Publish Agentic SFT temperature-0.7 agentic rerun
Browse files[["scicoding-agent", 100, 37, 0], ["terminal-bench-2.1", 178, 49, 0]]; Replace both agentic exports; 1 user-authorized zero adjudications with original null verdicts preserved. Reasoning results unchanged.
- sft-agentic-stage2/README.md +6 -6
- sft-agentic-stage2/manifest.json +1 -1
- sft-agentic-stage2/provenance.json +0 -0
- sft-agentic-stage2/scicoding-agent/README.md +2 -2
- sft-agentic-stage2/scicoding-agent/provenance.json +1 -1
- sft-agentic-stage2/scicoding-agent/summary.json +1 -1
- sft-agentic-stage2/summary.json +1 -1
- sft-agentic-stage2/terminal-bench-2.1/README.md +2 -2
- sft-agentic-stage2/terminal-bench-2.1/judge_trajectories.jsonl +0 -0
- sft-agentic-stage2/terminal-bench-2.1/provenance.json +1 -1
- sft-agentic-stage2/terminal-bench-2.1/scoring_adjudications.json +1 -0
- sft-agentic-stage2/terminal-bench-2.1/scoring_adjudications.tar.gz +3 -0
- sft-agentic-stage2/terminal-bench-2.1/summary.json +1 -1
- sft-agentic-stage2/terminal-bench-2.1/trajectories.jsonl +2 -2
sft-agentic-stage2/README.md
CHANGED
|
@@ -1,16 +1,16 @@
|
|
| 1 |
-
# Agentic SFT
|
| 2 |
|
| 3 |
-
## Current agentic evaluation (2026-11-
|
| 4 |
|
| 5 |
| Benchmark | Attempts | Passed | avg@2 | Unscored |
|
| 6 |
|---|---:|---:|---:|---:|
|
| 7 |
-
|
|
| 8 |
-
|
|
| 9 |
|
| 10 |
-
All
|
| 11 |
|
| 12 |
Terminus 2; temperature **0.7**, top-p **0.95**, top-k **20**; **16 H100 GPUs**, TP2, eight engines, two tasks per engine. Each benchmark contains full trajectories, verifier records, native artifacts, frozen configs and provenance. These exports replace the earlier agentic results; old exports remain recoverable in repository history.
|
| 13 |
|
| 14 |
## Unchanged reasoning and code evaluations
|
| 15 |
|
| 16 |
-
The current results and all files in [arxivmath](arxivmath/), [frontierscience-olympiad](frontierscience-olympiad/), [frontierscience-research](frontierscience-research/), [hle-verified](hle-verified/) and
|
|
|
|
| 1 |
+
# Agentic SFT
|
| 2 |
|
| 3 |
+
## Current agentic evaluation (2026-11-15)
|
| 4 |
|
| 5 |
| Benchmark | Attempts | Passed | avg@2 | Unscored |
|
| 6 |
|---|---:|---:|---:|---:|
|
| 7 |
+
| scicoding-agent | 100 | 37 | 37.00% | 0 |
|
| 8 |
+
| terminal-bench-2.1 | 178 | 49 | 27.53% | 0 |
|
| 9 |
|
| 10 |
+
All attempts are finished and scored. Missing native verdicts were explicitly assigned zero by the user; their original null verdicts and errors are preserved alongside the adjudications. Every planned attempt stays in the denominator. Native verifier success counts regardless of agent stop reason. Exact unscored attempt IDs and native errors are retained in the exported judge records.
|
| 11 |
|
| 12 |
Terminus 2; temperature **0.7**, top-p **0.95**, top-k **20**; **16 H100 GPUs**, TP2, eight engines, two tasks per engine. Each benchmark contains full trajectories, verifier records, native artifacts, frozen configs and provenance. These exports replace the earlier agentic results; old exports remain recoverable in repository history.
|
| 13 |
|
| 14 |
## Unchanged reasoning and code evaluations
|
| 15 |
|
| 16 |
+
The current results and all files in [arxivmath](arxivmath/), [frontierscience-olympiad](frontierscience-olympiad/), [frontierscience-research](frontierscience-research/), [hle-verified](hle-verified/) and any previously published code-evaluation folders are preserved. Their own frozen configurations apply, not the agentic sampling settings above. See `summary.json` for the complete current results.
