Publish aligned H100 agentic reruns; replace stale results and audit configuration exceptions
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- AGENTIC_CONFIG_ALIGNMENT.md +25 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/README.md +16 -11
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/manifest.json +154 -1
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/provenance.json +0 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/recovery_provenance.tar.gz +0 -3
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/README.md +5 -1
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/configs.json +0 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/judge_trajectories.jsonl +0 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/native_artifacts.tar.gz +2 -2
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/provenance.json +8 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/summary.json +1 -1
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/trajectories.jsonl +2 -2
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/summary.json +51 -55
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/README.md +5 -1
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/configs.json +0 -0
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/judge_trajectories.jsonl +2 -2
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/native_artifacts.tar.gz +2 -2
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/provenance.json +39 -0
- sft-agentic-stage2/terminal-bench-2.1/recovery_original_attempts.tar.gz → NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/scoring_adjudications.tar.gz +2 -2
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/summary.json +1 -1
- NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/trajectories.jsonl +2 -2
- Qwen3.5-35B-A3B/README.md +14 -13
- Qwen3.5-35B-A3B/manifest.json +154 -1
- Qwen3.5-35B-A3B/provenance.json +0 -0
- Qwen3.5-35B-A3B/scicoding-agent/README.md +5 -1
- Qwen3.5-35B-A3B/scicoding-agent/configs.json +0 -0
- Qwen3.5-35B-A3B/scicoding-agent/judge_trajectories.jsonl +0 -0
- Qwen3.5-35B-A3B/scicoding-agent/native_artifacts.tar.gz +2 -2
- Qwen3.5-35B-A3B/scicoding-agent/provenance.json +8 -0
- Qwen3.5-35B-A3B/scicoding-agent/summary.json +1 -1
- Qwen3.5-35B-A3B/scicoding-agent/trajectories.jsonl +2 -2
- Qwen3.5-35B-A3B/summary.json +67 -34
- Qwen3.5-35B-A3B/terminal-bench-2.1/README.md +5 -1
- Qwen3.5-35B-A3B/terminal-bench-2.1/configs.json +0 -0
- Qwen3.5-35B-A3B/terminal-bench-2.1/judge_trajectories.jsonl +0 -0
- Qwen3.5-35B-A3B/terminal-bench-2.1/native_artifacts.tar.gz +2 -2
- Qwen3.5-35B-A3B/terminal-bench-2.1/provenance.json +54 -0
- Qwen3.5-35B-A3B/{agentic_recovery_provenance.tar.gz → terminal-bench-2.1/scoring_adjudications.tar.gz} +2 -2
- Qwen3.5-35B-A3B/terminal-bench-2.1/summary.json +1 -1
- Qwen3.5-35B-A3B/terminal-bench-2.1/trajectories.jsonl +2 -2
- README.md +20 -0
- gpt-oss-20b/README.md +15 -20
- gpt-oss-20b/configs.json +0 -0
- gpt-oss-20b/manifest.json +52 -60
- gpt-oss-20b/provenance.json +0 -0
- gpt-oss-20b/recovery-provenance/code-hashes.json +0 -1
- gpt-oss-20b/recovery-provenance/combined-summary.json +0 -1
- gpt-oss-20b/recovery-provenance/results-overlay.json +0 -1
- gpt-oss-20b/recovery-provenance/superseded-http400-attempts/summary.json +0 -1
- gpt-oss-20b/recovery-provenance/superseded-http400-attempts/trajectories.jsonl +0 -3
AGENTIC_CONFIG_ALIGNMENT.md
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Agentic configuration alignment audit — 2026-11-08
|
| 2 |
+
|
| 3 |
+
## Common sampling contract
|
| 4 |
+
|
| 5 |
+
For the self-hosted published agentic evaluations: temperature 1.0, top-p 0.95, top-k 20; Terminus 2; two independent attempts per task (avg@2). Each task's native verifier determines reward. Abnormal-but-verifier-passing runs remain passes under the published policy. Explicit timeout-zero adjudications remain separately identified.
|
| 6 |
+
|
| 7 |
+
## Replacement scope and evidence
|
| 8 |
+
|
| 9 |
+
Eight folders replaced: SciCoding-Agent and Terminal-Bench 2.1 for sft-agentic-stage2, NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16, Qwen3.5-35B-A3B, and gpt-oss-20b. All 1,112 effective native trial configs were audited before publication; each matches temperature 1.0, top-p 0.95, top-k 20 and Terminus 2. Frozen plans, per-file hashes, effective native agent configs and source provenance are published in each benchmark configs.json and provenance.json.
|
| 10 |
+
|
| 11 |
+
## Remaining execution differences — not hidden or retroactively edited
|
| 12 |
+
|
| 13 |
+
| Published group | Sampling T / p / k | Attempts per task | Tasks per engine | Evidence |
|
| 14 |
+
|---|---|---:|---:|---|
|
| 15 |
+
| Four aligned replacement models | 1.0 / 0.95 / 20 | 2 | 2 | Native configs for all 1,112 attempts; frozen contracts |
|
| 16 |
+
| RL stage 1 step 120 | 1.0 / 0.95 / 20 | 2 | 2 | Published benchmark-specific frozen plans/contracts |
|
| 17 |
+
| Qwen3.5-9B IT, SFT mix | 1.0 / 0.95 / 20 | 2 | 1 | Published benchmark-specific frozen plans/contracts |
|
| 18 |
+
| DeepSeek Flash 0731, Gemma-4-12B IT | 1.0 / 0.95 / 20 | 2 | 1 | Published root provenance agentic plans/contracts |
|
| 19 |
+
| GPT-5.6-Luna | Provider defaults; T/p unspecified | 4 | API-based, not directly comparable | Published evaluation_policy |
|
| 20 |
+
|
| 21 |
+
Sampling alignment is not full runtime equivalence: tensor parallelism, hardware/runtime versions, model-specific reasoning transport and historical concurrency may differ. Existing source configurations are authoritative. Luna remains an explicitly separate historical API baseline, not part of the aligned self-hosted cohort. No runs were relaunched solely to equalize historical concurrency.
|
| 22 |
+
|
| 23 |
+
## Preservation
|
| 24 |
+
|
| 25 |
+
The replacement is rebased onto the latest HF main revision. Reasoning, HLE, SciCode and other models' raw result files are untouched. Original agentic exports remain recoverable from Git history. Canonical current scores are in benchmark summary.json and refreshed parent summaries/manifests; do not use superseded cached comparison tables.
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/README.md
CHANGED
|
@@ -1,20 +1,25 @@
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
|
| 2 |
|
| 3 |
-
|
| 4 |
|
| 5 |
-
| Benchmark |
|
| 6 |
-
|---|---|
|
| 7 |
-
|
|
| 8 |
-
|
|
| 9 |
-
| frontierscience-research | See [frontierscience-research/summary.json](frontierscience-research/summary.json) |
|
| 10 |
-
| scicode-verified | See [scicode-verified/summary.json](scicode-verified/summary.json) |
|
| 11 |
-
| scicoding-agent | See [scicoding-agent/summary.json](scicoding-agent/summary.json) |
|
| 12 |
-
| terminal-bench-2.1 | See [terminal-bench-2.1/summary.json](terminal-bench-2.1/summary.json) |
|
| 13 |
|
| 14 |
-
|
| 15 |
|
| 16 |
-
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
## HLE-Verified (2026-10-30)
|
| 20 |
|
|
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
|
| 2 |
|
| 3 |
+
## Aligned agentic rerun — 2026-11-04
|
| 4 |
|
| 5 |
+
| Benchmark | avg@2 | User-adjudicated timeout zeros |
|
| 6 |
+
|---|---:|---:|
|
| 7 |
+
| scicoding-agent | 19/100 (19.00%) | 0 |
|
| 8 |
+
| terminal-bench-2.1 | 35/178 (19.66%) | 2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
+
H100; Terminus 2; temperature 1.0, top-p 0.95, top-k 20; concurrency 2 per engine. TP1 for GPT-OSS-20B; TP2 for the other models. Two independent attempts per task. All attempts are scored. Timeout zeros preserve their original native null verdict and evidence; they are not native test failures.
|
| 11 |
|
| 12 |
+
These benchmark folders replace the older inconsistent agentic exports. Complete trajectories, continuation chains, verifier artifacts and frozen configs are included. Older exports remain recoverable in repository revision `55136ae404538e1ebb3dfaba6261b529109883e9`.
|
| 13 |
|
| 14 |
+
## Unchanged non-agentic evaluations
|
| 15 |
+
|
| 16 |
+
These results are preserved from the latest pre-publication revision, not rolled back to the original export date.
