5a2c46f2fad44"
99
+ },
100
+ "outputs/weight_llama_sft_20260916/reward_w100_seed100": {
101
+ "adapter_model.safetensors": "48f7e097045c511fc478b36f87c6bea8dfd28f8a4eeb1ff7d33c8f6b91adf1fb",
102
+ "adapter_config.json": "43c6bcf1ba128112e094bbd0d13c197609cabee5d934557914d6443955fe59bf"
103
+ },
104
+ "outputs/weight_choice_only_sft_20260916/reward_w100_seed100": {
105
+ "adapter_model.safetensors": "8bccbfa172bf5b3a31842cbbaa3638f5693a7ddb9d4043985534be14563de0e9",
106
+ "adapter_config.json": "7a5bf4c5513609ea813bd7ecd4f928ee4c19e158afbaf92f02236cae3b593842"
107
+ },
108
+ "outputs/spatial_llama_v1/reward_w100_full_seed100": {
109
+ "adapter_model.safetensors": "b5d8919307ef4d949043d8645377919d5cd487fa67f9ed48902bca20cde15dc5",
110
+ "adapter_config.json": "c7dc160b7f1807c9faada301d5e18309da6d2b6106d97d7d0e03b60fc1666201"
111
+ },
112
+ "outputs/spatial_llama_v1/reward_w100_choice_only_seed100": {
113
+ "adapter_model.safetensors": "5be542c2523330f983a3c507d1203be555ca5343dca6b1812a1b71fb0d35bf3b",
114
+ "adapter_config.json": "130b6bb83cae77a72c83067a2922a9db421a9459744b1b92dc8680b77bee529b"
115
+ }
116
+ },
117
+ "scripts": {
118
+ "scripts/eval_weight_llama_test_20260925.py": "c000c82882cff9ad82f0a4208c663d8459fab41df110013fd9b7defff08cfbcc",
119
+ "scripts/eval_spatial_llama_test_20260925.py": "ce2ba9235e90a5a36851117df48219a58a5b526d55b2f530fe424bffede11f4d",
120
+ "scripts/slurm/eval_llama_test_main_20260925.sh": "ad7e9771ac0f843a77dbe40ceceea8669e1031e61d3c176a37a136cd41c70e49",
121
+ "src/mechcal/training/spatial_transcripts.py": "bb2e1f2393353b1c9bb0fc02904fe76081f6dd41b589e44aa3433ed6ed4460d8",
122
+ "src/mechcal/analysis/evaluate_centaur_reward_placeholder.py": "e60d6c8efbb2e7f519313aaa71dc5b80f3953e067d778ccdbd02fb4ffccfece1"
123
+ },
124
+ "payload_manifest_sha256": "f27ace490a6b76a2d433eb908350b3e6ae4768be8f9f83e2ce12ee325450681d"
125
+ }
review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_17.out ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ START family=task_a mode=full weight=050 shard=2/5
2
+ SAME_SHAPE_CAUSAL_AUDIT 0.0
3
+ PARTICIPANT_COMPLETE 050 full 2 seconds 60.6
4
+ PARTICIPANT_COMPLETE 050 full 7 seconds 124.8
5
+ PARTICIPANT_COMPLETE 050 full 12 seconds 192.9
6
+ PARTICIPANT_COMPLETE 050 full 17 seconds 262.4
7
+ PARTICIPANT_COMPLETE 050 full 22 seconds 332.5
8
+ PARTICIPANT_COMPLETE 050 full 27 seconds 402.2
9
+ PARTICIPANT_COMPLETE 050 full 32 seconds 472.4
10
