JLiangHe commited on
Commit
e56ddda
·
verified ·
1 Parent(s): 95b8779

Publish GPT-OSS-120B High non-agentic results and full trajectories

Browse files

Same benchmark export layout as Qwen/SFT. Research remains provisional (1799/1800 scored); SciCode complete. Agentic benchmarks not started. No reruns or score changes.

.gitattributes CHANGED
@@ -114,3 +114,8 @@ Qwen3.5-35B-A3B/frontierscience-research/trajectories.jsonl filter=lfs diff=lfs
114
  Qwen3.5-35B-A3B/scicode-verified/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
115
  Qwen3.5-35B-A3B/scicoding-agent/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
116
  Qwen3.5-35B-A3B/terminal-bench-2.1/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
114
  Qwen3.5-35B-A3B/scicode-verified/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
115
  Qwen3.5-35B-A3B/scicoding-agent/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
116
  Qwen3.5-35B-A3B/terminal-bench-2.1/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
117
+ gpt-oss-120b/arxivmath/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
118
+ gpt-oss-120b/frontierscience-olympiad/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
119
+ gpt-oss-120b/frontierscience-olympiad/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
120
+ gpt-oss-120b/frontierscience-research/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
121
+ gpt-oss-120b/frontierscience-research/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -26,3 +26,8 @@ Results are organized by model and benchmark. See each model README for protocol
26
  ## Qwen3.5-35B-A3B agentic suite completed (2026-10-29)
27
 
28
  [Full Qwen suite and trajectories](Qwen3.5-35B-A3B/README.md) now includes SciCoding-Agent 36/100 (36.00% avg@2) and TB 2.1 61/178 (34.27% avg@2). TB includes two explicit user-adjudicated zeros; original null native verdicts and timeout records are archived. Prior non-agentic results are unchanged.
 
 
 
 
 
 
26
  ## Qwen3.5-35B-A3B agentic suite completed (2026-10-29)
27
 
28
  [Full Qwen suite and trajectories](Qwen3.5-35B-A3B/README.md) now includes SciCoding-Agent 36/100 (36.00% avg@2) and TB 2.1 61/178 (34.27% avg@2). TB includes two explicit user-adjudicated zeros; original null native verdicts and timeout records are archived. Prior non-agentic results are unchanged.
29
+
30
+
31
+ ## GPT-OSS-120B High (2026-10-30)
32
+
33
+ [Results and full available trajectories](gpt-oss-120b/README.md): FS Olympiad 63.50%, provisional Research rubric 41.04% (1799/1800 scored), arXiv Math 31.55%, SciCode main problems 59.38%. One Research judgment was API-policy-rejected and remains unscored, not zero. TB 2.1 and SciCoding-Agent have not started.
gpt-oss-120b/README.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # GPT-OSS-120B — High reasoning
2
+
3
+ Non-agentic evaluation, published 2026-10-30. This is not a completed full-suite evaluation: TB 2.1 and SciCoding-Agent have not started and have no reported scores here.
4
+
5
+ | Benchmark | Result | Status |
6
+ |---|---|---|
7
+ | FrontierScience Olympiad avg@30 | 63.50% (1905/3000) | Complete |
8
+ | FrontierScience Research mean rubric | 41.04% | Provisional, 1799/1800 scored |
9
+ | Research trajectory score >=7/10 | 17.12% (308/1799) | Provisional |
10
+ | arXiv Math combined avg@4 | 31.55% (130/412) | Complete |
11
+ | SciCode-Verified main problems | 59.38% (38/64) | Complete |
12
+ | SciCode-Verified subproblems | 73.87% (212/287) | Complete |
13
+
14
+ arXiv releases: December 19/68, January 42/92, February 29/128, March 40/124. Combined accuracy is weighted by problems/attempts, not an unweighted mean of release accuracies.
15
+
16
+ Research uses all available scored trajectories for the table above; one judge API bio_policy rejection remains unscored, not zero. Question-weighted available-trial averages in benchmark_metrics are 41.05% rubric and 17.13% trajectory success. Neither success metric is the fraction of questions whose mean rubric exceeds 7. No missing judgment was fabricated or adjudicated.