|
| 17 |
+
|
| 18 |
+
- [arxivmath](arxivmath/summary.json)
|
| 19 |
+
- [frontierscience-olympiad](frontierscience-olympiad/summary.json)
|
| 20 |
+
- [frontierscience-research](frontierscience-research/summary.json)
|
| 21 |
+
- [scicode-verified](scicode-verified/summary.json)
|
| 22 |
+
- [hle-verified](hle-verified/summary.json)
|
| 23 |
|
| 24 |
## HLE-Verified (2026-10-30)
|
| 25 |
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/manifest.json
CHANGED
|
@@ -1 +1,154 @@
|
|
| 1 |
-
{
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"README.md": {
|
| 3 |
+
"size": 2408,
|
| 4 |
+
"sha256": "cdf801d6f9fed9b1a2c059b617564f839e3c8cdaa34c1727104a8ff72f1a8564"
|
| 5 |
+
},
|
| 6 |
+
"arxivmath/README.md": {
|
| 7 |
+
"size": 310,
|
| 8 |
+
"sha256": "137f3be1941e9875fbcdca77fc6745408c9bb94940c828e63d38685b2d8618a9"
|
| 9 |
+
},
|
| 10 |
+
"arxivmath/events.jsonl.gz": {
|
| 11 |
+
"size": 39276415,
|
| 12 |
+
"sha256": "5afc0105ffa39b476a00f469c0d048570a59e55ba08850aa4bfc2112c270c3e5"
|
| 13 |
+
},
|
| 14 |
+
"arxivmath/judge_trajectories.jsonl": {
|
| 15 |
+
"size": 222690,
|
| 16 |
+
"sha256": "bf85b0a5c53260e0b251c855876cdf265ce249ae6ece5f17a8df733b040e8449"
|
| 17 |
+
},
|
| 18 |
+
"arxivmath/summary.json": {
|
| 19 |
+
"size": 7057,
|
| 20 |
+
"sha256": "ab1e425cb4e830da49e75b8635910cc737310ca0ef3cb5dd7b298fb189316f45"
|
| 21 |
+
},
|
| 22 |
+
"arxivmath/trajectories.jsonl": {
|
| 23 |
+
"size": 70534675,
|
| 24 |
+
"sha256": "23874159feab0212859c246b2cdff6a7e3eec367407aac772fb06f57e0d5535f"
|
| 25 |
+
},
|
| 26 |
+
"frontierscience-olympiad/README.md": {
|
| 27 |
+
"size": 325,
|
| 28 |
+
"sha256": "3f6ec34d4ebf1b4eea5ea672e1bbcc5f9dad0d82169d7c43c9ac2ac16b23aac6"
|
| 29 |
+
},
|
| 30 |
+
"frontierscience-olympiad/events.jsonl.gz": {
|
| 31 |
+
"size": 218811226,
|
| 32 |
+
"sha256": "d8a8d92cdc8296ff38b7e3a7c74b0338c5792ad3857161ce2f7d51a21494cfc1"
|
| 33 |
+
},
|
| 34 |
+
"frontierscience-olympiad/judge_trajectories.jsonl": {
|
| 35 |
+
"size": 10815806,
|
| 36 |
+
"sha256": "a5faa8a15da6a6fd554c7e8a8267df349e0ad764bfc42203483f5cb87a39a82a"
|
| 37 |
+
},
|
| 38 |
+
"frontierscience-olympiad/summary.json": {
|
| 39 |
+
"size": 1819,
|
| 40 |
+
"sha256": "607892480c4b5352e2685f4b33ec356d07775f5417d7bcd39fd299a4316c37b1"
|
| 41 |
+
},
|
| 42 |
+
"frontierscience-olympiad/trajectories.jsonl": {
|
| 43 |
+
"size": 452095604,
|
| 44 |
+
"sha256": "cbaeb301c28eaa9688c13f307b04cea3eee4d0f9c56df8d44898ac53a531044a"
|
| 45 |
+
},
|
| 46 |
+
"frontierscience-research/README.md": {
|
| 47 |
+
"size": 325,
|
| 48 |
+
"sha256": "273395f47e9e7824017e5fcc307251f32f172d2340761181fa14307bb4aee1c9"
|
| 49 |
+
},
|
| 50 |
+
"frontierscience-research/events.jsonl.gz": {
|
| 51 |
+
"size": 112609792,
|
| 52 |
+
"sha256": "c459c0a4dcbfe22d5d2b42e9d657c14e788c916c4be93fe99a06e1f9a841985b"
|
| 53 |
+
},
|
| 54 |
+
"frontierscience-research/judge_trajectories.jsonl": {
|
| 55 |
+
"size": 28089196,
|
| 56 |
+
"sha256": "c331a94ee9c3ba10f02be2ef29657885b7d1a45dcf1afb5db15212d4a96543b5"
|
| 57 |
+
},
|
| 58 |
+
"frontierscience-research/summary.json": {
|
| 59 |
+
"size": 1886,
|
| 60 |
+
"sha256": "28eaa6f01079fbf9c0e09a30880a7fa3505c4f50b858bcca7ddf85bf6d9746c3"
|
| 61 |
+
},
|
| 62 |
+
"frontierscience-research/trajectories.jsonl": {
|
| 63 |
+
"size": 233262119,
|
| 64 |
+
"sha256": "b47f65b5c0dcd0c68373b4640807dd48c7ceac1d9cb65904dcf22a91eb170d8d"
|
| 65 |
+
},
|
| 66 |
+
"provenance.json": {
|
| 67 |
+
"size": 475719,
|
| 68 |
+
"sha256": "0b97c0af3c0f1908b813ec9d4fe744afe2465d4d30aca6ba88d89e2726c33caf"
|
| 69 |
+
},
|
| 70 |
+
"scicode-verified/README.md": {
|
| 71 |
+
"size": 317,
|
| 72 |
+
"sha256": "95a5d726d43f080acdc8b843f342260e733b7b6e0a72f2647960b206c10ccdec"
|
| 73 |
+
},
|
| 74 |
+
"scicode-verified/judge_trajectories.jsonl": {
|
| 75 |
+
"size": 287328,
|
| 76 |
+
"sha256": "5c48b70843fac0986d8ec7c39c3b5c14a7a92e387f48665b5dbbd1a7fe6b09f3"
|
| 77 |
+
},
|
| 78 |
+
"scicode-verified/native_artifacts.tar.gz": {
|
| 79 |
+
"size": 3152219,
|
| 80 |
+
"sha256": "e03cc9e03b35c7faf02232279f8e325d57d93ec3d77f6ed52aa025d7400f056c"
|
| 81 |
+
},
|
| 82 |
+
"scicode-verified/summary.json": {
|
| 83 |
+
"size": 525,
|
| 84 |
+
"sha256": "30c2b004b8e39187fe5715a55c510a3d87200ae7812b935ec6e71a1b0221a5ab"
|
| 85 |
+
},
|
| 86 |
+
"scicode-verified/trajectories.jsonl": {
|
| 87 |
+
"size": 12368826,
|
| 88 |
+
"sha256": "2bc11b389c7345b916fe78c75567731fd7035b3249f956b8d347fc1d264ab990"
|
| 89 |
+
},
|
| 90 |
+
"scicoding-agent/README.md": {
|
| 91 |
+
"size": 852,
|
| 92 |
+
"sha256": "b49d1e4f8f3628ccb6bb94e5d9b77f2c3c78e72456886a4a927b178593fa44c1"
|
| 93 |
+
},
|
| 94 |
+
"scicoding-agent/judge_trajectories.jsonl": {
|
| 95 |
+
"size": 427308,
|
| 96 |
+
"sha256": "35ecd4329881e3e89d5ac3031abcf8947f5fd3756e22250bc2bc5d64678ffac7"
|
| 97 |
+
},
|
| 98 |
+
"scicoding-agent/native_artifacts.tar.gz": {
|
| 99 |
+
"size": 8243560,
|
| 100 |
+
"sha256": "d0068883f0701fdf1cad072f050e31748780887c54adb57eb1f937c2dce7d1f9"
|
| 101 |
+
},
|
| 102 |
+
"scicoding-agent/summary.json": {
|
| 103 |
+
"size": 599,
|
| 104 |
+
"sha256": "b99179af6e304e427cc6f4a200f99fb4acced31841280e141858c1192b63ef00"
|
| 105 |
+
},
|
| 106 |
+
"scicoding-agent/trajectories.jsonl": {
|
| 107 |
+
"size": 38843777,
|
| 108 |
+
"sha256": "352c62194d397e84b1f1c628045e15a0594f95676a9c64cf7602dd6b1c741a8f"
|
| 109 |
+
},
|
| 110 |
+
"summary.json": {
|
| 111 |
+
"size": 18626,
|
| 112 |
+
"sha256": "e4eae8eb425e731e3502f8fa400b3648d41751bdc062c44b18951735762873b2"
|
| 113 |
+
},
|
| 114 |
+
"terminal-bench-2.1/README.md": {
|
| 115 |
+
"size": 855,
|
| 116 |
+
"sha256": "05c52257f5065e57e4c9bdf1c8beca02d0d609962ae53f5bcb25abb0afb102b0"
|
| 117 |
+
},
|
| 118 |
+
"terminal-bench-2.1/judge_trajectories.jsonl": {
|
| 119 |
+
"size": 15471933,
|
| 120 |
+
"sha256": "863c67aa40e75c22cf670b4c233832118940a2211e1c01f07631538b8098263c"
|
| 121 |
+
},
|
| 122 |
+
"terminal-bench-2.1/native_artifacts.tar.gz": {
|
| 123 |
+
"size": 842664153,
|
| 124 |
+
"sha256": "1c8df8a1f222a7c7644c744f08e023c1c22003e508788aa757903ab1981b2336"
|
| 125 |
+
},
|
| 126 |
+
"terminal-bench-2.1/summary.json": {
|
| 127 |
+
"size": 2374,
|
| 128 |
+
"sha256": "908667af59a5ed29fd0e4fd709a644d4006871a0d238dc0905ab2c7e902f1d11"
|
| 129 |
+
},
|
| 130 |
+
"terminal-bench-2.1/trajectories.jsonl": {
|
| 131 |
+
"size": 69871030,
|
| 132 |
+
"sha256": "f7cc05e0f31fbb9f72635ff858367848fe2a1f273f8d03b4a520a2081d330ad6"
|
| 133 |
+
},
|
| 134 |
+
"scicoding-agent/configs.json": {
|
| 135 |
+
"size": 243094,
|
| 136 |
+
"sha256": "7437167d2851ddba302a6e06e435a15afc6320b592a4d657809182a0344ac6c8"
|
| 137 |
+
},
|
| 138 |
+
"scicoding-agent/provenance.json": {
|
| 139 |
+
"size": 343,
|
| 140 |
+
"sha256": "5d27f5dde045e92c6858a2069809f9e5dd71f88bd51387e141aafebd6d63106d"
|
| 141 |
+
},
|
| 142 |
+
"terminal-bench-2.1/configs.json": {
|
| 143 |
+
"size": 242872,
|
| 144 |
+
"sha256": "fac78b0cf685f19a88939839bb8c408f3249cde91abb525b689f8bb37514c3a8"
|
| 145 |
+
},
|
| 146 |
+
"terminal-bench-2.1/provenance.json": {
|
| 147 |
+
"size": 2251,
|
| 148 |
+
"sha256": "f4022937818f5e410c615bab993135cc9e1c7215c84b35f7dc19b627958c59e0"
|
| 149 |
+
},
|
| 150 |
+
"terminal-bench-2.1/scoring_adjudications.tar.gz": {
|
| 151 |
+
"size": 61565,
|
| 152 |
+
"sha256": "cd0c89af900e55b2da3c81a2f04aa0807adc37b7dae1a5c666ffe07de3a35fb8"
|
| 153 |
+
}
|
| 154 |
+
}
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/provenance.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/recovery_provenance.tar.gz
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:89a919e3c02bff672ae7130e8fe9bb08287e38cb91851c787d48998ca8a4f98b
|
| 3 |
-
size 1494157
|
|
|
|
|
|
|
|
|
|
|
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/README.md
CHANGED
|
@@ -1,3 +1,7 @@
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 — scicoding-agent
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 — scicoding-agent
|
| 2 |
|
| 3 |
+
19/100 = 19.00% avg@2. Native scores: 100; explicitly user-adjudicated timeout zeros: 0; unscored: 0.
|
| 4 |
+
|
| 5 |
+
Aligned rerun: H100, Terminus 2, temperature 1.0, top-p 0.95, top-k 20, two active tasks per engine, two independent attempts per task. Exact model-specific serving, reasoning, compaction and timeout settings are frozen in configs.json.
|
| 6 |
+
|
| 7 |
+
trajectories.jsonl includes reasoning, tool exchanges and compaction continuation chains; judge_trajectories.jsonl contains native verifier outputs. native_artifacts.tar.gz preserves the available complete native trial artifacts. Timeout adjudications are not completed native verifier verdicts; original null rewards and evidence are preserved in scoring_adjudications.tar.gz where applicable. Every planned attempt remains in the denominator.
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/configs.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/judge_trajectories.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/native_artifacts.tar.gz
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:d0068883f0701fdf1cad072f050e31748780887c54adb57eb1f937c2dce7d1f9
|
| 3 |
+
size 8243560
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/provenance.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run",
|
| 3 |
+
"supersedes_hf_revision": "5da71d89fedc2a5e6c7c36c9883ed86d1d09f020",
|
| 4 |
+
"regenerated_for_publication": false,
|
| 5 |
+
"rejudged_for_publication": false,
|
| 6 |
+
"adjudications": {},
|
| 7 |
+
"native_scores_are_distinct_from_adjudicated_zeros": true
|
| 8 |
+
}
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed":
|
|
|
|
| 1 |
+
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 19, "avg_at_2": 0.19, "unscored": 0, "agent_timeouts": 12, "reward_positive_with_exception": 0, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "native_scored": 100, "adjudicated_zero": 0, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run", "sampling": {"temperature": 1.0, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "scoring_adjudications": {}}
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/scicoding-agent/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 sha256:352c62194d397e84b1f1c628045e15a0594f95676a9c64cf7602dd6b1c741a8f
|
| 3 |
+
size 38843777
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/summary.json
CHANGED
|
@@ -431,87 +431,83 @@
|
|
| 431 |
"attempts": 100,
|
| 432 |
"attempts_per_task": 2,
|
| 433 |
"metric": "avg@2",
|
| 434 |
-
"passed":
|
| 435 |
-
"avg_at_2": 0.
|
| 436 |
"unscored": 0,
|
| 437 |
-
"agent_timeouts":
|
| 438 |
-
"reward_positive_with_exception":
|
| 439 |
"scoring": "native_verifier",
|
| 440 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
|
|
|
|
|
|
|
|
|
| 441 |
"provisional": false,
|
| 442 |
-
"source_run": "/workspace/data/sci-llm-training/runtime/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 443 |
},
|
| 444 |
"terminal-bench-2.1": {
|
| 445 |
"tasks": 89,
|
| 446 |
"attempts": 178,
|
| 447 |
"attempts_per_task": 2,
|
| 448 |
"metric": "avg@2",
|
| 449 |
-
"passed":
|
| 450 |
-
"avg_at_2": 0.