|
sft-agentic-stage2/manifest.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"provenance.json": {"size": 428190, "sha256": "
|
|
|
|
| 1 |
+
{"provenance.json": {"size": 428190, "sha256": "598207fe2b4c6778e4cd50537c1771284328973d9c904732fe4010a8c0ae85b3"}, "summary.json": {"size": 12239, "sha256": "d55d92094adcab4858311b05c4658a66109e79c9d43c9a844365255657728ede"}, "README.md": {"size": 1407, "sha256": "aaf955cd086ac99635496896fe348a3a2402c2dfc9673fbadcb8a54098b4163d"}, "arxivmath/trajectories.jsonl": {"size": 359337367, "sha256": "78621a79dd483e4e9902ac13f87aed49bf897e240972841db41657ac424be0bc"}, "arxivmath/events.jsonl.gz": {"size": 227547208, "sha256": "51e923646de89919235080bbb70505738a2b271cf36233a777bd5d70babea182"}, "arxivmath/judge_trajectories.jsonl": {"size": 988260, "sha256": "39f7c0fe9416087530aaddae940295b59e2000fbb07e4c4c34d3f2de0245bf5e"}, "arxivmath/summary.json": {"size": 6214, "sha256": "96757d402960cf9cb5e53dd66d5912461aa83d5c9744dd4a092235d32bef4032"}, "scicode-verified/trajectories.jsonl": {"size": 17765829, "sha256": "9b7783fa9d68412eba0de2a4756df92a57cac5ec684f4600c72066207ac14c35"}, "scicode-verified/judge_trajectories.jsonl": {"size": 265790, "sha256": "c217059473dc469b072884eff7ef38df0c189b6dc71e2085889103761cd1cd8c"}, "scicode-verified/summary.json": {"size": 397, "sha256": "aa94cae0d1138750754ee264178c470005c6498144c508e902c3bca246e211e9"}, "scicode-verified/native_artifacts.tar.gz": {"size": 5416773, "sha256": "0d4161f5ae5c506a201738ee1714ef22748cd0e5d802d658f852ecb59b4396d6"}, "frontierscience-research/events.jsonl.gz": {"size": 138682005, "sha256": "7c3466823c6a06be0b1272cf34ca045067d7d9683246c82c23b0159ba7d93985"}, "frontierscience-research/judge_trajectories.jsonl": {"size": 27430579, "sha256": "0e9a9b0ae45e480a3b2d13f82e680f3fe86ee305005e16c5d2ffd6df5cf35614"}, "frontierscience-research/trajectories.jsonl": {"size": 241276630, "sha256": "764d808291ee04c16944e2305187e8505c580e430c636838db26efd21ca82c5e"}, "frontierscience-research/summary.json": {"size": 1684, "sha256": "46c13f2a1039aea38325614a117372178b3bbf1e2fabf9554d59180b12a3f067"}, "frontierscience-olympiad/summary.json": {"size": 1615, "sha256": "8b1e27cb156b89316f670d09e0a0ee5399ee9a2c8cfd91fe34eaacd3226224dd"}, "frontierscience-olympiad/judge_trajectories.jsonl": {"size": 10518578, "sha256": "55a5fccfc2437239980936f42d88346d93c366815b8d96693ea8e8751b3f8db1"}, "frontierscience-olympiad/events.jsonl.gz": {"size": 127421615, "sha256": "02a076bcffc8ab1632f80228ac546ef20e20e3169015d4e0fa18bbec91ae2879"}, "frontierscience-olympiad/trajectories.jsonl": {"size": 214973672, "sha256": "740cec9cdfafc1b7300bbbc6ed19203dff6d145a22a0cf8c73f6ace9550374b0"}, "arxivmath/README.md": {"size": 978, "sha256": "5355aba84c8fb94ca780bf4c05625534bee6242c5189206d72136626b9a42ba7"}, "arxivmath/configs.json": {"size": 15298, "sha256": "831d0dd998d8bc5c17382020a0eea12f9168869cc266a86fdc467ecb4d83d590"}, "arxivmath/provenance.json": {"size": 1285, "sha256": "98db9d4c3dfac615fe4f3913483a523a5009e5f1879c71389cb1520a7fb304aa"}, "scicoding-agent/README.md": {"size": 532, "sha256": "b6c3799bdeba5b09c0c0422c1a6a847516ade5e0df8d38179ceb85c46b71e57b"}, "scicoding-agent/configs.json": {"size": 215601, "sha256": "a618136ef82a408dbea96a30963379c8bdb6a15a2c7887b7146b3ee936015efb"}, "scicoding-agent/judge_trajectories.jsonl": {"size": 711576, "sha256": "dbe02aba1d5d1f148399085ce6626333f2ed4a4df6cfd5bc16bd44819b334185"}, "scicoding-agent/native_artifacts.tar.gz": {"size": 124167602, "sha256": "c5ee81819b0218164e2f9046909283d7d8a622a471c9732597f945292a322a38"}, "scicoding-agent/provenance.json": {"size": 368, "sha256": "dba4a47843ceba47add0c308ba798ef644e73e21ada11086a8cd58f86d186c69"}, "scicoding-agent/summary.json": {"size": 602, "sha256": "7885cce27e013260898b4cf520d2e84f60a3a9dce33765f4bbe2a1d0d4a63132"}, "scicoding-agent/trajectories.jsonl": {"size": 63926829, "sha256": "fafaf94b30e13949b25bf42d11347ce6888f15c7f5d54b8a1642660fedaf9543"}, "terminal-bench-2.1/README.md": {"size": 535, "sha256": "908052537ceebbf27cfcc4b2effcca0bdfe8d0d03b09b441406b19bf19baeba6"}, "terminal-bench-2.1/configs.json": {"size": 215601, "sha256": "a618136ef82a408dbea96a30963379c8bdb6a15a2c7887b7146b3ee936015efb"}, "terminal-bench-2.1/judge_trajectories.jsonl": {"size": 5527253, "sha256": "ebc2bab6e92c04f1263ccd96b4dd74ffadee2612db9f532b158bf928320e3294"}, "terminal-bench-2.1/native_artifacts.tar.gz": {"size": 285288063, "sha256": "1f50c8c66247fdd6ee833cbb4e2bf8a5b888ffab377eb5e6f5c152dd0d57de15"}, "terminal-bench-2.1/provenance.json": {"size": 