17
+
18
+ Protocol: checkpoint revision b5c939de8f754692c1647ca79fbf85e8c1e70f8a; native MXFP4 weights, high reasoning effort, temperature 1.0, top-p 1.0, 98,304 output-token cap. Serving: 32 H100 GPUs, TP2, 16 engines, 32 concurrent reasoning requests per engine. FS uses GPT-5.6 Luna, arXiv Math deterministic grading, and SciCode native executable verifiers. Exact frozen configs and source hashes are in provenance.json.
19
+
20
+ Each benchmark follows the existing repository format: trajectories.jsonl, judge_trajectories.jsonl, summary.json, and complete source-event or native-verifier archives. Full available reasoning and final responses are preserved; credential-shaped secrets are redacted. No inference, grading, or score changes were performed for publication. Cap hits: Olympiad 0/3000; scored Research 0/1799; arXiv 3/412; SciCode 0/287.
gpt-oss-120b/arxivmath/README.md ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ # GPT-OSS-120B — arxivmath
2
+
3
+ Full model responses and reasoning are in trajectories.jsonl; aligned judging or verifier records are in judge_trajectories.jsonl. Summary and archives preserve original scores and provenance. No generation or scoring was rerun.
gpt-oss-120b/arxivmath/events.jsonl.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42e48ad68afdd0059f28cbde22cbde7fc57f8e76a58cf5b6ffc8db7a546e9ba4
3
+ size 30251627
gpt-oss-120b/arxivmath/judge_trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
gpt-oss-120b/arxivmath/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"benchmark": "arxivmath", "releases": [{"total_records": 68, "completed": 68, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.27941176470588236, "accuracy": 0.27941176470588236, "mean_latency_seconds": 548.7706443442576, "finish_reason_counts": {"length": 1, "stop": 67}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 17, "n_trajectories_scored": 68, "reported_accuracy": 0.27941176470588236, "competition_metrics": {"arxiv/december": {"n_problems_scored": 17, "n_trajectories_scored": 68, "reported_accuracy": 0.27941176470588236}}}, "benchmark": "matharena-arxivmath-december", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "5efc5255c3eacd3bfc37a5cdfd1b13bb0e5865a4b03654f20c458e2dafaac5bc", "scoring_hash": "a3e2b1125bcb225515024853daf2e70da3b83f9b59856ada480339b2eb926661", "selected": 68, "incomplete": 0, "generated_now": 68, "scored_now": 68, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 68, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:49:07.475469+00:00"}, {"total_records": 128, "completed": 128, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.2265625, "accuracy": 0.2265625, "mean_latency_seconds": 483.31701234840966, "finish_reason_counts": {"stop": 128}, "length_limited_response_count": 0, "observed_max_completion_tokens": 92524, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 32, "n_trajectories_scored": 128, "reported_accuracy": 0.2265625, "competition_metrics": {"arxiv/february": {"n_problems_scored": 32, "n_trajectories_scored": 128, "reported_accuracy": 0.2265625}}}, "benchmark": "matharena-arxivmath-february", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "5ea28462ffe02741b342ec49d760377462d2a6394a3ed7302117d1456db78c49", "scoring_hash": "17f11f51e8e8063d80cf440ac0ca8e971e8fee7ed79f96a8d9867e872d10faa5", "selected": 128, "incomplete": 0, "generated_now": 128, "scored_now": 128, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 128, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:48:39.019286+00:00"}, {"total_records": 92, "completed": 92, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.45652173913043476, "accuracy": 0.45652173913043476, "mean_latency_seconds": 492.4658350913454, "finish_reason_counts": {"length": 1, "stop": 91}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 23, "n_trajectories_scored": 92, "reported_accuracy": 0.45652173913043476, "competition_metrics": {"arxiv/january": {"n_problems_scored": 23, "n_trajectories_scored": 92, "reported_accuracy": 0.45652173913043476}}}, "benchmark": "matharena-arxivmath-january", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "7c0996301787f473cd18987db61036d6b712bf52e04f16c0de9b02dc70509d8c", "scoring_hash": "e09d955ed6a3fccc283a4ea838cd7802e2927196a3fa75d3fb16070088ce2bb8", "selected": 92, "incomplete": 0, "generated_now": 