|
| 451 |
"unscored": 0,
|
| 452 |
-
"agent_timeouts":
|
| 453 |
-
"reward_positive_with_exception":
|
| 454 |
-
"scoring": "
|
| 455 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
|
|
|
|
|
|
|
|
|
| 456 |
"provisional": false,
|
| 457 |
-
"source_run": "/workspace/data/sci-llm-training/runtime/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 458 |
"scoring_adjudications": {
|
| 459 |
"mailman--attempt-002": {
|
| 460 |
-
"timestamp": "2026-
|
| 461 |
-
"
|
| 462 |
-
"native_reward": null,
|
| 463 |
-
"original_verifier_result": null,
|
| 464 |
-
"method": "explicit_user_authorized_zero_for_unscored_attempt",
|
| 465 |
"authorization": "[REDACTED_CREDENTIAL]",
|
| 466 |
-
"reason": "Original attempt ended without a native verifier reward; this is an adjudicated zero, not a completed native test verdict.",
|
| 467 |
-
"exception_type": "VerifierTimeoutError",
|
| 468 |
-
"backup": "/workspace/data/sci-llm-training/runtime/nemotron35-lightning-full-suite-20261029/operations/user-zero-adjudication-20261029/terminal-bench-2.1--mailman--attempt-002",
|
| 469 |
-
"original_sha256": "ef5726e0fa911c30c412c7f7cbe08181541c85351260634745e65e0d147b7ab4",
|
| 470 |
-
"generation_or_verifier_rerun": false
|
| 471 |
-
},
|
| 472 |
-
"pytorch-model-recovery--attempt-001": {
|
| 473 |
-
"timestamp": "2026-10-29T19:11:17.511954+00:00",
|
| 474 |
"reward": 0.0,
|
| 475 |
"native_reward": null,
|
| 476 |
"original_verifier_result": null,
|
| 477 |
-
"
|
| 478 |
-
"
|
| 479 |
-
"
|
| 480 |
-
"
|
| 481 |
-
"
|
| 482 |
-
"
|
| 483 |
-
"
|
| 484 |
},
|
| 485 |
"pytorch-model-recovery--attempt-002": {
|
| 486 |
-
"timestamp": "2026-
|
| 487 |
-
"
|
| 488 |
-
"native_reward": null,
|
| 489 |
-
"original_verifier_result": null,
|
| 490 |
-
"method": "explicit_user_authorized_zero_for_unscored_attempt",
|
| 491 |
"authorization": "[REDACTED_CREDENTIAL]",
|
| 492 |
-
"reason": "Original attempt ended without a native verifier reward; this is an adjudicated zero, not a completed native test verdict.",
|
| 493 |
-
"exception_type": "AgentTimeoutError",
|
| 494 |
-
"backup": "/workspace/data/sci-llm-training/runtime/nemotron35-lightning-full-suite-20261029/operations/user-zero-adjudication-20261029/terminal-bench-2.1--pytorch-model-recovery--attempt-002",
|
| 495 |
-
"original_sha256": "7fae00d6c6402beb1764a220d36240d2234c301adbcb73a2386afc5c69e971a5",
|
| 496 |
-
"generation_or_verifier_rerun": false
|
| 497 |
-
},
|
| 498 |
-
"torch-pipeline-parallelism--attempt-002": {
|
| 499 |
-
"timestamp": "2026-10-29T19:11:17.623946+00:00",
|
| 500 |
"reward": 0.0,
|
| 501 |
"native_reward": null,
|
| 502 |
"original_verifier_result": null,
|
| 503 |
-
"
|
| 504 |
-
"
|
| 505 |
-
"
|
| 506 |
-
"
|
| 507 |
-
"
|
| 508 |
-
"
|
| 509 |
-
"
|
| 510 |
}
|
| 511 |
-
}
|
| 512 |
-
"native_scored": 174,
|
| 513 |
-
"adjudicated_zero": 4,
|
| 514 |
-
"scoring_note": "Four missing native verdicts assigned zero by explicit user instruction; original exceptions and null rewards preserved in native_artifacts.tar.gz and recovery_provenance.tar.gz."
|
| 515 |
},
|
| 516 |
"hle-verified": {
|
| 517 |
"gold": {
|
|
|
|
| 431 |
"attempts": 100,
|
| 432 |
"attempts_per_task": 2,
|
| 433 |
"metric": "avg@2",
|
| 434 |
+
"passed": 19,
|
| 435 |
+
"avg_at_2": 0.19,
|
| 436 |
"unscored": 0,
|
| 437 |
+
"agent_timeouts": 12,
|
| 438 |
+
"reward_positive_with_exception": 0,
|
| 439 |
"scoring": "native_verifier",
|
| 440 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
| 441 |
+
"native_scored": 100,
|
| 442 |
+
"adjudicated_zero": 0,
|
| 443 |
+
"grading_complete": true,
|
| 444 |
"provisional": false,
|
| 445 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run",
|
| 446 |
+
"sampling": {
|
| 447 |
+
"temperature": 1.0,
|
| 448 |
+
"top_p": 0.95,
|
| 449 |
+
"top_k": 20
|
| 450 |
+
},
|
| 451 |
+
"hardware": "H100",
|
| 452 |
+
"tasks_per_engine": 2,
|
| 453 |
+
"scoring_adjudications": {}
|
| 454 |
},
|
| 455 |
"terminal-bench-2.1": {
|
| 456 |
"tasks": 89,
|
| 457 |
"attempts": 178,
|
| 458 |
"attempts_per_task": 2,
|
| 459 |
"metric": "avg@2",
|
| 460 |
+
"passed": 35,
|
| 461 |
+
"avg_at_2": 0.19662921348314608,
|
| 462 |
"unscored": 0,
|
| 463 |
+
"agent_timeouts": 57,
|
| 464 |
+
"reward_positive_with_exception": 0,
|
| 465 |
+
"scoring": "native_verifier_with_explicit_user_adjudication",
|
| 466 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
| 467 |
+
"native_scored": 176,
|
| 468 |
+
"adjudicated_zero": 2,
|
| 469 |
+
"grading_complete": true,
|
| 470 |
"provisional": false,
|
| 471 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run",
|
| 472 |
+
"sampling": {
|
| 473 |
+
"temperature": 1.0,
|
| 474 |
+
"top_p": 0.95,
|
| 475 |
+
"top_k": 20
|
| 476 |
+
},
|
| 477 |
+
"hardware": "H100",
|
| 478 |
+
"tasks_per_engine": 2,
|
| 479 |
"scoring_adjudications": {
|
| 480 |
"mailman--attempt-002": {
|
| 481 |
+
"timestamp": "2026-11-04T14:50:07.041310+00:00",
|
| 482 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
|
|
|
|
|
|
|
|
|
| 483 |
"authorization": "[REDACTED_CREDENTIAL]",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 484 |
"reward": 0.0,
|
| 485 |
"native_reward": null,
|
| 486 |
"original_verifier_result": null,
|
| 487 |
+
"original_exception_type": "AgentTimeoutError",
|
| 488 |
+
"verifier_stage": "tests_started",
|
| 489 |
+
"timeout_evidence": "Trial mailman__a7D3i4S failed: Verifier execution timed out after 1800.0 seconds",
|
| 490 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 491 |
+
"generation_or_verifier_rerun": false,
|
| 492 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--mailman--attempt-002",
|
| 493 |
+
"original_result_sha256": "23e2436ef457f7d4568691c3af19ee69c1e516cc7fd2dc13409b09a360836204"
|
| 494 |
},
|
| 495 |
"pytorch-model-recovery--attempt-002": {
|
| 496 |
+
"timestamp": "2026-11-04T14:50:07.080752+00:00",
|
| 497 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
|
|
|
|
|
|
|
|
|
| 498 |
"authorization": "[REDACTED_CREDENTIAL]",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 499 |
"reward": 0.0,
|
| 500 |
"native_reward": null,
|
| 501 |
"original_verifier_result": null,
|
| 502 |
+
"original_exception_type": "AgentTimeoutError",
|
| 503 |
+
"verifier_stage": "setup",
|
| 504 |
+
"timeout_evidence": "Trial pytorch-model-recovery__2VTmgHe failed: Verifier execution timed out after 900.0 seconds",
|
| 505 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 506 |
+
"generation_or_verifier_rerun": false,
|
| 507 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--pytorch-model-recovery--attempt-002",
|
| 508 |
+
"original_result_sha256": "357cd361d362020796c1c29f1a67cd06adac685d7e295ecb7d46c5feb7d80693"
|
| 509 |
}
|
| 510 |
+
}
|
|
|
|
|
|
|
|
|
|
| 511 |
},
|
| 512 |
"hle-verified": {
|
| 513 |
"gold": {
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/README.md
CHANGED
|
@@ -1,3 +1,7 @@
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 — terminal-bench-2.1
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 — terminal-bench-2.1
|
| 2 |
|
| 3 |
+
35/178 = 19.66% avg@2. Native scores: 176; explicitly user-adjudicated timeout zeros: 2; unscored: 0.
|
| 4 |
+
|
| 5 |
+
Aligned rerun: H100, Terminus 2, temperature 1.0, top-p 0.95, top-k 20, two active tasks per engine, two independent attempts per task. Exact model-specific serving, reasoning, compaction and timeout settings are frozen in configs.json.
|
| 6 |
+
|
| 7 |
+
trajectories.jsonl includes reasoning, tool exchanges and compaction continuation chains; judge_trajectories.jsonl contains native verifier outputs. native_artifacts.tar.gz preserves the available complete native trial artifacts. Timeout adjudications are not completed native verifier verdicts; original null rewards and evidence are preserved in scoring_adjudications.tar.gz where applicable. Every planned attempt remains in the denominator.