367, "sha256": "d0c4b2db78da8518259c93d18ade1bcf84658e6b6c84416ed2764dcd3f5da96a"}, "terminal-bench-2.1/scoring_adjudications.json": {"size": 5226, "sha256": "9280c3343e8009be58a3f455512ec830e36e5adb7d13fcc45051347259b29eb1"}, "terminal-bench-2.1/scoring_adjudications.tar.gz": {"size": 33503, "sha256": "2e0a73d4edac8bbe364b81224c5719c6bd6d75d0a231f722c54a51602bb2e91c"}, "terminal-bench-2.1/summary.json": {"size": 649, "sha256": "b3e4826b2698096a34150e8fb60342cd9c6537a07afd01ca09993b546e5228a9"}, "terminal-bench-2.1/trajectories.jsonl": {"size": 87319910, "sha256": "1f1d2c140ad9de823bf3d76494fc0f6cfab9a5c0fce86e4240a5447bb00b2959"}}
|
sft-agentic-stage2/provenance.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
sft-agentic-stage2/scicoding-agent/README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
-
# scicoding-agent:
|
| 2 |
|
| 3 |
-
37/100 passed (37.00% avg@2); 0 unscored. All planned attempts remain in the denominator.
|
| 4 |
|
| 5 |
Terminus 2; temperature 0.7, top-p 0.95, top-k 20; H100, TP2, eight engines, concurrency two per engine. Frozen plans and native artifacts are included. Native verifier success counts even after abnormal agent stops.
|
|
|
|
| 1 |
+
# scicoding-agent: Agentic SFT rerun
|
| 2 |
|
| 3 |
+
37/100 passed (37.00% avg@2); 0 unscored. All planned attempts remain in the denominator. Originally missing verdicts are explicitly user-adjudicated zero. Where applicable, original null verdicts and errors are preserved in scoring_adjudications.json and its backup archive.
|
| 4 |
|
| 5 |
Terminus 2; temperature 0.7, top-p 0.95, top-k 20; H100, TP2, eight engines, concurrency two per engine. Frozen plans and native artifacts are included. Native verifier success counts even after abnormal agent stops.
|
sft-agentic-stage2/scicoding-agent/provenance.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "published_date": "2026-11-
|
|
|
|
| 1 |
+
{"source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "published_date": "2026-11-15", "replaces_revision": "f74262971417a10dc91b9084c0f6658ffcbf7d57", "exporter_sha256": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15", "generation_or_judgment_rerun": false, "manual_zero_adjudications": false}
|
sft-agentic-stage2/scicoding-agent/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 37, "avg_at_2": 0.37, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 13, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2}
|
|
|
|
| 1 |
+
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 37, "avg_at_2": 0.37, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 13, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "adjudicated_zero": 0, "native_scored": 100}
|
sft-agentic-stage2/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"checkpoint_step": 1144, "full_suite_complete": false, "benchmarks": {"arxivmath": {"benchmark": "arxivmath", "releases": [{"total_records": 340, "completed": 340, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.27647058823529413, "accuracy": 0.27647058823529413, "mean_latency_seconds": 1102.0164873120102, "finish_reason_counts": {"length": 50, "stop": 290}, "length_limited_response_count": 50, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 17, "n_trajectories_scored": 340, "reported_accuracy": 0.27647058823529413, "competition_metrics": {"arxiv/december": {"n_problems_scored": 17, "n_trajectories_scored": 340, "reported_accuracy": 0.27647058823529413}}}, "benchmark": "matharena-arxivmath-december", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "4873a8007169ad6dc5abe49e396175a317e1ae178b8305c197728fc35f6dcf04", "scoring_hash": "948a078d5ceacd79f3ae2d4c1a1d8f0b67c71fcf3e637a8cb30d126f2ed47d30", "selected": 340, "incomplete": 0, "generated_now": 340, "scored_now": 340, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 340, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:10:02.287090+00:00"}, {"total_records": 640, "completed": 640, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.2109375, "accuracy": 0.2109375, "mean_latency_seconds": 1015.557109405839, "finish_reason_counts": {"length": 81, "stop": 559}, "length_limited_response_count": 81, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 32, "n_trajectories_scored": 640, "reported_accuracy": 0.2109375, "competition_metrics": {"arxiv/february": {"n_problems_scored": 32, "n_trajectories_scored": 640, "reported_accuracy": 