92, "scored_now": 92, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 92, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:46:58.870497+00:00"}, {"total_records": 124, "completed": 124, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.3225806451612903, "accuracy": 0.3225806451612903, "mean_latency_seconds": 413.8366440370812, "finish_reason_counts": {"length": 1, "stop": 123}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 31, "n_trajectories_scored": 124, "reported_accuracy": 0.3225806451612903, "competition_metrics": {"arxiv/march": {"n_problems_scored": 31, "n_trajectories_scored": 124, "reported_accuracy": 0.3225806451612903}}}, "benchmark": "matharena-arxivmath-march", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "d896a1939ae59e96c6d2a794f9b4991af8bca2e5b3a3c8174346c908b4ca1879", "scoring_hash": "2dd97483dbb1256f5c3074e2aedd129a0b86523ee141a54afcf1e6d115ea8db8", "selected": 124, "incomplete": 0, "generated_now": 124, "scored_now": 124, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 124, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:49:35.432425+00:00"}], "total_records": 412, "completed": 412, "accuracy": 0.3155339805825243, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}
gpt-oss-120b/arxivmath/trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c38d9892f8d3241c8dd2d459559f831297ea244072f00daebb09c1969485ffbe
3
+ size 52359960
gpt-oss-120b/frontierscience-olympiad/README.md ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ # GPT-OSS-120B — frontierscience-olympiad
2
+
3
+ Full model responses and reasoning are in trajectories.jsonl; aligned judging or verifier records are in judge_trajectories.jsonl. Summary and archives preserve original scores and provenance. No generation or scoring was rerun.
gpt-oss-120b/frontierscience-olympiad/events.jsonl.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ceaeb636950a52501eb439f55d0e22c2b72ddc029ac83d2059656731567334d2
3
+ size 87961290
gpt-oss-120b/frontierscience-olympiad/judge_trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:367c43c754b1dc247969507425e103fc716d4f1fd2a47f04b3cc309504f91d9b
3
+ size 11017767
gpt-oss-120b/frontierscience-olympiad/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"total_records": 3000, "completed": 3000, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.635, "accuracy": 0.635, "mean_latency_seconds": 172.2245705660623, "finish_reason_counts": {"stop": 3000}, "length_limited_response_count": 0, "observed_max_completion_tokens": 88869, "judge_usage": {"judged_record_count": 3000, "request_count": 3001, "retry_request_count": 1, "prompt_tokens": 2918032, "cached_prompt_tokens": 5160, "cache_write_prompt_tokens": 1682117, "completion_tokens": 852262, "reasoning_tokens": 608745, "total_tokens": 3770294}, "judge_api_usage": {"request_count": 3001, "prompt_tokens": 2918032, "cached_prompt_tokens": 5160, "cache_write_prompt_tokens": 1682117, "completion_tokens": 852262, "reasoning_tokens": 608745, "total_tokens": 3770294}, "benchmark_metrics": {"track": "olympiad", "n_problems_scored": 100, "n_trajectories_scored": 3000, "aggregation": "avg@30", "reported_accuracy": 0.635, "mean_rubric_score": null, "success_threshold": 0.5, "community_profile": null}, "benchmark": "frontierscience-olympiad", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "7fb559870bd997006d944fcb333d94642ed958267906654b7b9b14013e91c650", "scoring_hash": "c15a7c1751abfa79a462596bae6c2f2b2b33d2a1f26b30753d0cff221feee016", "selected": 3000, "incomplete": 0, "generated_now": 3000, "scored_now": 3000, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 3000, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T04:08:24.980052+00:00", "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}
gpt-oss-120b/frontierscience-olympiad/trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ad96dadc377bbf66bc378ce19b8785952ad9cde7e6ff753f0a54b3b4ef178c3
3
+ size 156271238
gpt-oss-120b/frontierscience-research/README.md ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ # GPT-OSS-120B — frontierscience-research
2
+
3
+ Full model responses and reasoning are in trajectories.jsonl; aligned judging or verifier records are in judge_trajectories.jsonl. Summary and archives preserve original scores and provenance. No generation or scoring was rerun.