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/configs.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/judge_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 sha256:863c67aa40e75c22cf670b4c233832118940a2211e1c01f07631538b8098263c
|
| 3 |
+
size 15471933
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/native_artifacts.tar.gz
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 sha256:1c8df8a1f222a7c7644c744f08e023c1c22003e508788aa757903ab1981b2336
|
| 3 |
+
size 842664153
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/provenance.json
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run",
|
| 3 |
+
"supersedes_hf_revision": "5da71d89fedc2a5e6c7c36c9883ed86d1d09f020",
|
| 4 |
+
"regenerated_for_publication": false,
|
| 5 |
+
"rejudged_for_publication": false,
|
| 6 |
+
"adjudications": {
|
| 7 |
+
"mailman--attempt-002": {
|
| 8 |
+
"timestamp": "2026-11-04T14:50:07.041310+00:00",
|
| 9 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 10 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 11 |
+
"reward": 0.0,
|
| 12 |
+
"native_reward": null,
|
| 13 |
+
"original_verifier_result": null,
|
| 14 |
+
"original_exception_type": "AgentTimeoutError",
|
| 15 |
+
"verifier_stage": "tests_started",
|
| 16 |
+
"timeout_evidence": "Trial mailman__a7D3i4S failed: Verifier execution timed out after 1800.0 seconds",
|
| 17 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 18 |
+
"generation_or_verifier_rerun": false,
|
| 19 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--mailman--attempt-002",
|
| 20 |
+
"original_result_sha256": "23e2436ef457f7d4568691c3af19ee69c1e516cc7fd2dc13409b09a360836204"
|
| 21 |
+
},
|
| 22 |
+
"pytorch-model-recovery--attempt-002": {
|
| 23 |
+
"timestamp": "2026-11-04T14:50:07.080752+00:00",
|
| 24 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 25 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 26 |
+
"reward": 0.0,
|
| 27 |
+
"native_reward": null,
|
| 28 |
+
"original_verifier_result": null,
|
| 29 |
+
"original_exception_type": "AgentTimeoutError",
|
| 30 |
+
"verifier_stage": "setup",
|
| 31 |
+
"timeout_evidence": "Trial pytorch-model-recovery__2VTmgHe failed: Verifier execution timed out after 900.0 seconds",
|
| 32 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 33 |
+
"generation_or_verifier_rerun": false,
|
| 34 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--pytorch-model-recovery--attempt-002",
|
| 35 |
+
"original_result_sha256": "357cd361d362020796c1c29f1a67cd06adac685d7e295ecb7d46c5feb7d80693"
|
| 36 |
+
}
|
| 37 |
+
},
|
| 38 |
+
"native_scores_are_distinct_from_adjudicated_zeros": true
|
| 39 |
+
}
|
sft-agentic-stage2/terminal-bench-2.1/recovery_original_attempts.tar.gz → NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/scoring_adjudications.tar.gz
RENAMED
|
@@ -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:cd0c89af900e55b2da3c81a2f04aa0807adc37b7dae1a5c666ffe07de3a35fb8
|
| 3 |
+
size 61565
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/terminal-bench-2.1/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed":
|
|
|
|
| 1 |
+
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 35, "avg_at_2": 0.19662921348314608, "unscored": 0, "agent_timeouts": 57, "reward_positive_with_exception": 0, "scoring": "native_verifier_with_explicit_user_adjudication", "scoring_policy": "native-verifier-after-agent-stop-v1", "native_scored": 176, "adjudicated_zero": 2, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/run", "sampling": {"temperature": 1.0, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "scoring_adjudications": {"mailman--attempt-002": {"timestamp": "2026-11-04T14:50:07.041310+00:00", "method": "user_authorized_zero_for_verifier_timeout", "authorization": "[REDACTED_CREDENTIAL]", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_type": "AgentTimeoutError", "verifier_stage": "tests_started", "timeout_evidence": "Trial mailman__a7D3i4S failed: Verifier execution timed out after 1800.0 seconds", "reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.", "generation_or_verifier_rerun": false, "backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--mailman--attempt-002", "original_result_sha256": "23e2436ef457f7d4568691c3af19ee69c1e516cc7fd2dc13409b09a360836204"}, "pytorch-model-recovery--attempt-002": {"timestamp": "2026-11-04T14:50:07.080752+00:00", "method": "user_authorized_zero_for_verifier_timeout", "authorization": "[REDACTED_CREDENTIAL]", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_type": "AgentTimeoutError", "verifier_stage": "setup", "timeout_evidence": "Trial pytorch-model-recovery__2VTmgHe failed: Verifier execution timed out after 900.0 seconds", "reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.", "generation_or_verifier_rerun": false, "backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/nemotron35/archive/user-timeout-zero-20261104/terminal-bench-2.1--pytorch-model-recovery--attempt-002", "original_result_sha256": "357cd361d362020796c1c29f1a67cd06adac685d7e295ecb7d46c5feb7d80693"}}}
|
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/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:f7cc05e0f31fbb9f72635ff858367848fe2a1f273f8d03b4a520a2081d330ad6
|
| 3 |
+
size 69871030
|
Qwen3.5-35B-A3B/README.md
CHANGED
|
@@ -1,24 +1,25 @@
|
|
| 1 |
# Qwen3.5-35B-A3B
|
| 2 |
|
| 3 |
-
|
| 4 |
|
| 5 |
-
| Benchmark |
|
| 6 |
-
|---|---|
|
| 7 |
-
|
|
| 8 |
-
|
|
| 9 |
-
| arXiv Math | 148/412 = 35.92% |
|
| 10 |
-
| SciCode-Verified | 183/287 steps (63.76%); 31/64 main problems (48.44%) |
|
| 11 |
-
| SciCoding-Agent | 36/100 = 36.00% avg@2 |
|
| 12 |
-
| Terminal-Bench 2.1 | 61/178 = 34.27% avg@2 |
|
| 13 |
|
| 14 |
-
|
| 15 |
|
| 16 |
-
|
| 17 |
|
| 18 |
-
|
| 19 |
|
| 20 |
-
|
| 21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
|
| 23 |
## HLE-Verified (2026-10-30)
|
| 24 |
|
|
|
|
| 1 |
# Qwen3.5-35B-A3B
|
| 2 |
|
| 3 |
+
## Aligned agentic rerun — 2026-11-04
|
| 4 |
|
| 5 |
+
| Benchmark | avg@2 | User-adjudicated timeout zeros |
|
| 6 |
+
|---|---:|---:|
|
| 7 |
+
| scicoding-agent | 41/100 (41.00%) | 0 |
|
| 8 |
+
| terminal-bench-2.1 | 62/178 (34.83%) | 3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
+
H100; Terminus 2; temperature 1.0, top-p 0.95, top-k 20; concurrency 2 per engine. TP1 for GPT-OSS-20B; TP2 for the other models. Two independent attempts per task. All attempts are scored. Timeout zeros preserve their original native null verdict and evidence; they are not native test failures.
|
| 11 |
|
| 12 |
+
These benchmark folders replace the older inconsistent agentic exports. Complete trajectories, continuation chains, verifier artifacts and frozen configs are included. Older exports remain recoverable in repository revision `55136ae404538e1ebb3dfaba6261b529109883e9`.
|
| 13 |
|
| 14 |
+
## Unchanged non-agentic evaluations
|
| 15 |
|
| 16 |
+
These results are preserved from the latest pre-publication revision, not rolled back to the original export date.
|
| 17 |
|
| 18 |
+
- [arxivmath](arxivmath/summary.json)
|
| 19 |
+
- [frontierscience-olympiad](frontierscience-olympiad/summary.json)
|
| 20 |
+
- [frontierscience-research](frontierscience-research/summary.json)
|
| 21 |
+
- [scicode-verified](scicode-verified/summary.json)
|
| 22 |
+
- [hle-verified](hle-verified/summary.json)
|
| 23 |
|
| 24 |
## HLE-Verified (2026-10-30)
|
| 25 |
|
Qwen3.5-35B-A3B/manifest.json
CHANGED
|
@@ -1 +1,154 @@
|
|
| 1 |
-
{
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"README.md": {
|
| 3 |
+
"size": 2381,
|
| 4 |
+
"sha256": "6b34d85eeb43fffa77eb12a7683b406a8eb3715f2e577e8617d3b65d7b3387c3"
|
| 5 |
+
},
|
| 6 |
+
"arxivmath/README.md": {
|
| 7 |
+
"size": 283,
|
| 8 |
+
"sha256": "fbc08a92971edfb3dbfce8873e64288ff926c1570801ad0076cc12238a938c03"
|
| 9 |
+
},
|
| 10 |
+
"arxivmath/events.jsonl.gz": {
|
| 11 |
+
"size": 45507013,
|
| 12 |
+
"sha256": "06637e4af4da6937c321e6f8dbc7ef2bec75dd5b0624e1ad780f0a646b63457b"
|
| 13 |
+
},
|
| 14 |
+
"arxivmath/judge_trajectories.jsonl": {
|
| 15 |
+
"size": 209839,
|
| 16 |
+
"sha256": "c09da06cd077975f6b8b30587e0873916e543f717803203b1f8c4722d922d1b5"
|
| 17 |
+
},
|
| 18 |
+
"arxivmath/summary.json": {
|
| 19 |
+
"size": 6864,
|
| 20 |
+
"sha256": "fae7d9750f876d99f8768424bd6c88c9187012b35b9b60eea004a53e9b2984a4"
|
| 21 |
+
},
|
| 22 |
+
"arxivmath/trajectories.jsonl": {
|
| 23 |
+
"size": 68307168,
|
| 24 |
+
"sha256": "82e2dbd2309f509aa6f5934caef3e7c445ca63179df9f2a251b41bbefbefbff7"
|
| 25 |
+
},
|
| 26 |
+
"frontierscience-olympiad/README.md": {
|
| 27 |
+
"size": 298,
|
| 28 |
+
"sha256": "65c4a593da8583b4a84a2fb52bde30d93206078e879a67c458cb71b78dd8000f"
|
| 29 |
+
},
|
| 30 |
+
"frontierscience-olympiad/events.jsonl.gz": {
|
| 31 |
+
"size": 221366393,
|
| 32 |
+
"sha256": "0cf0b2749b2eb12861b3ca721396802dbaf1996e4bdc7945b8c9866fd2ed3eb6"
|
| 33 |
+
},
|
| 34 |
+
"frontierscience-olympiad/judge_trajectories.jsonl": {
|
| 35 |
+
"size": 11199188,
|
| 36 |
+
"sha256": "c7a9a5c5f5acdc1cfa403d75380c41aa39209a447776e3f62af32b50b6d55e51"
|
| 37 |
+
},
|
| 38 |
+
"frontierscience-olympiad/summary.json": {
|
| 39 |
+
"size": 1816,
|
| 40 |
+
"sha256": "fa932c961ef4d5d10e70287179aa27ecdffb56265cd5d8690d1fb434299537c8"
|
| 41 |
+
},
|
| 42 |
+
"frontierscience-olympiad/trajectories.jsonl": {
