Datasets:

DONG19
/

diagnosis-advantage-grpo

prompt listlengths 1 1	data_source large_stringclasses 1 value	reward_model dict	extra_info dict	uid large_stringlengths 9 112
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["0\n","0\n","0\n0\n0\n0\n0\n","0\n","999\n","994\n24\n1137\n867\n147(...TRUNCATED)	{"problem_id":"1240_B. Sequence Sorting","variant":"A","time_limit":2.0,"test_inputs":["1\n1\n1\n","(...TRUNCATED)	1240_B. Sequence Sorting
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["0\n","0\n","0\n0\n0\n0\n0\n","0\n","999\n","994\n24\n1137\n867\n147(...TRUNCATED)	{"problem_id":"1240_B. Sequence Sorting","variant":"B","time_limit":2.0,"test_inputs":["1\n1\n1\n","(...TRUNCATED)	1240_B. Sequence Sorting
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["0 0 1\n","0 0 1\n","0 0 2\n","0 0 3\n","465782 191000430048 1910008(...TRUNCATED)	{"problem_id":"818_A. Diplomas and Certificates","variant":"A","time_limit":1.0,"test_inputs":["1 1\(...TRUNCATED)	818_A. Diplomas and Certificates
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["0 0 1\n","0 0 1\n","0 0 2\n","0 0 3\n","465782 191000430048 1910008(...TRUNCATED)	{"problem_id":"818_A. Diplomas and Certificates","variant":"B","time_limit":1.0,"test_inputs":["1 1\(...TRUNCATED)	818_A. Diplomas and Certificates
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["1\n","1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n","1\n1\n1\n1\n1\n1\n1\n1\n1\n1(...TRUNCATED)	{"problem_id":"975_C. Valhalla Siege","variant":"A","time_limit":2.0,"test_inputs":["1 1\n699817411\(...TRUNCATED)	975_C. Valhalla Siege
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["1\n","1\n1\n1\n1\n1\n1\n1\n1\n1\n1\n","1\n1\n1\n1\n1\n1\n1\n1\n1\n1(...TRUNCATED)	{"problem_id":"975_C. Valhalla Siege","variant":"B","time_limit":2.0,"test_inputs":["1 1\n699817411\(...TRUNCATED)	975_C. Valhalla Siege
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["680\n","19340\n","887208598\n","570417588\n","739547285\n","2463057(...TRUNCATED)	{"problem_id":"584_B. Kolya and Tanya ","variant":"A","time_limit":1.0,"test_inputs":["2\n","3\n","9(...TRUNCATED)	584_B. Kolya and Tanya
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["680\n","19340\n","887208598\n","570417588\n","739547285\n","2463057(...TRUNCATED)	{"problem_id":"584_B. Kolya and Tanya ","variant":"B","time_limit":1.0,"test_inputs":["2\n","3\n","9(...TRUNCATED)	584_B. Kolya and Tanya
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["NO\n","NO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\n","NO\nNO\nYES\nNO\n(...TRUNCATED)	{"problem_id":"1509_B. TMT Document","variant":"A","time_limit":1.0,"test_inputs":["1\n3\nMMM\n","10(...TRUNCATED)	1509_B. TMT Document
[{"role":"user","content":"You are given a competitive programming problem and a Python 3 solution t(...TRUNCATED)	codecontests-o	{"style":"rule","ground_truth":["NO\n","NO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\nNO\n","NO\nNO\nYES\nNO\n(...TRUNCATED)	{"problem_id":"1509_B. TMT Document","variant":"B","time_limit":1.0,"test_inputs":["1\n3\nMMM\n","10(...TRUNCATED)	1509_B. TMT Document

End of preview. Expand in Data Studio

Diagnosis-Advantage RLVR dataset

Training/eval data for the Diagnosis-Advantage RLVR recipe (verl GRPO): given a correct-but-slow Python solution to a competitive-programming problem, the model must produce a faster, still-correct optimization. Each problem appears as two prompt variants sharing one uid (route-B paired rollouts):

variant A — diagnose then optimize (emit a <diagnosis>…</diagnosis> block, then optimized code)
variant B — optimize directly (no diagnosis) — the counterfactual baseline

split	rows	problems (×2 variants)
train	15,686	7,843
validation	184	92
test	240	120

Columns

prompt — chat-format [{role: user, content}]; the content inlines the slow seed solution.
data_source — "codecontests-o".
reward_model — {style: "rule", ground_truth: [expected stdout per test]}.
extra_info — {problem_id, variant ("A"/"B"), time_limit, test_inputs (stdin per test), slow_code (the seed), n_tests}.
uid — problem_id (shared by a problem's A and B rows; the GRPO group key).

To get the diagnosis vs no-diagnosis test split, filter extra_info.variant == "A" / "B".

Provenance & notes

Derived from caijanfeng/CodeContests-O (test cases) joined with deepmind/code_contests Python-3 seed solutions (the slow seed). Seeds capped at 6000 chars.
Individual test stdin/stdout are capped at 128 KB (drops pathological multi-MB corner cases that blow up RAM during sandboxed execution); ~88% of tests retained.
Scoring is done at train/eval time via SandboxFusion (correctness + process_time_ns CPU-time for the speedup signal); this dataset ships only the prompts + tests.

Downloads last month: 144