Text Classification
PEFT
Safetensors
English
sieve
decision-model
calibration
lora
multiple-choice
qwen3.5
Eval Results (legacy)
Instructions to use sthanika-ai/Sieve-9B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use sthanika-ai/Sieve-9B with PEFT:
from peft import PeftModel from transformers import AutoModel base_model = AutoModel.from_pretrained("Qwen/Qwen3.5-9B") model = PeftModel.from_pretrained(base_model, "sthanika-ai/Sieve-9B") - Notebooks
- Google Colab
- Kaggle
Model card: coverage sentence; quoted task names in the metadata; leaderboard scores named as such
96cdfc2 verified Download provenance.json from sthanika-ai/Sieve-9B: direct link, hf CLI and curl.
- Browser
- Download file 3.84 kB
-
https://huggingface.co/sthanika-ai/Sieve-9B/resolve/main/provenance.json
- Command line
-
hf download hf://sthanika-ai/Sieve-9B/provenance.json
-
curl -L -o provenance.json https://huggingface.co/sthanika-ai/Sieve-9B/resolve/main/provenance.json
3.84 kB
| { | |
| "built": "2026-09-29T11:54:53", | |
| "base_model": "Qwen/Qwen3.5-9B", | |
| "base_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a", | |
| "files_sha256": { | |
| ".gitattributes": "34448b82c17d60fec9b65b1f093c115ddbaadc04beb1b0140b6bfed2e012a930", | |
| "LICENSE": "6f11f9076e09fc1c0ffd1aaa7295bedea4fff14d93a008e0e2b42488d2f1ab30", | |
| "README.md": "06f0160b98f5abbdc019aa00123b208330a699e656228686535dd4e0993cad97", | |
| "adapter_config.json": "260d5a5261989a804e328355981c5e7af2cc40279fe78f43905b4bcd8b286933", | |
| "adapter_model.safetensors": "f14e56ebf16e431be2d55b774dd0bbafe9b297274d821ffae50aebba6a6d7978", | |
| "head.pt": "c8a943e41f4dd40939fac36f943a97d6a1bc9c5adfbe2e72cc76d6f266a822c3", | |
| "result.json": "3151a08a4086d808498a4ce050a0a6b2080c7d55a043493fb23897af78afd988", | |
| "tokenizer.json": "5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42", | |
| "tokenizer_config.json": "316230d6a809701f4db5ea8f8fc862bc3a6f3229c937c174e674ff3ca0a64ac8", | |
| "train.log": "1ec42a059fa1283d869e7b436d32d9c958c7b3f64a1997a8b70145923e82d8f1", | |
| "training_config.json": "57320b4470cc2cdd5f33c3d1b07f625c7da3018f0ab319f8ec33e133259a2157", | |
| "training_metrics.json": "64a9e952350934ea6749c8d8af7421eac1cb96e217b77591cdaf968b1c2ffab9" | |
| }, | |
| "weights": "adapter_model.safetensors and adapter_config.json are the trained weight average, unchanged. head.pt has the same weights; only the calibration temperature was added.", | |
| "head_pt_before_temperature_was_stored_sha256": "03971bf5b568a47217edd39ffcc427db711e94a53fae1dddc4388675e6aedc53", | |
| "training_data": { | |
| "published": false, | |
| "decision_v7_train": { | |
| "file": "decision_v7_train.jsonl", | |
| "sha256": "7ed5254b5cb5291baefaceb09edf7e13110258211518c8038f4a12c11bd628ad", | |
| "records": 12576 | |
| }, | |
| "date_policy_unknowable": { | |
| "file": "date_policy_unknowable.jsonl", | |
| "sha256": "afd8502d162163605ac446439e32c7b9083302dd78a6bfdc5e30075619e98437", | |
| "records": 1425 | |
| }, | |
| "date_policy": { | |
| "file": "date_policy.jsonl", | |
| "sha256": "9d068b1e7d382992add46e93b22a9d539036f4db78b576ea07c0954534b3d401", | |
| "records": 900 | |
| }, | |
| "rule_structures": { | |
| "file": "rule_structures.jsonl", | |
| "sha256": "bb7e0314ce99138a4c435a1fe789f703d02fe40aeae14664824cb9e74136975a", | |
| "records": 1920 | |
| } | |
| }, | |
| "environment": { | |
| "torch": "2.8.0+cu128", | |
| "transformers": "5.17.0", | |
| "peft": "0.21.0", | |
| "flash-linear-attention": "0.5.2" | |
| }, | |
| "contamination_audit": "Every training file was checked against seven benchmark partitions (decision-v7 development, calibration and test; transfer-v4 development and test; transfer-v9 development; the held-out decision benchmark) by exact (state, question) pair, exact state and a sentence-order-insensitive state hash. decision_v7_train, date_policy and rule_structures: 0 overlaps. date_policy_unknowable (run A only): 4 near-duplicate states (3 in the sealed decision-v7 test, from the intact controls; 1 in transfer-v9), and its unknowable records share families with transfer-v9's unknowable items, which are therefore excluded from every transfer-v9 number.", | |
| "verification": "On 2026-09-29 these files, loaded with the published code (github.com/sthanika-ai/Sieve; adapter merged, bf16, one A100), were run on 364 Decision Index 0.2.1 requests sampled from all 91 tracks (1,098 questions, 45 requests with several questions). Every answer matched the recorded run, and the largest probability difference was 0.", | |
| "decision_index": { | |
| "edition": "0.2.1", | |
| "decision_index": 41.71, | |
| "raw_index": 56.47, | |
| "scored_requests": 150317, | |
| "answered": 150317, | |
| "kit_commit": "87d4650b42b377c0291a89c1f1a879f9b31082bf", | |
| "engine": "sieve.decision_index_engine:SieveEngine", | |
| "run_recorded_adapter_sha256": "f14e56ebf16e431be2d55b774dd0bbafe9b297274d821ffae50aebba6a6d7978", | |
| "run_recorded_head_sha256": "c8a943e41f4dd40939fac36f943a97d6a1bc9c5adfbe2e72cc76d6f266a822c3" | |
| } | |
| } |