|
| 43 |
+
"size": 415242177,
|
| 44 |
+
"sha256": "8e957f49fde64012c4342c2a4f9df30a90a5dc49f403ef71e273b2ef12c28c09"
|
| 45 |
+
},
|
| 46 |
+
"frontierscience-research/README.md": {
|
| 47 |
+
"size": 298,
|
| 48 |
+
"sha256": "bf03e2df51aeaef984f808121ea8cc989a54f7408ad76439b5f1c2c67cf37aa7"
|
| 49 |
+
},
|
| 50 |
+
"frontierscience-research/events.jsonl.gz": {
|
| 51 |
+
"size": 84174734,
|
| 52 |
+
"sha256": "1683f466afe4236fdc7eb9d3ed3ae4d1a197e0dfb003a4bbe0fb7345f87a35e5"
|
| 53 |
+
},
|
| 54 |
+
"frontierscience-research/judge_trajectories.jsonl": {
|
| 55 |
+
"size": 27240844,
|
| 56 |
+
"sha256": "24bae13ca7c3127086e67dac0d0338b28f25c4d47d1343d95b7f2cd7fa76ab00"
|
| 57 |
+
},
|
| 58 |
+
"frontierscience-research/summary.json": {
|
| 59 |
+
"size": 1814,
|
| 60 |
+
"sha256": "cb8f52cb5e843d812737cea079e15059ea9b8fe0c0af4781eeedc0865e60d746"
|
| 61 |
+
},
|
| 62 |
+
"frontierscience-research/trajectories.jsonl": {
|
| 63 |
+
"size": 134991835,
|
| 64 |
+
"sha256": "2f7a4bfa1acdfd6df1bbb76d10dab0cfeafd7d2213361e8801b73d6894dc2e8e"
|
| 65 |
+
},
|
| 66 |
+
"provenance.json": {
|
| 67 |
+
"size": 458887,
|
| 68 |
+
"sha256": "7dd10d58597a39872eb26e5fa8bddbbff37a02d909c1c7fd049c0900bae9d430"
|
| 69 |
+
},
|
| 70 |
+
"scicode-verified/README.md": {
|
| 71 |
+
"size": 290,
|
| 72 |
+
"sha256": "a6ca43c681dd439800fa6c2351ec11097a84a1e712b649cfd8a98a6c3f8e825e"
|
| 73 |
+
},
|
| 74 |
+
"scicode-verified/judge_trajectories.jsonl": {
|
| 75 |
+
"size": 257330,
|
| 76 |
+
"sha256": "179110ff78d7d8031d2240a901b4ae8ca59c28b8ecb4d32f38f1500e9d529afa"
|
| 77 |
+
},
|
| 78 |
+
"scicode-verified/native_artifacts.tar.gz": {
|
| 79 |
+
"size": 4560090,
|
| 80 |
+
"sha256": "2f5b28ed78e7987373d0cc5dbc7d86056cee1dcb60feae894f300dc5fbd294d6"
|
| 81 |
+
},
|
| 82 |
+
"scicode-verified/summary.json": {
|
| 83 |
+
"size": 515,
|
| 84 |
+
"sha256": "ca97e85e8ebc69ec8d1a1066ff8863315dd5548979acd3484e57e2b7adf483d7"
|
| 85 |
+
},
|
| 86 |
+
"scicode-verified/trajectories.jsonl": {
|
| 87 |
+
"size": 12209668,
|
| 88 |
+
"sha256": "c79e5f2809badd82c5905a29f762c507bcf6eba7fda6510f864805b2f8cac12a"
|
| 89 |
+
},
|
| 90 |
+
"summary.json": {
|
| 91 |
+
"size": 19432,
|
| 92 |
+
"sha256": "abb77ead5303bb72b2cbf50fc77fff4462bcb35598950f20868511615b342f52"
|
| 93 |
+
},
|
| 94 |
+
"scicoding-agent/README.md": {
|
| 95 |
+
"size": 825,
|
| 96 |
+
"sha256": "f2f51d98f49eaa9660f1bc9da282f13468c581bfde8dc2f9b3a2479ec4f83c6d"
|
| 97 |
+
},
|
| 98 |
+
"scicoding-agent/judge_trajectories.jsonl": {
|
| 99 |
+
"size": 197921,
|
| 100 |
+
"sha256": "148d248bcf908c387c0c54f7b7444242c67c35f28d2ce5da2db178f5fdb235cf"
|
| 101 |
+
},
|
| 102 |
+
"scicoding-agent/native_artifacts.tar.gz": {
|
| 103 |
+
"size": 9589060,
|
| 104 |
+
"sha256": "bcf7dc01e20d600ee4ec9cafbce675137587c833c74b3081f2a9d24f6ec82e5a"
|
| 105 |
+
},
|
| 106 |
+
"scicoding-agent/summary.json": {
|
| 107 |
+
"size": 598,
|
| 108 |
+
"sha256": "86c659b3ee830da3a5a689e884b1e1664dc8afc0d45ab39455ba9e43e3883080"
|
| 109 |
+
},
|
| 110 |
+
"scicoding-agent/trajectories.jsonl": {
|
| 111 |
+
"size": 31867666,
|
| 112 |
+
"sha256": "ff7503e444f39fce214901e89da80744f4f509157c21edee1fc5542b884db69a"
|
| 113 |
+
},
|
| 114 |
+
"terminal-bench-2.1/README.md": {
|
| 115 |
+
"size": 828,
|
| 116 |
+
"sha256": "b8720cf50c6a7b7f9203bf5c30fd6966ae91cbba8e0c50a950aa6958e9d198ac"
|
| 117 |
+
},
|
| 118 |
+
"terminal-bench-2.1/judge_trajectories.jsonl": {
|
| 119 |
+
"size": 10064312,
|
| 120 |
+
"sha256": "d5980a13c054ff490d9b1ad277bea5d277c8706fa87a4f2e9f8d5ea497c93fe8"
|
| 121 |
+
},
|
| 122 |
+
"terminal-bench-2.1/native_artifacts.tar.gz": {
|
| 123 |
+
"size": 128430903,
|
| 124 |
+
"sha256": "b996d9c97eff15000f9121ab745d0bd38d07591a7fcd867a85ecfb34e0b69fba"
|
| 125 |
+
},
|
| 126 |
+
"terminal-bench-2.1/summary.json": {
|
| 127 |
+
"size": 3339,
|
| 128 |
+
"sha256": "f638a007f1543423f51cfa6ccd3261a2e7a936f0a6c774b57ba2b6e1effdfc0b"
|
| 129 |
+
},
|
| 130 |
+
"terminal-bench-2.1/trajectories.jsonl": {
|
| 131 |
+
"size": 59085851,
|
| 132 |
+
"sha256": "03ffe164de34b51bc9a3880585aa3428fef5b088a20fa5188cd116c83d42bbf0"
|
| 133 |
+
},
|
| 134 |
+
"scicoding-agent/configs.json": {
|
| 135 |
+
"size": 242809,
|
| 136 |
+
"sha256": "0d10d846fe71098308c8a427d944b9e1b5d65d0aaf2436abc27ca2b54aceb1a6"
|
| 137 |
+
},
|
| 138 |
+
"scicoding-agent/provenance.json": {
|
| 139 |
+
"size": 343,
|
| 140 |
+
"sha256": "4b5775f675092dc04d29892a38b69d0a727ddbc52d512a4caabdc83214d7bafc"
|
| 141 |
+
},
|
| 142 |
+
"terminal-bench-2.1/configs.json": {
|
| 143 |
+
"size": 242587,
|
| 144 |
+
"sha256": "d709d3908a1c7b7ab3b58c37e07655bd899d285b80d0cb9f4103f730f3fe0743"
|
| 145 |
+
},
|
| 146 |
+
"terminal-bench-2.1/provenance.json": {
|
| 147 |
+
"size": 3304,
|
| 148 |
+
"sha256": "0cd8f348442e52bbeac7e21cbb31577ff1912928e03be07714f17da632e1494a"
|
| 149 |
+
},
|
| 150 |
+
"terminal-bench-2.1/scoring_adjudications.tar.gz": {
|
| 151 |
+
"size": 69984,
|
| 152 |
+
"sha256": "a10f1f0f02dd9e08fc4754a5f25d933557312a0af377bc10f607510e95e6bca6"
|
| 153 |
+
}
|
| 154 |
+
}
|
Qwen3.5-35B-A3B/provenance.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
Qwen3.5-35B-A3B/scicoding-agent/README.md
CHANGED
|
@@ -1,3 +1,7 @@
|
|
| 1 |
# Qwen3.5-35B-A3B — scicoding-agent
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Qwen3.5-35B-A3B — scicoding-agent
|
| 2 |
|
| 3 |
+
41/100 = 41.00% avg@2. Native scores: 100; explicitly user-adjudicated timeout zeros: 0; unscored: 0.
|
| 4 |
+
|
| 5 |
+
Aligned rerun: H100, Terminus 2, temperature 1.0, top-p 0.95, top-k 20, two active tasks per engine, two independent attempts per task. Exact model-specific serving, reasoning, compaction and timeout settings are frozen in configs.json.
|
| 6 |
+
|
| 7 |
+
trajectories.jsonl includes reasoning, tool exchanges and compaction continuation chains; judge_trajectories.jsonl contains native verifier outputs. native_artifacts.tar.gz preserves the available complete native trial artifacts. Timeout adjudications are not completed native verifier verdicts; original null rewards and evidence are preserved in scoring_adjudications.tar.gz where applicable. Every planned attempt remains in the denominator.
|
Qwen3.5-35B-A3B/scicoding-agent/configs.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
Qwen3.5-35B-A3B/scicoding-agent/judge_trajectories.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
Qwen3.5-35B-A3B/scicoding-agent/native_artifacts.tar.gz
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:bcf7dc01e20d600ee4ec9cafbce675137587c833c74b3081f2a9d24f6ec82e5a
|
| 3 |
+
size 9589060
|
Qwen3.5-35B-A3B/scicoding-agent/provenance.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run",
|
| 3 |
+
"supersedes_hf_revision": "5da71d89fedc2a5e6c7c36c9883ed86d1d09f020",
|
| 4 |
+
"regenerated_for_publication": false,
|
| 5 |
+
"rejudged_for_publication": false,
|
| 6 |
+
"adjudications": {},
|
| 7 |
+
"native_scores_are_distinct_from_adjudicated_zeros": true
|
| 8 |
+
}
|
Qwen3.5-35B-A3B/scicoding-agent/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed":
|
|
|
|
| 1 |
+
{"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 41, "avg_at_2": 0.41, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 0, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "native_scored": 100, "adjudicated_zero": 0, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run", "sampling": {"temperature": 1.0, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "scoring_adjudications": {}}
|
Qwen3.5-35B-A3B/scicoding-agent/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:ff7503e444f39fce214901e89da80744f4f509157c21edee1fc5542b884db69a
|
| 3 |
+
size 31867666
|
Qwen3.5-35B-A3B/summary.json
CHANGED
|
@@ -431,65 +431,98 @@
|
|
| 431 |
"attempts": 178,
|
| 432 |
"attempts_per_task": 2,
|
| 433 |
"metric": "avg@2",
|
| 434 |
-
"passed":
|
| 435 |
-
"avg_at_2": 0.
|
| 436 |
"unscored": 0,
|
| 437 |
-
"agent_timeouts":
|
| 438 |
-
"reward_positive_with_exception":
|
| 439 |
-
"scoring": "
|
| 440 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
| 441 |
-
"
|
|
|
|
| 442 |
"grading_complete": true,
|
| 443 |
-
"
|
| 444 |
-
"
|
| 445 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 446 |
"scoring_adjudications": {
|
| 447 |
-
"
|
| 448 |
-
"timestamp": "2026-
|
|
|
|
|
|
|
| 449 |
"reward": 0.0,
|
| 450 |
"native_reward": null,
|
| 451 |
"original_verifier_result": null,
|
| 452 |
-
"
|
| 453 |
-
"
|
| 454 |
-
"
|
| 455 |
-
"
|
| 456 |
-
"
|
| 457 |
-
"
|
| 458 |
-
"
|
| 459 |
},
|
| 460 |
-
"
|
| 461 |
-
"timestamp": "2026-
|
|
|
|
|
|
|
| 462 |
"reward": 0.0,
|
| 463 |
"native_reward": null,
|
| 464 |
"original_verifier_result": null,
|
| 465 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 466 |
"authorization": "[REDACTED_CREDENTIAL]",
|
| 467 |
-
"
|
| 468 |
-
"
|
| 469 |
-
"
|
| 470 |
-
"
|
| 471 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 472 |
}
|
| 473 |
-
}
|
| 474 |
-
"scoring_note": "Two explicit user-adjudicated zeros, not native test failures. Original null rewards preserved in agentic_recovery_provenance.tar.gz."
|
| 475 |
},
|
| 476 |
"scicoding-agent": {
|
| 477 |
"tasks": 50,
|
| 478 |
"attempts": 100,
|
| 479 |
"attempts_per_task": 2,
|
| 480 |
"metric": "avg@2",
|
| 481 |
-
"passed":
|
| 482 |
-
"avg_at_2": 0.