4
+
5
+ Provisional: 1,799/1,800 scored. The API-policy-rejected judgment remains null, not zero.
gpt-oss-120b/frontierscience-research/events.jsonl.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c54fdf83fd2f6f61a2a3b7d17e0a3c416168db664138c39a056682e1eefa705f
3
+ size 77694189
gpt-oss-120b/frontierscience-research/judge_trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63bba62ad3c22672cc343a3c077490e8ae61c9ecada86dad827dc54ff1adc4b3
3
+ size 29069960
gpt-oss-120b/frontierscience-research/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"total_records": 1800, "completed": 1799, "query_errors": 0, "score_errors": 1, "generated_unscored": 0, "mean_score": 0.41037548638132293, "accuracy": 0.17120622568093385, "mean_latency_seconds": 205.21365755201853, "finish_reason_counts": {"stop": 1799}, "length_limited_response_count": 0, "observed_max_completion_tokens": 75536, "judge_usage": {"judged_record_count": 1799, "request_count": 1835, "retry_request_count": 36, "prompt_tokens": 13256308, "cached_prompt_tokens": 251895, "cache_write_prompt_tokens": 12992300, "completion_tokens": 3386921, "reasoning_tokens": 2074633, "total_tokens": 16643229}, "judge_api_usage": {"request_count": 1835, "prompt_tokens": 13256308, "cached_prompt_tokens": 251895, "cache_write_prompt_tokens": 12992300, "completion_tokens": 3386921, "reasoning_tokens": 2074633, "total_tokens": 16643229}, "benchmark_metrics": {"track": "research", "n_problems_scored": 60, "n_trajectories_scored": 1799, "aggregation": "average_trials", "reported_accuracy": 0.17130268199233717, "mean_rubric_score": 0.4105160344827586, "success_threshold": 0.7, "community_profile": null}, "benchmark": "frontierscience-research", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "8aeabb7df69c1d0e05e6691291e2354bd9eac4c81a8ef3d4eceb0d8e3cc65ca1", "scoring_hash": "ceea6075bb3e7655a14d6ad2f07750f7b1b588c24da569c70a81bc6665831750", "selected": 1800, "incomplete": 1, "generated_now": 1800, "scored_now": 1799, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 1800, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:55:06.048144+00:00", "provisional": true, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run", "scoring_note": "One generated trajectory was rejected by the judge API (bio_policy) and remains unscored, not zero. Trajectory-level means use 1799 scored responses; benchmark_metrics average available trials within each of 60 questions."}
gpt-oss-120b/frontierscience-research/trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbe5435927dd950d5c6bf25db424b9b66ab0fa83b488c7fc3db5b780eb23dab2
3
+ size 149044468
gpt-oss-120b/manifest.