0.2109375}}}, "benchmark": "matharena-arxivmath-february", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "fdaa08e51eb4942125e4f22d297e5d58c62b36d392a534c98ad8fd072116bb7f", "scoring_hash": "d6faa135fc0741929b9bc2bf9f551058db3c7b3403b2db67ba25209e4b22dce2", "selected": 640, "incomplete": 0, "generated_now": 640, "scored_now": 640, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 640, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T12:02:16.977867+00:00"}, {"total_records": 460, "completed": 460, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.40652173913043477, "accuracy": 0.40652173913043477, "mean_latency_seconds": 1071.42053608192, "finish_reason_counts": {"length": 63, "stop": 397}, "length_limited_response_count": 63, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 23, "n_trajectories_scored": 460, "reported_accuracy": 0.40652173913043477, "competition_metrics": {"arxiv/january": {"n_problems_scored": 23, "n_trajectories_scored": 460, "reported_accuracy": 0.40652173913043477}}}, "benchmark": "matharena-arxivmath-january", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "e37e0e691696c151ba2c379184cfdcc810cf6cb1ffa777c2063f6c5b8ae837d1", "scoring_hash": "3ef4967c2d483f17357a50c735b64f26e2fb272115af2308f31fcdd651e272e1", "selected": 460, "incomplete": 0, "generated_now": 460, "scored_now": 460, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 460, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:59:46.640134+00:00"}, {"total_records": 620, "completed": 620, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.18870967741935485, "accuracy": 0.18870967741935485, "mean_latency_seconds": 826.4639452366793, "finish_reason_counts": {"length": 23, "stop": 597}, "length_limited_response_count": 23, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 31, "n_trajectories_scored": 620, "reported_accuracy": 0.18870967741935485, "competition_metrics": {"arxiv/march": {"n_problems_scored": 31, "n_trajectories_scored": 620, "reported_accuracy": 0.18870967741935485}}}, "benchmark": "matharena-arxivmath-march", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "c56931c7271c41c1c492d563f541528d6520bd0cec43cee9a32000393a951256", "scoring_hash": "bed079f3e4e0c4b044de96788da4a217fc2c7cc3b027fcc0a99d91188f155fbd", "selected": 620, "incomplete": 0, "generated_now": 620, "scored_now": 620, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 620, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:59:54.027895+00:00"}], "total_records": 2060, "completed": 2060, "accuracy": 0.2587378640776699, "correct": 533.0, "mean_score": 0.2587378640776699, "aggregation": "avg@20", "trials": 20, "questions": 103, "unscored": 0, "provisional": false, "publication_date": "2026-11-12", "source_run": "/workspace/data/sci-llm-training/runtime/arxiv-avg20-20261110/sft-agentic"}, "scicode-verified": {"updated_at": 1793119295.3793085, "expected_problems": 64, "expected_steps": 287, "generated_steps": 287, "scored_steps": 287, "passed_steps": 127, "completed_problems": 64, "passed_problems": 8, "subproblem_accuracy": 0.4425087108013937, "main_problem_accuracy": 0.125, "complete": true, "cap_hits": 0, "mean_output_tokens": 15117.979094076654, "grading_infrastructure_errors": 0, "errors": []}, "scicoding-agent": {"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 37, "avg_at_2": 0.37, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 13, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2}, "frontierscience-research": {"total_records": 1800, "completed": 1800, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.29590305555555557, "accuracy": 0.07333333333333333, "mean_latency_seconds": 505.6271946268312, "finish_reason_counts": {"length": 25, "stop": 1775}, "length_limited_response_count": 25, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 1800, "request_count": 1820, "retry_request_count": 20, "prompt_tokens": 9522776, "cached_prompt_tokens": 124123, "cache_write_prompt_tokens": 9371007, "completion_tokens": 2939042, "reasoning_tokens": 1718732, "total_tokens": 12461818}, "judge_api_usage": {"request_count": 1820, "prompt_tokens": 9522776, "cached_prompt_tokens": 124123, "cache_write_prompt_tokens": 9371007, "completion_tokens": 2939042, "reasoning_tokens": 1718732, "total_tokens": 12461818}, "benchmark_metrics": {"track": "research", "n_problems_scored": 60, "n_trajectories_scored": 1800, "aggregation": "avg@30", "reported_accuracy": 0.07333333333333332, "mean_rubric_score": 0.29590305555555557, "success_threshold": 