|
| 483 |
"unscored": 0,
|
| 484 |
-
"agent_timeouts":
|
| 485 |
"reward_positive_with_exception": 0,
|
| 486 |
"scoring": "native_verifier",
|
| 487 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
| 488 |
-
"provisional": false,
|
| 489 |
-
"grading_complete": true,
|
| 490 |
"native_scored": 100,
|
| 491 |
"adjudicated_zero": 0,
|
| 492 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 493 |
},
|
| 494 |
"hle-verified": {
|
| 495 |
"gold": {
|
|
|
|
| 431 |
"attempts": 178,
|
| 432 |
"attempts_per_task": 2,
|
| 433 |
"metric": "avg@2",
|
| 434 |
+
"passed": 62,
|
| 435 |
+
"avg_at_2": 0.34831460674157305,
|
| 436 |
"unscored": 0,
|
| 437 |
+
"agent_timeouts": 53,
|
| 438 |
+
"reward_positive_with_exception": 3,
|
| 439 |
+
"scoring": "native_verifier_with_explicit_user_adjudication",
|
| 440 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
| 441 |
+
"native_scored": 175,
|
| 442 |
+
"adjudicated_zero": 3,
|
| 443 |
"grading_complete": true,
|
| 444 |
+
"provisional": false,
|
| 445 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run",
|
| 446 |
+
"sampling": {
|
| 447 |
+
"temperature": 1.0,
|
| 448 |
+
"top_p": 0.95,
|
| 449 |
+
"top_k": 20
|
| 450 |
+
},
|
| 451 |
+
"hardware": "H100",
|
| 452 |
+
"tasks_per_engine": 2,
|
| 453 |
"scoring_adjudications": {
|
| 454 |
+
"extract-moves-from-video--attempt-001": {
|
| 455 |
+
"timestamp": "2026-11-04T14:50:09.132779+00:00",
|
| 456 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 457 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 458 |
"reward": 0.0,
|
| 459 |
"native_reward": null,
|
| 460 |
"original_verifier_result": null,
|
| 461 |
+
"original_exception_type": "AgentTimeoutError",
|
| 462 |
+
"verifier_stage": "setup",
|
| 463 |
+
"timeout_evidence": "Trial extract-moves-from-video__Ezi83im failed: Verifier execution timed out after 1800.0 seconds",
|
| 464 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 465 |
+
"generation_or_verifier_rerun": false,
|
| 466 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--extract-moves-from-video--attempt-001",
|
| 467 |
+
"original_result_sha256": "254fc7bfe27701cbd6c144a789b0edfd8ed218d9f3534c937710659525644c28"
|
| 468 |
},
|
| 469 |
+
"feal-differential-cryptanalysis--attempt-002": {
|
| 470 |
+
"timestamp": "2026-11-04T14:50:09.171586+00:00",
|
| 471 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 472 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 473 |
"reward": 0.0,
|
| 474 |
"native_reward": null,
|
| 475 |
"original_verifier_result": null,
|
| 476 |
+
"original_exception_type": "AgentTimeoutError",
|
| 477 |
+
"verifier_stage": "tests_started",
|
| 478 |
+
"timeout_evidence": "Trial feal-differential-cryptanalysis__6p5YiS7 failed: Verifier execution timed out after 1800.0 seconds",
|
| 479 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 480 |
+
"generation_or_verifier_rerun": false,
|
| 481 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002",
|
| 482 |
+
"original_result_sha256": "3c75e6d74012ab4f62d691a5fafb7c9742e4c5f6359b1f52b67a5ea024c7404c"
|
| 483 |
+
},
|
| 484 |
+
"winning-avg-corewars--attempt-002": {
|
| 485 |
+
"timestamp": "2026-11-04T14:50:09.209266+00:00",
|
| 486 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 487 |
"authorization": "[REDACTED_CREDENTIAL]",
|
| 488 |
+
"reward": 0.0,
|
| 489 |
+
"native_reward": null,
|
| 490 |
+
"original_verifier_result": null,
|
| 491 |
+
"original_exception_type": "AgentTimeoutError",
|
| 492 |
+
"verifier_stage": "tests_started",
|
| 493 |
+
"timeout_evidence": "Trial winning-avg-corewars__DwjY6rN failed: Verifier execution timed out after 3600.0 seconds",
|
| 494 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 495 |
+
"generation_or_verifier_rerun": false,
|
| 496 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--winning-avg-corewars--attempt-002",
|
| 497 |
+
"original_result_sha256": "598e0762f9cff542abe368a2e28a542fb0292f0399bf2e9d01af7dbdc10e85f7"
|
| 498 |
}
|
| 499 |
+
}
|
|
|
|
| 500 |
},
|
| 501 |
"scicoding-agent": {
|
| 502 |
"tasks": 50,
|
| 503 |
"attempts": 100,
|
| 504 |
"attempts_per_task": 2,
|
| 505 |
"metric": "avg@2",
|
| 506 |
+
"passed": 41,
|
| 507 |
+
"avg_at_2": 0.41,
|
| 508 |
"unscored": 0,
|
| 509 |
+
"agent_timeouts": 0,
|
| 510 |
"reward_positive_with_exception": 0,
|
| 511 |
"scoring": "native_verifier",
|
| 512 |
"scoring_policy": "native-verifier-after-agent-stop-v1",
|
|
|
|
|
|
|
| 513 |
"native_scored": 100,
|
| 514 |
"adjudicated_zero": 0,
|
| 515 |
+
"grading_complete": true,
|
| 516 |
+
"provisional": false,
|
| 517 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run",
|
| 518 |
+
"sampling": {
|
| 519 |
+
"temperature": 1.0,
|
| 520 |
+
"top_p": 0.95,
|
| 521 |
+
"top_k": 20
|
| 522 |
+
},
|
| 523 |
+
"hardware": "H100",
|
| 524 |
+
"tasks_per_engine": 2,
|
| 525 |
+
"scoring_adjudications": {}
|
| 526 |
},
|
| 527 |
"hle-verified": {
|
| 528 |
"gold": {
|
Qwen3.5-35B-A3B/terminal-bench-2.1/README.md
CHANGED
|
@@ -1,3 +1,7 @@
|
|
| 1 |
# Qwen3.5-35B-A3B — terminal-bench-2.1
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Qwen3.5-35B-A3B — terminal-bench-2.1
|
| 2 |
|
| 3 |
+
62/178 = 34.83% avg@2. Native scores: 175; explicitly user-adjudicated timeout zeros: 3; unscored: 0.
|
| 4 |
+
|
| 5 |
+
Aligned rerun: H100, Terminus 2, temperature 1.0, top-p 0.95, top-k 20, two active tasks per engine, two independent attempts per task. Exact model-specific serving, reasoning, compaction and timeout settings are frozen in configs.json.
|
| 6 |
+
|
| 7 |
+
trajectories.jsonl includes reasoning, tool exchanges and compaction continuation chains; judge_trajectories.jsonl contains native verifier outputs. native_artifacts.tar.gz preserves the available complete native trial artifacts. Timeout adjudications are not completed native verifier verdicts; original null rewards and evidence are preserved in scoring_adjudications.tar.gz where applicable. Every planned attempt remains in the denominator.
|
Qwen3.5-35B-A3B/terminal-bench-2.1/configs.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
Qwen3.5-35B-A3B/terminal-bench-2.1/judge_trajectories.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
Qwen3.5-35B-A3B/terminal-bench-2.1/native_artifacts.tar.gz
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:b996d9c97eff15000f9121ab745d0bd38d07591a7fcd867a85ecfb34e0b69fba
|
| 3 |
+
size 128430903
|
Qwen3.5-35B-A3B/terminal-bench-2.1/provenance.json
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run",
|
| 3 |
+
"supersedes_hf_revision": "5da71d89fedc2a5e6c7c36c9883ed86d1d09f020",
|
| 4 |
+
"regenerated_for_publication": false,
|
| 5 |
+
"rejudged_for_publication": false,
|
| 6 |
+
"adjudications": {
|
| 7 |
+
"extract-moves-from-video--attempt-001": {
|
| 8 |
+
"timestamp": "2026-11-04T14:50:09.132779+00:00",
|
| 9 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 10 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 11 |
+
"reward": 0.0,
|
| 12 |
+
"native_reward": null,
|
| 13 |
+
"original_verifier_result": null,
|
| 14 |
+
"original_exception_type": "AgentTimeoutError",
|
| 15 |
+
"verifier_stage": "setup",
|
| 16 |
+
"timeout_evidence": "Trial extract-moves-from-video__Ezi83im failed: Verifier execution timed out after 1800.0 seconds",
|
| 17 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 18 |
+
"generation_or_verifier_rerun": false,
|
| 19 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--extract-moves-from-video--attempt-001",
|
| 20 |
+
"original_result_sha256": "254fc7bfe27701cbd6c144a789b0edfd8ed218d9f3534c937710659525644c28"
|
| 21 |
+
},
|
| 22 |
+
"feal-differential-cryptanalysis--attempt-002": {
|
| 23 |
+
"timestamp": "2026-11-04T14:50:09.171586+00:00",
|
| 24 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 25 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 26 |
+
"reward": 0.0,
|
| 27 |
+
"native_reward": null,
|
| 28 |
+
"original_verifier_result": null,
|
| 29 |
+
"original_exception_type": "AgentTimeoutError",
|
| 30 |
+
"verifier_stage": "tests_started",
|
| 31 |
+
"timeout_evidence": "Trial feal-differential-cryptanalysis__6p5YiS7 failed: Verifier execution timed out after 1800.0 seconds",
|
| 32 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 33 |
+
"generation_or_verifier_rerun": false,
|
| 34 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002",
|
| 35 |
+
"original_result_sha256": "3c75e6d74012ab4f62d691a5fafb7c9742e4c5f6359b1f52b67a5ea024c7404c"
|
| 36 |
+
},
|
| 37 |
+
"winning-avg-corewars--attempt-002": {
|
| 38 |
+
"timestamp": "2026-11-04T14:50:09.209266+00:00",
|
| 39 |
+
"method": "user_authorized_zero_for_verifier_timeout",
|
| 40 |
+
"authorization": "[REDACTED_CREDENTIAL]",
|
| 41 |
+
"reward": 0.0,
|
| 42 |
+
"native_reward": null,
|
| 43 |
+
"original_verifier_result": null,
|
| 44 |
+
"original_exception_type": "AgentTimeoutError",
|
| 45 |
+
"verifier_stage": "tests_started",
|
| 46 |
+
"timeout_evidence": "Trial winning-avg-corewars__DwjY6rN failed: Verifier execution timed out after 3600.0 seconds",
|
| 47 |
+
"reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.",
|
| 48 |
+
"generation_or_verifier_rerun": false,
|
| 49 |
+
"backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--winning-avg-corewars--attempt-002",
|
| 50 |
+
"original_result_sha256": "598e0762f9cff542abe368a2e28a542fb0292f0399bf2e9d01af7dbdc10e85f7"
|
| 51 |
+
}
|
| 52 |
+
},
|
| 53 |
+
"native_scores_are_distinct_from_adjudicated_zeros": true
|
| 54 |
+
}
|
Qwen3.5-35B-A3B/{agentic_recovery_provenance.tar.gz → terminal-bench-2.1/scoring_adjudications.tar.gz}
RENAMED
|
@@ -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:a10f1f0f02dd9e08fc4754a5f25d933557312a0af377bc10f607510e95e6bca6
|
| 3 |
+
size 69984
|
Qwen3.5-35B-A3B/terminal-bench-2.1/summary.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed":
|
|
|
|
| 1 |
+
{"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 62, "avg_at_2": 0.34831460674157305, "unscored": 0, "agent_timeouts": 53, "reward_positive_with_exception": 3, "scoring": "native_verifier_with_explicit_user_adjudication", "scoring_policy": "native-verifier-after-agent-stop-v1", "native_scored": 175, "adjudicated_zero": 3, "grading_complete": true, "provisional": false, "source_run": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/run", "sampling": {"temperature": 1.0, "top_p": 0.95, "top_k": 20}, "hardware": "H100", "tasks_per_engine": 2, "scoring_adjudications": {"extract-moves-from-video--attempt-001": {"timestamp": "2026-11-04T14:50:09.132779+00:00", "method": "user_authorized_zero_for_verifier_timeout", "authorization": "[REDACTED_CREDENTIAL]", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_type": "AgentTimeoutError", "verifier_stage": "setup", "timeout_evidence": "Trial extract-moves-from-video__Ezi83im failed: Verifier execution timed out after 1800.0 seconds", "reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.", "generation_or_verifier_rerun": false, "backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--extract-moves-from-video--attempt-001", "original_result_sha256": "254fc7bfe27701cbd6c144a789b0edfd8ed218d9f3534c937710659525644c28"}, "feal-differential-cryptanalysis--attempt-002": {"timestamp": "2026-11-04T14:50:09.171586+00:00", "method": "user_authorized_zero_for_verifier_timeout", "authorization": "[REDACTED_CREDENTIAL]", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_type": "AgentTimeoutError", "verifier_stage": "tests_started", "timeout_evidence": "Trial feal-differential-cryptanalysis__6p5YiS7 failed: Verifier execution timed out after 1800.0 seconds", "reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.", "generation_or_verifier_rerun": false, "backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002", "original_result_sha256": "3c75e6d74012ab4f62d691a5fafb7c9742e4c5f6359b1f52b67a5ea024c7404c"}, "winning-avg-corewars--attempt-002": {"timestamp": "2026-11-04T14:50:09.209266+00:00", "method": "user_authorized_zero_for_verifier_timeout", "authorization": "[REDACTED_CREDENTIAL]", "reward": 0.0, "native_reward": null, "original_verifier_result": null, "original_exception_type": "AgentTimeoutError", "verifier_stage": "tests_started", "timeout_evidence": "Trial winning-avg-corewars__DwjY6rN failed: Verifier execution timed out after 3600.0 seconds", "reason": "Explicit user timeout policy, including setup timeouts; not a completed native verifier verdict.", "generation_or_verifier_rerun": false, "backup": "/workspace/data/sci-llm-training/runtime/agentic-h100-alignment-20261104/qwen35-35b/archive/user-timeout-zero-20261104/terminal-bench-2.1--winning-avg-corewars--attempt-002", "original_result_sha256": "598e0762f9cff542abe368a2e28a542fb0292f0399bf2e9d01af7dbdc10e85f7"}}}
|
Qwen3.5-35B-A3B/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 sha256:03ffe164de34b51bc9a3880585aa3428fef5b088a20fa5188cd116c83d42bbf0
|
| 3 |
+
size 59085851
|
README.md
CHANGED
|
@@ -1,3 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Science evaluation results
|
| 2 |
|
| 3 |
Results are organized by model and benchmark. See each model README for protocol and completeness caveats.