+ PARTICIPANT_COMPLETE 050 full 37 seconds 542.5
11
+ PARTICIPANT_COMPLETE 050 full 42 seconds 612.1
12
+ PARTICIPANT_COMPLETE 050 full 47 seconds 681.2
13
+ PARTICIPANT_COMPLETE 050 full 52 seconds 750.4
14
+ PARTICIPANT_COMPLETE 050 full 57 seconds 819.8
15
+ PARTICIPANT_COMPLETE 050 full 62 seconds 889.4
16
+ PARTICIPANT_COMPLETE 050 full 67 seconds 959.3
17
+ PARTICIPANT_COMPLETE 050 full 72 seconds 1029.4
18
+ PARTICIPANT_COMPLETE 050 full 77 seconds 1099.6
19
+ PARTICIPANT_COMPLETE 050 full 82 seconds 1169.8
20
+ PARTICIPANT_COMPLETE 050 full 87 seconds 1240.1
21
+ PARTICIPANT_COMPLETE 050 full 92 seconds 1310.3
22
+ PARTICIPANT_COMPLETE 050 full 97 seconds 1380.3
23
+ PARTICIPANT_COMPLETE 050 full 102 seconds 1449.9
24
+ PARTICIPANT_COMPLETE 050 full 107 seconds 1518.7
25
+ PARTICIPANT_COMPLETE 050 full 112 seconds 1586.4
26
+ PARTICIPANT_COMPLETE 050 full 117 seconds 1650.4
27
+ PARTICIPANT_COMPLETE 050 full 122 seconds 1718.1
28
+ PARTICIPANT_COMPLETE 050 full 127 seconds 1787.2
29
+ PARTICIPANT_COMPLETE 050 full 132 seconds 1856.9
30
+ PARTICIPANT_COMPLETE 050 full 137 seconds 1926.7
31
+ PARTICIPANT_COMPLETE 050 full 142 seconds 1997.0
32
+ PARTICIPANT_COMPLETE 050 full 147 seconds 2067.3
33
+ PARTICIPANT_COMPLETE 050 full 152 seconds 2137.8
34
+ PARTICIPANT_COMPLETE 050 full 157 seconds 2208.3
35
+ PARTICIPANT_COMPLETE 050 full 162 seconds 2278.5
36
+ PARTICIPANT_COMPLETE 050 full 167 seconds 2348.2
37
+ PARTICIPANT_COMPLETE 050 full 172 seconds 2417.8
38
+ PARTICIPANT_COMPLETE 050 full 177 seconds 2487.5
39
+ PARTICIPANT_COMPLETE 050 full 182 seconds 2557.4
review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_28.out ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ START family=task_a mode=full weight=090 shard=3/5
2
+ SAME_SHAPE_CAUSAL_AUDIT 0.0
3
+ PARTICIPANT_COMPLETE 090 full 3 seconds 63.0
4
+ PARTICIPANT_COMPLETE 090 full 8 seconds 128.7
5
+ PARTICIPANT_COMPLETE 090 full 13 seconds 197.5
6
+ PARTICIPANT_COMPLETE 090 full 18 seconds 267.5
7
+ PARTICIPANT_COMPLETE 090 full 23 seconds 337.9
8
+ PARTICIPANT_COMPLETE 090 full 28 seconds 407.9
9
+ PARTICIPANT_COMPLETE 090 full 33 seconds 477.6
10
+ PARTICIPANT_COMPLETE 090 full 38 seconds 547.0
11
+ PARTICIPANT_COMPLETE 090 full 43 seconds 616.6
12
+ PARTICIPANT_COMPLETE 090 full 48 seconds 686.5
13
+ PARTICIPANT_COMPLETE 090 full 53 seconds 756.6
14
+ PARTICIPANT_COMPLETE 090 full 58 seconds 826.8
15
+ PARTICIPANT_COMPLETE 090 full 63 seconds 897.1