0.7, "community_profile": null}, "benchmark": "frontierscience-research", "model": "final-906a58d3597d", "execution_mode": "non_agent", "generation_hash": "9a53ae203c2f853346e633dc54dcc75c5a80b034f5c7c0dfcdc7a3157ca88a61", "scoring_hash": "ceea6075bb3e7655a14d6ad2f07750f7b1b588c24da569c70a81bc6665831750", "selected": 1800, "incomplete": 0, "generated_now": 1800, "scored_now": 1800, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 1800, "generation_concurrency": 1536, "judge_concurrency": 256, "updated_at": "2026-10-27T16:08:48.021990+00:00"}, "frontierscience-olympiad": {"total_records": 3000, "completed": 3000, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.614, "accuracy": 0.614, "mean_latency_seconds": 341.9275932600237, "finish_reason_counts": {"length": 18, "stop": 2982}, "length_limited_response_count": 18, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 3000, "request_count": 3000, "retry_request_count": 0, "prompt_tokens": 2460555, "cached_prompt_tokens": 4873, "cache_write_prompt_tokens": 625442, "completion_tokens": 822086, "reasoning_tokens": 596849, "total_tokens": 3282641}, "judge_api_usage": {"request_count": 3000, "prompt_tokens": 2460555, "cached_prompt_tokens": 4873, "cache_write_prompt_tokens": 625442, "completion_tokens": 822086, "reasoning_tokens": 596849, "total_tokens": 3282641}, "benchmark_metrics": {"track": "olympiad", "n_problems_scored": 100, "n_trajectories_scored": 3000, "aggregation": "avg@30", "reported_accuracy": 0.614, "mean_rubric_score": null, "success_threshold": 0.5, "community_profile": null}, "benchmark": "frontierscience-olympiad", "model": "final-906a58d3597d", "execution_mode": "non_agent", "generation_hash": "3cf730a629fac5a44682882a563d9765f9effe71282b54a3b4257c954f2b94bb", "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": 1536, "judge_concurrency": 256, "updated_at": "2026-10-27T16:17:54.676278+00:00"}, "terminal-bench-2.1": {"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 49, "avg_at_2": 0.2752808988764045, "unscored": 1, "agent_timeouts": 111, "reward_positive_with_exception": 6, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": false, "provisional": true, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2}, "hle-verified": {"gold": {"correct": 165, "total": 575, "accuracy": 0.28695652173913044, "cap_hits": 27}, "revision": {"correct": 335, "total": 1025, "accuracy": 0.32682926829268294, "cap_hits": 56}, "correct": 500, "total": 1600, "overall_accuracy": 0.3125, "all_complete": true, "metric": "question-weighted accuracy; one response per question", "provenance": "hle-verified/provenance.json", "artifacts": "hle-verified/"}}, "pending_benchmarks": [], "arxiv_avg20_publication": {"date": "2026-11-12", "path": "arxivmath/", "correct": 533.0, "total": 2060}, "generation_finished": true, "grading_complete": false, "provisional": true, "latest_agentic_rerun": {"published_at": "2026-11-13", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "benchmarks": ["scicoding-agent", "terminal-bench-2.1"], "unscored": 1}}
|
|
|
|
| 1 |
+
{"checkpoint_step": 1144, "full_suite_complete": false, "benchmarks": {"arxivmath": {"benchmark": "arxivmath", "releases": [{"total_records": 340, "completed": 340, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.27647058823529413, "accuracy": 0.27647058823529413, "mean_latency_seconds": 1102.0164873120102, "finish_reason_counts": {"length": 50, "stop": 290}, "length_limited_response_count": 50, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 17, "n_trajectories_scored": 340, "reported_accuracy": 0.27647058823529413, "competition_metrics": {"arxiv/december": {"n_problems_scored": 17, "n_trajectories_scored": 340, "reported_accuracy": 0.27647058823529413}}}, "benchmark": "matharena-arxivmath-december", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "4873a8007169ad6dc5abe49e396175a317e1ae178b8305c197728fc35f6dcf04", "scoring_hash": "948a078d5ceacd79f3ae2d4c1a1d8f0b67c71fcf3e637a8cb30d126f2ed47d30", "selected": 340, "incomplete": 0, "generated_now": 340, "scored_now": 340, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 340, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:10:02.287090+00:00"}, {"total_records": 640, "completed": 640, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.2109375, "accuracy": 0.2109375, "mean_latency_seconds": 1015.557109405839, "finish_reason_counts": {"length": 81, "stop": 559}, "length_limited_response_count": 81, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 32, "n_trajectories_scored": 640, "reported_accuracy": 