|
|
|
|
| 1 |
+
|
| 2 |
+
|
| 3 |
+
## Current aligned agentic reruns — published 2026-11-08
|
| 4 |
+
|
| 5 |
+
These eight benchmark folders supersede their older exports and any historical numbers below. Full trajectories, verifier outputs, native artifacts and effective native configurations are included. No new generation or scoring was performed for publication.
|
| 6 |
+
|
| 7 |
+
| Model | SciCoding-Agent avg@2 | Terminal-Bench 2.1 avg@2 |
|
| 8 |
+
|---|---:|---:|
|
| 9 |
+
| [sft-agentic-stage2](sft-agentic-stage2) | 34/100 (34.00%) | 49/178 (27.53%) |
|
| 10 |
+
| [NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16](NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16) | 19/100 (19.00%) | 35/178 (19.66%) |
|
| 11 |
+
| [Qwen3.5-35B-A3B](Qwen3.5-35B-A3B) | 41/100 (41.00%) | 62/178 (34.83%) |
|
| 12 |
+
| [gpt-oss-20b](gpt-oss-20b) | 22/100 (22.00%) | 12/178 (6.74%) |
|
| 13 |
+
|
| 14 |
+
All 1,112 attempts are scored: 1,103 native scores and nine explicitly adjudicated timeout zeros. The metric remains verifier pass; the proposed additional normal-completion requirement is NOT applied here.
|
| 15 |
+
|
| 16 |
+
The reruns use H100, Terminus 2, temperature 1.0, top-p 0.95, top-k 20, concurrency 2, avg@2. TP is model-specific (GPT-OSS-20B TP1; others TP2). Read [AGENTIC_CONFIG_ALIGNMENT.md](AGENTIC_CONFIG_ALIGNMENT.md) for remaining historical execution differences and the Luna exception. Do not claim that every historical run has an identical full configuration.
|
| 17 |
+
|
| 18 |
+
## Previously published repository documentation
|
| 19 |
+
|
| 20 |
+
The following is retained for other benchmarks and provenance. For the four models' agentic results, use the current table and benchmark folders above.
|
| 21 |
# Science evaluation results
|
| 22 |
|
| 23 |
Results are organized by model and benchmark. See each model README for protocol and completeness caveats.
|
gpt-oss-20b/README.md
CHANGED
|
@@ -1,30 +1,25 @@
|
|
| 1 |
-
# gpt-oss-20b
|
| 2 |
|
| 3 |
-
|
| 4 |
|
| 5 |
-
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
-
|
| 8 |
|
| 9 |
-
|
| 10 |
|
| 11 |
-
##
|
| 12 |
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
Research mean rubric and trajectory success are separate metrics. Agentic avg@2 averages both attempts, not pass@2. Unscored attempts are retained, explicitly identified, and not silently treated as native failures.
|
| 16 |
-
|
| 17 |
-
## Results
|
| 18 |
-
|
| 19 |
-
| Benchmark | Metric | Result | Scoring |
|
| 20 |
-
|---|---|---:|---|
|
| 21 |
-
| arxivmath | avg@4 | 21.36% | 412/412 scored |
|
| 22 |
-
| frontierscience-olympiad | avg@30 | 53.87% | 3000/3000 scored |
|
| 23 |
-
| frontierscience-research | Mean rubric (scored trajectories) | 32.77% | 1799/1800 scored |
|
| 24 |
-
| scicode-verified | Main-problem accuracy | 48.44% | 64/64 problems |
|
| 25 |
-
| terminal-bench-2.1 | avg@2 | 9.55% | 176/178 scored |
|
| 26 |
-
| scicoding-agent | avg@2 | 22.00% | 100/100 scored |
|
| 27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
## HLE-Verified (2026-10-30)
|
| 30 |
|
|
|
|
| 1 |
+
# gpt-oss-20b
|
| 2 |
|
| 3 |
+
## Aligned agentic rerun — 2026-11-04
|
| 4 |
|
| 5 |
+
| Benchmark | avg@2 | User-adjudicated timeout zeros |
|
| 6 |
+
|---|---:|---:|
|
| 7 |
+
| scicoding-agent | 22/100 (22.00%) | 0 |
|
| 8 |
+
| terminal-bench-2.1 | 12/178 (6.74%) | 2 |
|
| 9 |
|
| 10 |
+
H100; Terminus 2; temperature 1.0, top-p 0.95, top-k 20; concurrency 2 per engine. TP1 for GPT-OSS-20B; TP2 for the other models. Two independent attempts per task. All attempts are scored. Timeout zeros preserve their original native null verdict and evidence; they are not native test failures.
|
| 11 |
|
| 12 |
+
These benchmark folders replace the older inconsistent agentic exports. Complete trajectories, continuation chains, verifier artifacts and frozen configs are included. Older exports remain recoverable in repository revision `55136ae404538e1ebb3dfaba6261b529109883e9`.
|
| 13 |
|
| 14 |
+
## Unchanged non-agentic evaluations
|
| 15 |
|
| 16 |
+
These results are preserved from the latest pre-publication revision, not rolled back to the original export date.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
- [arxivmath](arxivmath/summary.json)
|
| 19 |
+
- [frontierscience-olympiad](frontierscience-olympiad/summary.json)
|
| 20 |
+
- [frontierscience-research](frontierscience-research/summary.json)
|
| 21 |
+
- [scicode-verified](scicode-verified/summary.json)
|
| 22 |
+
- [hle-verified](hle-verified/summary.json)
|
| 23 |
|
| 24 |
## HLE-Verified (2026-10-30)
|
| 25 |
|
gpt-oss-20b/configs.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
gpt-oss-20b/manifest.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"README.md": {
|
| 3 |
-
"size":
|
| 4 |
-
"sha256": "
|
| 5 |
},
|
| 6 |
"arxivmath/events.jsonl.gz": {
|
| 7 |
"size": 59938582,
|
|
@@ -20,8 +20,8 @@
|
|
| 20 |
"sha256": "dab2b1bb1533fc247b018d60da373595e943c7534a18ed100b752708555e45ab"
|
| 21 |
},
|
| 22 |
"configs.json": {
|
| 23 |
-
"size":
|
| 24 |
-
"sha256": "
|
| 25 |
},
|
| 26 |
"frontierscience-olympiad/events.jsonl.gz": {
|
| 27 |
"size": 209349166,
|
|
@@ -56,44 +56,8 @@
|
|
| 56 |
"sha256": "c144bb03f7970f6a14f954e4d209e603388c298a014060c0671fb1627cee657a"
|
| 57 |
},
|
| 58 |
"provenance.json": {
|
| 59 |
-
"size":
|
| 60 |
-
"sha256": "
|
| 61 |
-
},
|
| 62 |
-
"recovery-provenance/code-hashes.json": {
|
| 63 |
-
"size": 11716,
|
| 64 |
-
"sha256": "3de02d0b9bb51db75ce4d7e85c4cdce4d0a930ac3a436d862a08b0f25142ad21"
|
| 65 |
-
},
|
| 66 |
-
"recovery-provenance/combined-summary.json": {
|
| 67 |
-
"size": 500,
|
| 68 |
-
"sha256": "dd7a367f9554378a6bcedcd8f6c7f0bb4a542b09bf4d1b0000537e44d55fb74c"
|
| 69 |
-
},
|
| 70 |
-
"recovery-provenance/results-overlay.json": {
|
| 71 |
-
"size": 40842,
|
| 72 |
-
"sha256": "e3e639b6d270f5c5fa588d96bc84c65b659b18dd26f98ace3f5c88ee81d57555"
|
| 73 |
-
},
|
| 74 |
-
"recovery-provenance/superseded-http400-attempts/judge_trajectories.jsonl": {
|
| 75 |
-
"size": 3040081,
|
| 76 |
-
"sha256": "9cee04fb28b756ce7b69d53823e9bda58be78f101bf4c61bbf725e9a93bd92c9"
|
| 77 |
-
},
|
| 78 |
-
"recovery-provenance/superseded-http400-attempts/native_artifacts.tar.gz": {
|
| 79 |
-
"size": 3827325,
|
| 80 |
-
"sha256": "1b3cf58e916884db00bcce4047db7158fec16f27098478be2f46262bab53d566"
|
| 81 |
-
},
|
| 82 |
-
"recovery-provenance/superseded-http400-attempts/summary.json": {
|
| 83 |
-
"size": 285,
|
| 84 |
-
"sha256": "484b88ee041ebb0d8f41f5005a87f3427cf6d0d537de13460c287abf70733f48"
|
| 85 |
-
},
|
| 86 |
-
"recovery-provenance/superseded-http400-attempts/trajectories.jsonl": {
|
| 87 |
-
"size": 12341402,
|
| 88 |
-
"sha256": "42a78ca139195a06f2737de65bf0526d538ce4b9c354d70f23e2634c0822bdb2"
|
| 89 |
-
},
|
| 90 |
-
"recovery-provenance/targets.json": {
|
| 91 |
-
"size": 47419,
|
| 92 |
-
"sha256": "424c93becfdda0270d36c86d5e32ccaea5f54abe2ef98e7578b4b4859393400a"
|
| 93 |
-
},
|
| 94 |
-
"recovery-provenance/validation.json": {
|
| 95 |
-
"size": 1816,
|
| 96 |
-
"sha256": "b4631b2d9bedd321f97568958bf087b4f8e47a3fc239f3481acd4b2bc6fe7154"
|
| 97 |
},
|
| 98 |
"scicode-verified/judge_trajectories.jsonl": {
|
| 99 |
"size": 247496,
|
|
@@ -112,39 +76,67 @@
|
|
| 112 |
"sha256": "1ed2cb2eb5d55a8e4d7f3d82a270f204669225f93a022d792ca4bbe0025794a6"
|
| 113 |
},
|
| 114 |
"scicoding-agent/judge_trajectories.jsonl": {
|
| 115 |
-
"size":
|
| 116 |
-
"sha256": "
|
| 117 |
},
|
| 118 |
"scicoding-agent/native_artifacts.tar.gz": {
|
| 119 |
-
"size":
|
| 120 |
-
"sha256": "
|
| 121 |
},
|
| 122 |
"scicoding-agent/summary.json": {
|
| 123 |
-
"size":
|
| 124 |
-
"sha256": "
|
| 125 |
},
|
| 126 |
"scicoding-agent/trajectories.jsonl": {
|
| 127 |
-
"size":
|
| 128 |
-
"sha256": "
|
| 129 |
},
|
| 130 |
"summary.json": {
|
| 131 |
-
"size":
|
| 132 |
-
"sha256": "
|
| 133 |
},
|
| 134 |
"terminal-bench-2.1/judge_trajectories.jsonl": {
|
| 135 |
-
"size":
|
| 136 |
-
"sha256": "
|
| 137 |
},
|
| 138 |
"terminal-bench-2.1/native_artifacts.tar.gz": {
|
| 139 |
-
"size":
|
| 140 |
-
"sha256": "
|
| 141 |
},
|
| 142 |
"terminal-bench-2.1/summary.json": {
|
| 143 |
-
"size":
|
| 144 |
-
"sha256": "
|
| 145 |
},
|
| 146 |
"terminal-bench-2.1/trajectories.jsonl": {
|
| 147 |
-
"size":
|
| 148 |
-
"sha256": "