16
+ PARTICIPANT_COMPLETE 090 full 68 seconds 967.6
17
+ PARTICIPANT_COMPLETE 090 full 73 seconds 1038.0
18
+ PARTICIPANT_COMPLETE 090 full 78 seconds 1108.5
19
+ PARTICIPANT_COMPLETE 090 full 83 seconds 1178.9
20
+ PARTICIPANT_COMPLETE 090 full 88 seconds 1249.1
21
+ PARTICIPANT_COMPLETE 090 full 93 seconds 1319.0
22
+ PARTICIPANT_COMPLETE 090 full 98 seconds 1388.5
23
+ PARTICIPANT_COMPLETE 090 full 103 seconds 1458.5
24
+ PARTICIPANT_COMPLETE 090 full 108 seconds 1528.6
25
+ PARTICIPANT_COMPLETE 090 full 113 seconds 1598.8
26
+ PARTICIPANT_COMPLETE 090 full 118 seconds 1669.1
27
+ PARTICIPANT_COMPLETE 090 full 123 seconds 1739.1
28
+ PARTICIPANT_COMPLETE 090 full 128 seconds 1808.8
29
+ PARTICIPANT_COMPLETE 090 full 133 seconds 1878.1
30
+ PARTICIPANT_COMPLETE 090 full 138 seconds 1947.2
31
+ PARTICIPANT_COMPLETE 090 full 143 seconds 2016.4
32
+ PARTICIPANT_COMPLETE 090 full 148 seconds 2085.5
33
+ PARTICIPANT_COMPLETE 090 full 153 seconds 2154.7
34
+ PARTICIPANT_COMPLETE 090 full 158 seconds 2224.0
35
+ PARTICIPANT_COMPLETE 090 full 163 seconds 2292.8
36
+ PARTICIPANT_COMPLETE 090 full 168 seconds 2361.1
37
+ PARTICIPANT_COMPLETE 090 full 173 seconds 2428.9
38
+ PARTICIPANT_COMPLETE 090 full 178 seconds 2496.4
39
+ PARTICIPANT_COMPLETE 090 full 183 seconds 2563.8
review_artifacts/outputs/llama_test_20260925/resume_audit_logs/test-A-full_13587155_4.out ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ START family=task_a mode=full weight=000 shard=4/5
2
+ SAME_SHAPE_CAUSAL_AUDIT 0.0
3
+ PARTICIPANT_COMPLETE 000 full 4 seconds 65.8
4
+ PARTICIPANT_COMPLETE 000 full 9 seconds 131.1
5
+ PARTICIPANT_COMPLETE 000 full 14 seconds 198.2
6
+ PARTICIPANT_COMPLETE 000 full 19 seconds 265.9
7
+ PARTICIPANT_COMPLETE 000 full 24 seconds 333.9
8
+ PARTICIPANT_COMPLETE 000 full 29 seconds 401.4
9
+ PARTICIPANT_COMPLETE 000 full 34 seconds 469.1
10
+ PARTICIPANT_COMPLETE 000 full 39 seconds 536.9
11
+ PARTICIPANT_COMPLETE 000 full 44 seconds 605.2
12
+ PARTICIPANT_COMPLETE 000 full 49 seconds 674.3
13
+ PARTICIPANT_COMPLETE 000 full 54 seconds 743.8
14
+ PARTICIPANT_COMPLETE 000 full 59 seconds 813.7
15
+ PARTICIPANT_COMPLETE 000 full 64 seconds 883.7
16
+ PARTICIPANT_COMPLETE 000 full 69 seconds 953.7
17
+ PARTICIPANT_COMPLETE 000 full 74 seconds 1023.9
18
+ PARTICIPANT_COMPLETE 000 full 79 seconds 1093.9
19
+ PARTICIPANT_COMPLETE 000 full 84 seconds 1164.1
20
+ PARTICIPANT_COMPLETE 000 full 89 seconds 1234.1