0.2109375, "competition_metrics": {"arxiv/february": {"n_problems_scored": 32, "n_trajectories_scored": 640, "reported_accuracy": 0.2109375}}}, "benchmark": "matharena-arxivmath-february", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "fdaa08e51eb4942125e4f22d297e5d58c62b36d392a534c98ad8fd072116bb7f", "scoring_hash": "d6faa135fc0741929b9bc2bf9f551058db3c7b3403b2db67ba25209e4b22dce2", "selected": 640, "incomplete": 0, "generated_now": 640, "scored_now": 640, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 640, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T12:02:16.977867+00:00"}, {"total_records": 460, "completed": 460, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.40652173913043477, "accuracy": 0.40652173913043477, "mean_latency_seconds": 1071.42053608192, "finish_reason_counts": {"length": 63, "stop": 397}, "length_limited_response_count": 63, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 23, "n_trajectories_scored": 460, "reported_accuracy": 0.40652173913043477, "competition_metrics": {"arxiv/january": {"n_problems_scored": 23, "n_trajectories_scored": 460, "reported_accuracy": 0.40652173913043477}}}, "benchmark": "matharena-arxivmath-january", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "e37e0e691696c151ba2c379184cfdcc810cf6cb1ffa777c2063f6c5b8ae837d1", "scoring_hash": "3ef4967c2d483f17357a50c735b64f26e2fb272115af2308f31fcdd651e272e1", "selected": 460, "incomplete": 0, "generated_now": 460, "scored_now": 460, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 460, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:59:46.640134+00:00"}, {"total_records": 620, "completed": 620, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.18870967741935485, "accuracy": 0.18870967741935485, "mean_latency_seconds": 826.4639452366793, "finish_reason_counts": {"length": 23, "stop": 597}, "length_limited_response_count": 23, "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}, "benchmark_metrics": {"competition": "arxivmath", "aggregation": "avg@20", "n_problems_scored": 31, "n_trajectories_scored": 620, "reported_accuracy": 0.18870967741935485, "competition_metrics": {"arxiv/march": {"n_problems_scored": 31, "n_trajectories_scored": 620, "reported_accuracy": 0.18870967741935485}}}, "benchmark": "matharena-arxivmath-march", "model": "sft-agentic-arxiv-avg20", "execution_mode": "non_agent", "generation_hash": "c56931c7271c41c1c492d563f541528d6520bd0cec43cee9a32000393a951256", "scoring_hash": "bed079f3e4e0c4b044de96788da4a217fc2c7cc3b027fcc0a99d91188f155fbd", "selected": 620, "incomplete": 0, "generated_now": 620, "scored_now": 620, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 620, "generation_concurrency": 256, "judge_concurrency": 1, "updated_at": "2026-11-11T11:59:54.027895+00:00"}], "total_records": 2060, "completed": 2060, "accuracy": 0.2587378640776699, "correct": 533.0, "mean_score": 0.2587378640776699, "aggregation": "avg@20", "trials": 20, "questions": 103, "unscored": 0, "provisional": false, "publication_date": "2026-11-12", "source_run": "/workspace/data/sci-llm-training/runtime/arxiv-avg20-20261110/sft-agentic"}, "scicode-verified": {"updated_at": 1793119295.3793085, "expected_problems": 64, "expected_steps": 287, "generated_steps": 287, "scored_steps": 287, "passed_steps": 127, "completed_problems": 64, "passed_problems": 8, "subproblem_accuracy": 0.4425087108013937, "main_problem_accuracy": 0.125, "complete": true, "cap_hits": 0, "mean_output_tokens": 15117.979094076654, "grading_infrastructure_errors": 0, "errors": []}, "scicoding-agent": {"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 37, "avg_at_2": 0.37, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 13, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "adjudicated_zero": 0, "native_scored": 100}, "frontierscience-research": {"total_records": 1800, "completed": 1800, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.29590305555555557, "accuracy": 0.07333333333333333, "mean_latency_seconds": 505.6271946268312, "finish_reason_counts": {"length": 25, "stop": 1775}, "length_limited_response_count": 25, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 1800, "request_count": 1820, "retry_request_count": 20, "prompt_tokens": 9522776, "cached_prompt_tokens": 124123, "cache_write_prompt_tokens": 9371007, "completion_tokens": 2939042, "reasoning_tokens": 1718732, "total_tokens": 12461818}, "judge_api_usage": {"request_count": 1820, "prompt_tokens": 9522776, "cached_prompt_tokens": 