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"README.md": {"size": 2053, "sha256": "090bd405cec50c879df65762e5dcfcc3a13fdd9a2651deb5f9f93aa36d9be4ec"}, "arxivmath/README.md": {"size": 259, "sha256": "76987fa32370457a756c1b52c7467c12db5de6ce6930d427ff767dca58cd7fad"}, "arxivmath/events.jsonl.gz": {"size": 30251627, "sha256": "42e48ad68afdd0059f28cbde22cbde7fc57f8e76a58cf5b6ffc8db7a546e9ba4"}, "arxivmath/judge_trajectories.jsonl": {"size": 221577, "sha256": "560593e58c58cd3e9aebb683c154cea60a02d5cbc564662e6a269907f2304e2d"}, "arxivmath/summary.json": {"size": 6912, "sha256": "fb75c05964ec35d873c23cbb12c9e600e4fb796ed1c161cca9ab6ed458ad804f"}, "arxivmath/trajectories.jsonl": {"size": 52359960, "sha256": "c38d9892f8d3241c8dd2d459559f831297ea244072f00daebb09c1969485ffbe"}, "frontierscience-olympiad/README.md": {"size": 274, "sha256": "c79c9f840e0da523f233fedf27735b3ed85efcb39895216c0c2bc3070253c486"}, "frontierscience-olympiad/events.jsonl.gz": {"size": 87961290, "sha256": "ceaeb636950a52501eb439f55d0e22c2b72ddc029ac83d2059656731567334d2"}, "frontierscience-olympiad/judge_trajectories.jsonl": {"size": 11017767, "sha256": "367c43c754b1dc247969507425e103fc716d4f1fd2a47f04b3cc309504f91d9b"}, "frontierscience-olympiad/summary.json": {"size": 1760, "sha256": "9de439b022c40f42c732c7a90136301f20cc43487f30c1915e7c67459f152e6a"}, "frontierscience-olympiad/trajectories.jsonl": {"size": 156271238, "sha256": "5ad96dadc377bbf66bc378ce19b8785952ad9cde7e6ff753f0a54b3b4ef178c3"}, "frontierscience-research/README.md": {"size": 365, "sha256": "b6fbb52c9eb9cb3c135ba543e27b4dc0fe1b89f548c42e45b6734c34a3a9769e"}, "frontierscience-research/events.jsonl.gz": {"size": 77694189, "sha256": "c54fdf83fd2f6f61a2a3b7d17e0a3c416168db664138c39a056682e1eefa705f"}, "frontierscience-research/judge_trajectories.jsonl": {"size": 29069960, "sha256": "63bba62ad3c22672cc343a3c077490e8ae61c9ecada86dad827dc54ff1adc4b3"}, "frontierscience-research/summary.json": {"size": 2079, "sha256": "d29552c87b538056e748f9179ffff2a2ae90270d4f46a17a2a62b9bd58869ed1"}, "frontierscience-research/trajectories.jsonl": {"size": 149044468, "sha256": "dbe5435927dd950d5c6bf25db424b9b66ab0fa83b488c7fc3db5b780eb23dab2"}, "provenance.json": {"size": 237984, "sha256": "e1bebaed4730659bfdecd770876f5220c70f9ee2531dc71e0ad823882a5e3fc4"}, "scicode-verified/README.md": {"size": 266, "sha256": "acca9bea612922ce540929a331bd97bd0b04c20471ad83aee20f6baf76f45bbb"}, "scicode-verified/judge_trajectories.jsonl": {"size": 233574, "sha256": "1f198d055e8c7a0e40dbb530db03bc647c4658acfb7e45fc5adc9ea1447817a0"}, "scicode-verified/native_artifacts.tar.gz": {"size": 2844707, "sha256": "00aa23d20f464a4b1efcec8ebe591d681e7f33e665f2ae17c3cb4bbf38e3e88f"}, "scicode-verified/summary.json": {"size": 515, "sha256": "1e1b8c73c9d25c59e19c7a1cafe480371586df4e909933ecebde270dd35f3ea7"}, "scicode-verified/trajectories.jsonl": {"size": 9182302, "sha256": "1b9e8f711d8fcba068d0a89e7543593a02c239bf345d707ea9882e443c644627"}, "summary.json": {"size": 11524, "sha256": "17f0f4d0c17e38b7e1a867d3379d012d789534e2772ad35eda917afe4d740183"}}
gpt-oss-120b/provenance.json ADDED
The diff for this file is too large to render. See raw diff
 
gpt-oss-120b/scicode-verified/README.md ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ # GPT-OSS-120B — scicode-verified
2
+
3
+ Full model responses and reasoning are in trajectories.jsonl; aligned judging or verifier records are in judge_trajectories.jsonl. Summary and archives preserve original scores and provenance. No generation or scoring was rerun.