21
+ PARTICIPANT_COMPLETE 000 full 94 seconds 1303.7
22
+ PARTICIPANT_COMPLETE 000 full 99 seconds 1373.2
23
+ PARTICIPANT_COMPLETE 000 full 104 seconds 1442.2
24
+ PARTICIPANT_COMPLETE 000 full 109 seconds 1511.7
25
+ PARTICIPANT_COMPLETE 000 full 114 seconds 1581.6
26
+ PARTICIPANT_COMPLETE 000 full 119 seconds 1651.5
27
+ PARTICIPANT_COMPLETE 000 full 124 seconds 1721.7
28
+ PARTICIPANT_COMPLETE 000 full 129 seconds 1792.1
29
+ PARTICIPANT_COMPLETE 000 full 134 seconds 1862.4
30
+ PARTICIPANT_COMPLETE 000 full 139 seconds 1932.8
31
+ PARTICIPANT_COMPLETE 000 full 144 seconds 2003.2
32
+ PARTICIPANT_COMPLETE 000 full 149 seconds 2073.6
33
+ PARTICIPANT_COMPLETE 000 full 154 seconds 2143.6
34
+ PARTICIPANT_COMPLETE 000 full 159 seconds 2213.1
35
+ PARTICIPANT_COMPLETE 000 full 164 seconds 2282.4
review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/COMPLETE_shard0.json ADDED
@@ -0,0 +1,269 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "mode": "choice",
3
+ "weight": "000",
4
+ "split": "test",
5
+ "smoke": false,
6
+ "operations": null,
7
+ "indices": [
8
+ 0,
9
+ 1,
10
+ 2,
11
+ 3,
12
+ 4,
13
+ 5,
14
+ 6,
15
+ 7,
16
+ 8,
17
+ 9,
18
+ 10,
19
+ 11,
20
+ 12,
21
+ 13,
22
+ 14,
23
+ 15,
24
+ 16,
25
+ 17,
26
+ 18,
27
+ 19,
28
+ 20,
29
+ 21,
30
+ 22,
31
+ 23,
32
+ 24,
33
+ 25,
34
+ 26,
35
+ 27,
36
+ 28,
37
+ 29,
38
+ 30,
39
+ 31,
40
+ 32,
41
+ 33,
42
+ 34,
43
+ 35,
44
+ 36,
45
+ 37,
46
+ 38,
47
+ 39,
48
+ 40,
49
+ 41,
50
+ 42,
51
+ 43,
52
+ 44,
53
+ 45,
54
+ 46,
55
+ 47,
56
+ 48,
57
+ 49,
58
+ 50,
59
+ 51,
60
+ 52,
61
+ 53,
62
+ 54,
63
+ 55,
64
+ 56,
65
+ 57,
66
+ 58,
67
+ 59,
68
+ 60,
69
+ 61,
70
+ 62,
71
+ 63,
72
+ 64,
73
+ 65,
74
+ 66,
75
+ 67,
76
+ 68,
77
+ 69,
78
+ 70,
79
+ 71,
80
+ 72,
81
+ 73,
82
+ 74,
83
+ 75,
84
+ 76,
85
+ 77,
86
+ 78,
87
+ 79,
88
+ 80,
89
+ 81,
90
+ 82,
91
+ 83,
92
+ 84,
93
+ 85,
94
+ 86,
95
+ 87,
96
+ 88,
97
+ 89,
98
+ 90,
99
+ 91,
100
+ 92,
101
+ 93,
102
+ 94,
103
+ 95,
104
+ 96,
105
+ 97,
106
+ 98,
107
+ 99,
108
+ 100,
109
+ 101,
110
+ 102,
111
+ 103,
112
+ 104,
113
+ 105,
114
+ 106,
115
+ 107,
116
+ 108,
117
+ 109,
118
+ 110,
119
+ 111,
120
+ 112,
121
+ 113,
122
+ 114,
123
+ 115,
124
+ 116,
125
+ 117,
126
+ 118,
127
+ 119,
128
+ 120,
129
+ 121,
130
+ 122,
131
+ 123,
132
+ 124,
133
+ 125,
134
+ 126,
135
+ 127,
136
+ 128,
137
+ 129,
138
+ 130,
139
+ 131,
140
+ 132,
141
+ 133,
142
+ 134,
143
+ 135,
144
+ 136,
145
+ 137,
146
+ 138,
147
+ 139,
148
+ 140,