124123, "cache_write_prompt_tokens": 9371007, "completion_tokens": 2939042, "reasoning_tokens": 1718732, "total_tokens": 12461818}, "benchmark_metrics": {"track": "research", "n_problems_scored": 60, "n_trajectories_scored": 1800, "aggregation": "avg@30", "reported_accuracy": 0.07333333333333332, "mean_rubric_score": 0.29590305555555557, "success_threshold": 0.7, "community_profile": null}, "benchmark": "frontierscience-research", "model": "final-906a58d3597d", "execution_mode": "non_agent", "generation_hash": "9a53ae203c2f853346e633dc54dcc75c5a80b034f5c7c0dfcdc7a3157ca88a61", "scoring_hash": "ceea6075bb3e7655a14d6ad2f07750f7b1b588c24da569c70a81bc6665831750", "selected": 1800, "incomplete": 0, "generated_now": 1800, "scored_now": 1800, "cached_completed": 0, "stage": "all", "worker_count": 1, "worker_index": 0, "worker_owned_predictions": 1800, "generation_concurrency": 1536, "judge_concurrency": 256, "updated_at": "2026-10-27T16:08:48.021990+00:00"}, "frontierscience-olympiad": {"total_records": 3000, "completed": 3000, "query_errors": 0, "score_errors": 0, "generated_unscored": 0, "mean_score": 0.614, "accuracy": 0.614, "mean_latency_seconds": 341.9275932600237, "finish_reason_counts": {"length": 18, "stop": 2982}, "length_limited_response_count": 18, "observed_max_completion_tokens": 98304, "judge_usage": {"judged_record_count": 3000, "request_count": 3000, "retry_request_count": 0, "prompt_tokens": 2460555, "cached_prompt_tokens": 4873, "cache_write_prompt_tokens": 625442, "completion_tokens": 822086, "reasoning_tokens": 596849, "total_tokens": 3282641}, "judge_api_usage": {"request_count": 3000, "prompt_tokens": 2460555, "cached_prompt_tokens": 4873, "cache_write_prompt_tokens": 625442, "completion_tokens": 822086, "reasoning_tokens": 596849, "total_tokens": 3282641}, "benchmark_metrics": {"track": "olympiad", "n_problems_scored": 100, "n_trajectories_scored": 3000, "aggregation": "avg@30", "reported_accuracy": 0.614, "mean_rubric_score": null, "success_threshold": 0.5, "community_profile": null}, "benchmark": "frontierscience-olympiad", "model": "final-906a58d3597d", "execution_mode": "non_agent", "generation_hash": "3cf730a629fac5a44682882a563d9765f9effe71282b54a3b4257c954f2b94bb", "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": 1536, "judge_concurrency": 256, "updated_at": "2026-10-27T16:17:54.676278+00:00"}, "terminal-bench-2.1": {"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 49, "avg_at_2": 0.2752808988764045, "unscored": 0, "agent_timeouts": 111, "reward_positive_with_exception": 6, "scoring": "native_verifier_with_explicit_user_adjudication", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "adjudicated_zero": 1, "native_scored": 177}, "hle-verified": {"gold": {"correct": 165, "total": 575, "accuracy": 0.28695652173913044, "cap_hits": 27}, "revision": {"correct": 335, "total": 1025, "accuracy": 0.32682926829268294, "cap_hits": 56}, "correct": 500, "total": 1600, "overall_accuracy": 0.3125, "all_complete": true, "metric": "question-weighted accuracy; one response per question", "provenance": "hle-verified/provenance.json", "artifacts": "hle-verified/"}}, "pending_benchmarks": [], "arxiv_avg20_publication": {"date": "2026-11-12", "path": "arxivmath/", "correct": 533.0, "total": 2060}, "generation_finished": true, "grading_complete": true, "provisional": false, "latest_agentic_rerun": {"published_at": "2026-11-15", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "benchmarks": ["scicoding-agent", "terminal-bench-2.1"], "unscored": 0}}
|
sft-agentic-stage2/terminal-bench-2.1/README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
-
# terminal-bench-2.1:
|
| 2 |
|
| 3 |
-
49/178 passed (27.53% avg@2);
|
| 4 |
|
| 5 |
Terminus 2; temperature 0.7, top-p 0.95, top-k 20; H100, TP2, eight engines, concurrency two per engine. Frozen plans and native artifacts are included. Native verifier success counts even after abnormal agent stops.
|
|
|
|
| 1 |
+
# terminal-bench-2.1: Agentic SFT rerun
|
| 2 |
|
| 3 |
+
49/178 passed (27.53% avg@2); 0 unscored. All planned attempts remain in the denominator. Originally missing verdicts are explicitly user-adjudicated zero. Where applicable, original null verdicts and errors are preserved in scoring_adjudications.json and its backup archive.
|
| 4 |
|
| 5 |
Terminus 2; temperature 0.7, top-p 0.95, top-k 20; H100, TP2, eight engines, concurrency two per engine. Frozen plans and native artifacts are included. Native verifier success counts even after abnormal agent stops.