gpt-oss-120b/scicode-verified/judge_trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
gpt-oss-120b/scicode-verified/native_artifacts.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00aa23d20f464a4b1efcec8ebe591d681e7f33e665f2ae17c3cb4bbf38e3e88f
3
+ size 2844707
gpt-oss-120b/scicode-verified/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"updated_at": 1793332711.9944172, "expected_problems": 64, "expected_steps": 287, "generated_steps": 287, "scored_steps": 287, "passed_steps": 212, "completed_problems": 64, "passed_problems": 38, "subproblem_accuracy": 0.7386759581881533, "main_problem_accuracy": 0.59375, "complete": true, "cap_hits": 0, "mean_output_tokens": 5008.02787456446, "grading_infrastructure_errors": 0, "errors": [], "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}
gpt-oss-120b/scicode-verified/trajectories.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
gpt-oss-120b/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model": "openai/gpt-oss-120b", "reasoning_effort": "high", "full_suite_complete": false, "benchmarks": {"frontierscience-olympiad": {"total_records": 3000, "completed": 3000, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.635, "accuracy": 0.635, "mean_latency_seconds": 172.2245705660623, "finish_reason_counts": {"stop": 3000}, "length_limited_response_count": 0, "observed_max_completion_tokens": 88869, "judge_usage": {"judged_record_count": 3000, "request_count": 3001, "retry_request_count": 1, "prompt_tokens": 2918032, "cached_prompt_tokens": 5160, "cache_write_prompt_tokens": 1682117, "completion_tokens": 852262, "reasoning_tokens": 608745, "total_tokens": 3770294}, "judge_api_usage": {"request_count": 3001, "prompt_tokens": 2918032, "cached_prompt_tokens": 5160, "cache_write_prompt_tokens": 1682117, "completion_tokens": 852262, "reasoning_tokens": 608745, "total_tokens": 3770294}, "benchmark_metrics": {"track": "olympiad", "n_problems_scored": 100, "n_trajectories_scored": 3000, "aggregation": "avg@30", "reported_accuracy": 0.635, "mean_rubric_score": null, "success_threshold": 0.5, "community_profile": null}, "benchmark": "frontierscience-olympiad", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "7fb559870bd997006d944fcb333d94642ed958267906654b7b9b14013e91c650", "scoring_hash": "c15a7c1751abfa79a462596bae6c2f2b2b33d2a1f26b30753d0cff221feee016", "selected": 3000, "incomplete": 0, "generated_now": 3000, "scored_now": 3000, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 3000, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T04:08:24.980052+00:00", "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}, "frontierscience-research": {"total_records": 1800, "completed": 1799, "query_errors": 0, "score_errors": 1, "generated_unscored": 0, "mean_score": 0.41037548638132293, "accuracy": 0.17120622568093385, "mean_latency_seconds": 205.21365755201853, "finish_reason_counts": {"stop": 1799}, "length_limited_response_count": 0, "observed_max_completion_tokens": 75536, "judge_usage": {"judged_record_count": 1799, "request_count": 1835, "retry_request_count": 36, "prompt_tokens": 13256308, "cached_prompt_tokens": 251895, "cache_write_prompt_tokens": 12992300, "completion_tokens": 3386921, "reasoning_tokens": 2074633, "total_tokens": 16643229}, "judge_api_usage": {"request_count": 1835, "prompt_tokens": 13256308, "cached_prompt_tokens": 251895, "cache_write_prompt_tokens": 12992300, "completion_tokens": 3386921, "reasoning_tokens": 2074633, "total_tokens": 16643229}, "benchmark_metrics": {"track": "research", "n_problems_scored": 60, "n_trajectories_scored": 1799, "aggregation": "average_trials", "reported_accuracy": 0.17130268199233717, "mean_rubric_score": 0.4105160344827586, "success_threshold": 0.7, "community_profile": null}, "benchmark": "frontierscience-research", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "8aeabb7df69c1d0e05e6691291e2354bd9eac4c81a8ef3d4eceb0d8e3cc65ca1", "scoring_hash": "ceea6075bb3e7655a14d6ad2f07750f7b1b588c24da569c70a81bc6665831750", "selected": 1800, "incomplete": 1, "generated_now": 1800, "scored_now": 1799, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 1800, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:55:06.048144+00:00", "provisional": true, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run", "scoring_note": "One generated trajectory was rejected by the judge API (bio_policy) and remains unscored, not zero. Trajectory-level means use 1799 scored responses; benchmark_metrics average available trials within each of 60 questions."