149
+ 141,
150
+ 142,
151
+ 143,
152
+ 144,
153
+ 145,
154
+ 146,
155
+ 147,
156
+ 148,
157
+ 149,
158
+ 150,
159
+ 151,
160
+ 152,
161
+ 153,
162
+ 154,
163
+ 155,
164
+ 156,
165
+ 157,
166
+ 158,
167
+ 159,
168
+ 160,
169
+ 161,
170
+ 162,
171
+ 163,
172
+ 164,
173
+ 165,
174
+ 166,
175
+ 167,
176
+ 168,
177
+ 169,
178
+ 170,
179
+ 171,
180
+ 172,
181
+ 173,
182
+ 174,
183
+ 175,
184
+ 176,
185
+ 177,
186
+ 178,
187
+ 179,
188
+ 180,
189
+ 181,
190
+ 182,
191
+ 183,
192
+ 184,
193
+ 185,
194
+ 186,
195
+ 187,
196
+ 188,
197
+ 189,
198
+ 190,
199
+ 191,
200
+ 192,
201
+ 193,
202
+ 194,
203
+ 195,
204
+ 196,
205
+ 197,
206
+ 198,
207
+ 199,
208
+ 200,
209
+ 201,
210
+ 202,
211
+ 203,
212
+ 204,
213
+ 205,
214
+ 206,
215
+ 207,
216
+ 208,
217
+ 209,
218
+ 210,
219
+ 211,
220
+ 212,
221
+ 213,
222
+ 214,
223
+ 215,
224
+ 216,
225
+ 217,
226
+ 218,
227
+ 219,
228
+ 220,
229
+ 221,
230
+ 222,
231
+ 223,
232
+ 224,
233
+ 225,
234
+ 226,
235
+ 227,
236
+ 228,
237
+ 229,
238
+ 230,
239
+ 231,
240
+ 232,
241
+ 233,
242
+ 234,
243
+ 235,
244
+ 236,
245
+ 237,
246
+ 238,
247
+ 239,
248
+ 240,
249
+ 241,
250
+ 242,
251
+ 243,
252
+ 244,
253
+ 245,
254
+ 246,
255
+ 247,
256
+ 248,
257
+ 249
258
+ ],
259
+ "shard": 0,
260
+ "shards": 1,
261
+ "inference": "uncached_original_path",
262
+ "batch_size": 1,
263
+ "causal_logp_max_error": 0.0,
264
+ "source_sha256": "4402e83dec482c2edbc8c9693615bc7b9fe186781688967fac494608a2e7f0db",
265
+ "maps_sha256": "790e7bade985ec1f7e0c8d894ae678514b2507d2ad1b07e3bb007acfad15ec07",
266
+ "adapter": "outputs/weight_choice_only_sft_20260916/reward_w000_seed100",
267
+ "probability": "ABCD log probabilities under full vocabulary; normalize only in explicit conditional analysis",
268
+ "script_sha256": "c000c82882cff9ad82f0a4208c663d8459fab41df110013fd9b7defff08cfbcc"
269
+ }
review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_000.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af31f11ffead93794dae822fe95f989e3288a85a19f9a65632f7f28c0ec8ba85
3
+ size 3441
review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_001.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96ccf387d96f89494c8e89a62fee465df46f9693d53194c450d475ecc3601cf5
3
+ size 3407
review_artifacts/outputs/llama_test_20260925/task_a/production/reward_w000/choice/participant_002.npz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:49360d64a0d3a558234f716054aba84dcc2fe13e557ec7233d5986b77cea9c0f
3
+ size 3524