|
sft-agentic-stage2/terminal-bench-2.1/judge_trajectories.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
sft-agentic-stage2/terminal-bench-2.1/provenance.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "published_date": "2026-11-
|
|
|
|
| 1 |
+
{"source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "published_date": "2026-11-15", "replaces_revision": "f74262971417a10dc91b9084c0f6658ffcbf7d57", "exporter_sha256": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15", "generation_or_judgment_rerun": false, "manual_zero_adjudications": true}
|
sft-agentic-stage2/terminal-bench-2.1/scoring_adjudications.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"extract-moves-from-video--attempt-002": {"method": "explicit_user_authorized_zero", "authorization": "[REDACTED_CREDENTIAL]", "timestamp": "2026-11-15T17:56:30.367848+00:00", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_info": {"exception_type": "AgentTimeoutError", "exception_message": "Agent execution timed out after 1800.0 seconds", "exception_traceback": "Traceback (most recent call last):\n File \"/usr/local/lib/python3.12/asyncio/tasks.py\", line 520, in wait_for\n return await fut\n ^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1743, in run\n await self._run_agent_loop(\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1523, in _run_agent_loop\n timeout_occurred, terminal_output = await self._execute_commands(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1353, in _execute_commands\n return False, self._limit_output_length(await session.get_incremental_output())\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/agents/terminus_2/tmux_session.py\", line 859, in get_incremental_output\n current_buffer = await self.capture_pane(capture_entire=True)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/agents/terminus_2/tmux_session.py\", line 828, in capture_pane\n result = await self.environment.exec(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/environments/docker/docker.py\", line 1221, in exec\n return await self._compose_exec(\n ^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/environments/docker/docker.py\", line 1298, in _compose_exec\n return await self._run_docker_compose_command(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/environments/docker/docker.py\", line 697, in _run_docker_compose_command\n result = await self._collect_buffered_output(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/environments/docker/docker.py\", line 727, in _collect_buffered_output\n stdout_bytes, stderr_bytes = await process.communicate(input=stdin_data)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/usr/local/lib/python3.12/asyncio/subprocess.py\", line 201, in communicate\n stdin, stdout, stderr = await tasks.gather(stdin, stdout, stderr)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/usr/local/lib/python3.12/asyncio/subprocess.py\", line 181, in _read_stream\n output = await stream.read()\n ^^^^^^^^^^^^^^^^^^^\n File \"/usr/local/lib/python3.12/asyncio/streams.py\", line 706, in read\n block = await self.read(self._limit)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/usr/local/lib/python3.12/asyncio/streams.py\", line 713, in read\n await self._wait_for_data('read')\n File \"/usr/local/lib/python3.12/asyncio/streams.py\", line 545, in _wait_for_data\n await self._waiter\nasyncio.exceptions.CancelledError\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/trial/trial.py\", line 554, in _run_agent_phase\n await asyncio.wait_for(\n File \"/usr/local/lib/python3.12/asyncio/tasks.py\", line 519, in wait_for\n async with timeouts.timeout(timeout):\n ^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/usr/local/lib/python3.12/asyncio/timeouts.py\", line 115, in __aexit__\n raise TimeoutError from exc_val\nTimeoutError\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n File \"/workspace/data/sci-llm-training/runtime/rl-agentic-new-step40-agentic-20261112/code/eval-framework/scripts/agentic_recovery.py\", line 180, in run\n await original(*args, **kwargs)\n File \"/opt/scicoding/venvs/agentic-qualified-20261109/lib/python3.12/site-packages/harbor/trial/trial.py\", line 563, in _run_agent_phase\n raise AgentTimeoutError(\nharbor.trial.errors.AgentTimeoutError: Agent execution timed out after 1800.0 seconds\n", "occurred_at": "2026-11-13T12:44:55.574218Z"}, "original_result_sha256": "482c087ec9c9761411498f00514b491f699d986950a65319b0e922f1cf71cb47", "generation_or_verifier_rerun": false, "reason": "User-adjudicated zero for missing verdict; not a native completed verifier test verdict."}}
|
sft-agentic-stage2/terminal-bench-2.1/scoring_adjudications.tar.gz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2e0a73d4edac8bbe364b81224c5719c6bd6d75d0a231f722c54a51602bb2e91c
|
| 3 |
+
size 33503
|
sft-agentic-stage2/terminal-bench-2.1/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 49, "avg_at_2": 0.2752808988764045, "unscored":
|
|
|
|
| 1 |
+
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 49, "avg_at_2": 0.2752808988764045, "unscored": 0, "agent_timeouts": 111, "reward_positive_with_exception": 6, "scoring": "native_verifier_with_explicit_user_adjudication", "scoring_policy": "native-verifier-after-agent-stop-v1", "generation_finished": true, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-aligned-followups-20261113/01-sft-agentic", "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "adjudicated_zero": 1, "native_scored": 177}
|
sft-agentic-stage2/terminal-bench-2.1/trajectories.jsonl
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1f1d2c140ad9de823bf3d76494fc0f6cfab9a5c0fce86e4240a5447bb00b2959
|
| 3 |
+
size 87319910
|