}, "arxivmath": {"benchmark": "arxivmath", "releases": [{"total_records": 68, "completed": 68, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.27941176470588236, "accuracy": 0.27941176470588236, "mean_latency_seconds": 548.7706443442576, "finish_reason_counts": {"length": 1, "stop": 67}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 17, "n_trajectories_scored": 68, "reported_accuracy": 0.27941176470588236, "competition_metrics": {"arxiv/december": {"n_problems_scored": 17, "n_trajectories_scored": 68, "reported_accuracy": 0.27941176470588236}}}, "benchmark": "matharena-arxivmath-december", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "5efc5255c3eacd3bfc37a5cdfd1b13bb0e5865a4b03654f20c458e2dafaac5bc", "scoring_hash": "a3e2b1125bcb225515024853daf2e70da3b83f9b59856ada480339b2eb926661", "selected": 68, "incomplete": 0, "generated_now": 68, "scored_now": 68, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 68, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:49:07.475469+00:00"}, {"total_records": 128, "completed": 128, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.2265625, "accuracy": 0.2265625, "mean_latency_seconds": 483.31701234840966, "finish_reason_counts": {"stop": 128}, "length_limited_response_count": 0, "observed_max_completion_tokens": 92524, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 32, "n_trajectories_scored": 128, "reported_accuracy": 0.2265625, "competition_metrics": {"arxiv/february": {"n_problems_scored": 32, "n_trajectories_scored": 128, "reported_accuracy": 0.2265625}}}, "benchmark": "matharena-arxivmath-february", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "5ea28462ffe02741b342ec49d760377462d2a6394a3ed7302117d1456db78c49", "scoring_hash": "17f11f51e8e8063d80cf440ac0ca8e971e8fee7ed79f96a8d9867e872d10faa5", "selected": 128, "incomplete": 0, "generated_now": 128, "scored_now": 128, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 128, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:48:39.019286+00:00"}, {"total_records": 92, "completed": 92, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.45652173913043476, "accuracy": 0.45652173913043476, "mean_latency_seconds": 492.4658350913454, "finish_reason_counts": {"length": 1, "stop": 91}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 23, "n_trajectories_scored": 92, "reported_accuracy": 0.45652173913043476, "competition_metrics": {"arxiv/january": {"n_problems_scored": 23, "n_trajectories_scored": 92, "reported_accuracy": 0.45652173913043476}}}, "benchmark": "matharena-arxivmath-january", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "7c0996301787f473cd18987db61036d6b712bf52e04f16c0de9b02dc70509d8c", "scoring_hash": "e09d955ed6a3fccc283a4ea838cd7802e2927196a3fa75d3fb16070088ce2bb8", "selected": 92, "incomplete": 0, "generated_now": 92, "scored_now": 92, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 92, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:46:58.870497+00:00"}, {"total_records": 124, "completed": 124, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.3225806451612903, "accuracy": 0.3225806451612903, "mean_latency_seconds": 413.8366440370812, "finish_reason_counts": {"length": 1, "stop": 123}, "length_limited_response_count": 1, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 0, "request_count": 0, "retry_request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "judge_api_usage": {"request_count": 0, "prompt_tokens": 0, "cached_prompt_tokens": 0, "cache_write_prompt_tokens": 0, "completion_tokens": 0, "reasoning_tokens": 0, "total_tokens": 0}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@4", "n_problems_scored": 31, "n_trajectories_scored": 124, "reported_accuracy": 0.3225806451612903, "competition_metrics": {"arxiv/march": {"n_problems_scored": 31, "n_trajectories_scored": 124, "reported_accuracy": 0.3225806451612903}}}, "benchmark": "matharena-arxivmath-march", "model": "openai/gpt-oss-120b@b5c939de8f754692c1647ca79fbf85e8c1e70f8a", "execution_mode": "non_agent", "generation_hash": "d896a1939ae59e96c6d2a794f9b4991af8bca2e5b3a3c8174346c908b4ca1879", "scoring_hash": "2dd97483dbb1256f5c3074e2aedd129a0b86523ee141a54afcf1e6d115ea8db8", "selected": 124, "incomplete": 0, "generated_now": 124, "scored_now": 124, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 124, "generation_concurrency": 512, "judge_concurrency": 256, "updated_at": "2026-10-30T03:49:35.432425+00:00"}], "total_records": 412, "completed": 412, "accuracy": 0.3155339805825243, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}, "scicode-verified": {"updated_at": 1793332711.9944172, "expected_problems": 64, "expected_steps": 287, "generated_steps": 287, "scored_steps": 287, "passed_steps": 212, "completed_problems": 64, "passed_problems": 38, "subproblem_accuracy": 0.7386759581881533, "main_problem_accuracy": 0.59375, "complete": true, "cap_hits": 0, "mean_output_tokens": 5008.02787456446, "grading_infrastructure_errors": 0, "errors": [], "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/gpt-oss-120b-full-suite-20261030/run"}}, "not_started": ["terminal-bench-2.1", "scicoding-agent"]}