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 149 |
}
|
| 150 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"README.md": {
|
| 3 |
+
"size": 2376,
|
| 4 |
+
"sha256": "1c77405b443226bc2e6fc09554ca62748992370a8d6cb5f65a7e14d3f16df8ac"
|
| 5 |
},
|
| 6 |
"arxivmath/events.jsonl.gz": {
|
| 7 |
"size": 59938582,
|
|
|
|
| 20 |
"sha256": "dab2b1bb1533fc247b018d60da373595e943c7534a18ed100b752708555e45ab"
|
| 21 |
},
|
| 22 |
"configs.json": {
|
| 23 |
+
"size": 488633,
|
| 24 |
+
"sha256": "0bff54ba668d19c93b5a48b6021f2723b3a189fbc5e34308927c82dbadeeda2c"
|
| 25 |
},
|
| 26 |
"frontierscience-olympiad/events.jsonl.gz": {
|
| 27 |
"size": 209349166,
|
|
|
|
| 56 |
"sha256": "c144bb03f7970f6a14f954e4d209e603388c298a014060c0671fb1627cee657a"
|
| 57 |
},
|
| 58 |
"provenance.json": {
|
| 59 |
+
"size": 225920,
|
| 60 |
+
"sha256": "afc850a857e8825ff613d9fc4a3bccf0c4bcfdaff3bfad852f1c5e60cfd63eec"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
},
|
| 62 |
"scicode-verified/judge_trajectories.jsonl": {
|
| 63 |
"size": 247496,
|
|
|
|
| 76 |
"sha256": "1ed2cb2eb5d55a8e4d7f3d82a270f204669225f93a022d792ca4bbe0025794a6"
|
| 77 |
},
|
| 78 |
"scicoding-agent/judge_trajectories.jsonl": {
|
| 79 |
+
"size": 577093,
|
| 80 |
+
"sha256": "46e407e84ab790e87b6c98855f8af1faab95a03a6df73959e0cb7787c1b2c342"
|
| 81 |
},
|
| 82 |
"scicoding-agent/native_artifacts.tar.gz": {
|
| 83 |
+
"size": 2695384,
|
| 84 |
+
"sha256": "974132c245bc3ab31f1c5e95dc97c8aee3dc1dd580dc9149a4eee6bf1373af37"
|
| 85 |
},
|
| 86 |
"scicoding-agent/summary.json": {
|
| 87 |
+
"size": 599,
|
| 88 |
+
"sha256": "826d6445b5dacb0cffc7b5ce67824eaa34c399dbb57d8ffcf7b139142fd1275a"
|
| 89 |
},
|
| 90 |
"scicoding-agent/trajectories.jsonl": {
|
| 91 |
+
"size": 11682875,
|
| 92 |
+
"sha256": "a031d3112c34f8bfb7d42eb66deb59c9c3a7d0ab0eafe2e547ddcaaa931f3c42"
|
| 93 |
},
|
| 94 |
"summary.json": {
|
| 95 |
+
"size": 17994,
|
| 96 |
+
"sha256": "5e3d08b9308902681b44116f825408aa60d546c48da7dda50cdb8138f233960f"
|
| 97 |
},
|
| 98 |
"terminal-bench-2.1/judge_trajectories.jsonl": {
|
| 99 |
+
"size": 5921956,
|
| 100 |
+
"sha256": "c001492c957e73d83d2ca9ac96c0ec126caac87e9d5ff6107a7c083959590108"
|
| 101 |
},
|
| 102 |
"terminal-bench-2.1/native_artifacts.tar.gz": {
|
| 103 |
+
"size": 40629845,
|
| 104 |
+
"sha256": "f9fb0766842be64d39dd208b6005ed25b6a33df7855f863eb56da4663dcda817"
|
| 105 |
},
|
| 106 |
"terminal-bench-2.1/summary.json": {
|
| 107 |
+
"size": 2418,
|
| 108 |
+
"sha256": "c5f099e1676aae1353ddbe07078a31b62b351a43c05ef6c7ab23e247903280f6"
|
| 109 |
},
|
| 110 |
"terminal-bench-2.1/trajectories.jsonl": {
|
| 111 |
+
"size": 57489725,
|
| 112 |
+
"sha256": "5e4ed1a931f54b3416c168184a251371758333340cee37aa5e63ee3008c4d68c"
|
| 113 |
+
},
|
| 114 |
+
"scicoding-agent/README.md": {
|
| 115 |
+
"size": 821,
|
| 116 |
+
"sha256": "540014d1dda0c966cd39188101be771fa56ac652367eb925a2a43d66aa91f2b5"
|
| 117 |
+
},
|
| 118 |
+
"scicoding-agent/configs.json": {
|
| 119 |
+
"size": 242681,
|
| 120 |
+
"sha256": "fee437011bdca7be6ae93e9ff7cd3934304ce818f47897e37cef95061a0e405a"
|
| 121 |
+
},
|
| 122 |
+
"scicoding-agent/provenance.json": {
|
| 123 |
+
"size": 344,
|
| 124 |
+
"sha256": "b8942a40d223a8a52d3a26c4102d6df2e86eff3dc20058b21e92010a424eac95"
|
| 125 |
+
},
|
| 126 |
+
"terminal-bench-2.1/README.md": {
|
| 127 |
+
"size": 823,
|
| 128 |
+
"sha256": "09186dbd6c7b020a147718ae00684241e68e84933fa931651afd80b9e0b2e356"
|
| 129 |
+
},
|
| 130 |
+
"terminal-bench-2.1/configs.json": {
|
| 131 |
+
"size": 242459,
|
| 132 |
+
"sha256": "788216a5751a080a4cf462fec261c3940921e9fc3eeba4d1ebb2552b77f6372e"
|
| 133 |
+
},
|
| 134 |
+
"terminal-bench-2.1/provenance.json": {
|
| 135 |
+
"size": 2296,
|
| 136 |
+
"sha256": "df25169c7b0c2eaabb3a62d86b7f9c0900c25c435769d0bbdd8bd1b249478154"
|
| 137 |
+
},
|
| 138 |
+
"terminal-bench-2.1/scoring_adjudications.tar.gz": {
|
| 139 |
+
"size": 28419,
|
| 140 |
+
"sha256": "8b1363e6dc0164a2a3dc4378d4443402798961ef87aeeb1907023f1b9910e932"
|
| 141 |
}
|
| 142 |
}
|
gpt-oss-20b/provenance.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
gpt-oss-20b/recovery-provenance/code-hashes.json
DELETED
|
@@ -1 +0,0 @@
|
|
| 1 |
-
{"scripts/agentic_harmony.py": "c5778bdeeb90c280b004552a09adf65dc6856b2597127a16f902cfc37e74438a", "scripts/agentic_plan.py": "5b15625933e97fb2c9552b7d7933ab931b1a76a8c05f832cf5822481eba8777a", "scripts/agentic_queue.py": "a10c337b6c869837e3df6b720072801251c70cb9901de0af02cad4ebb7fd9b6f", "scripts/agentic_recovery.py": "5902ff269956a74a1334dd9a9a9375284d06c97c09d416df1b4e88540340c6bd", "scripts/agentic_sampling.py": "437f51d4ad3109d1736f7631bf4aa8bfd55b7bae1e85300077238cd7cf73efe5", "scripts/checkpoint_eval.py": "24d84a50494db7eacb2635bae6d52fc4797c1b5643e99794638c9a645def4665", "scripts/checkpoint_io.py": "8ded23c709c003d74c30ff1ec194245aa23d21452115641de1985f4ac3dc6b03", "scripts/endpoint_health.py": "31629053e17b571f33dabfde694363d1d7b38ea54519653be41a14ea50f59356", "scripts/eval_profile.py": "5a99d56637a8210cedd97c006793e6e3365f5515778ee744338045ee7749d211", "scripts/export_results.py": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15", "scripts/prepare_agentic_harbor.py": "30590b04d8189c64bb2d7a494d482a04757cbffefd6bbab8271b6f157cdecf66", "scripts/rescore_agentic_snapshot.py": "cdaabca41a1f74f525e33b19d1abc7737b9edbc9b644fc96f609bc3f91cc6768", "scripts/run_agentic_harbor.py": "2779eacbb427a623b29454fea987f54ff7b58df427b0c63cc4232f36b2ba23fa", "scripts/scicode_suite.py": "fd16a1fec3fbfe0358f5fab9d755d6ec5326ba6fdab5539cbd4e508a99243f0a", "scripts/start_checkpoint_eval.py": "0f7814c206b805aaf679caa49d2b68eaee258319ad8cbe1d6775cb89624001d9", "scripts/verifier_health.py": "046d069aa0ed0ebf238b410eab98a9e50074254aedd7370041b4191594893f60", "tests/test_agentic_harmony.py": "4434a7018c9c0b798ad9f93ec45d9f92d1de161e44caff1311504d1cc56635dc", "tests/test_agentic_queue_profiles.py": "7a8c94bb4c19a16d2e5cd7373f443d678d033c178f81e343d8ae55ce55b4a3c6", "tests/test_agentic_recovery.py": "5cd8b5d38cce2c1e0a0d25c5fdbf696cc00f1e54cbb92a1569e29c1252f3b663", "tests/test_arxivmath_selection.py": "0e3634c234f37de1f9468f3cdf16f5d3809b124a76e707ce30854a2e612d5362", "tests/test_checkpoint_groups.py": "4f7d68ac80e658f53ae67b79ae4c1f91fbfc4c39b62812bc49eeeda24c1bde0d", "tests/test_eval_profiles.py": "3762f96c33e9b80e7853c83d58a2b49baacbaad6c09ec83c631e31313a376dcf", "tests/test_export_results.py": "766bb3fa7ce580b8839c2e9248589f2aede7589b8eca3c3265b948e5926e3f3c", "tests/test_frontierscience_official.py": "10f71356dcb913c14bd32e872d04dd8f1d41d45fddf5d96f340b261296fc57b1", "tests/test_hle_verified.py": "427621e09c71e294a78ed45be51a5be4314f882840ec5f4531c4214741b829d5", "tests/test_judge_contract.py": "ccb087f205cd0a96668ade0ed6782d92058adf51ba3d88e381acad5eabc67190", "tests/test_layout.py": "e1925b7d173101ecb722ed637ee711e5ed2517c03ba7bc4148dba686944c55dd", "tests/test_one_line_eval.py": "144987804029ef32a46bdac1561585a0a22e54c123a44592f155dd99b41f5fa3", "tests/test_openai_backend.py": "53b8f6d658069b72d9188c280788201c775c2c8b2aefa8f42a9e9bd53149517d", "tests/test_prompt_stage_outcome.py": "8835df10b169c889dc2f93f6111d22a2fafee76ad6b3d1cf8d2f8383437d42de", "tests/test_rescore_agentic_snapshot.py": "ee882c65cde964d13e3a4f0f1284b260b5004ec96e60ece4d41c921d3491b5cf", "tests/test_scicode_suite.py": "51180f6a20e5f10a46e0d2796e88f2660793c6d2d03a5d8c303bcd8e63c17c06", "tests/test_scicoding_length_recovery.py": "d460ec5f5489d6f92905237584885407edd4f5ed63a6798ab8aa34d8c18ddb4b", "tests/test_scicoding_output_recovery.py": "01cbaed0ef6d4f535d8a767e708409cb61906257acdee12c87f9cafca83a6f45", "tests/test_scicoding_sandbox_identity.py": "31fb2036b98e19fbeea03e79246128ad0f8d02385aebe9d43d8e0099c1132e50", "tests/test_scicoding_tokenizer_provenance.py": "3eb776c10883333259dfd19ea705a4152225eddcf2f66c0edffc249b9b892a5c", "tests/test_server_lifecycle.py": "316880f817c995f012ba2ed3fd158776db1cd424b0f9e7af359d2a3e4af4b906", "tests/test_sft_final_checkpoint.py": "416a179105907016dfdbdd3bbd4668af7663271070b2271c4609db889e292ef4", "tests/test_suite_scheduling.py": "759376108d58a0e9ba290d47c6dabe1933da2d166b02c1f56aca2ee025c83099", "tests/test_terminal_bench_queue.py": "ef42c06385334aff5d08c73af1331459b2b7ff94a439580af78c351852f391fb", "tests/test_terminal_bench_sharding.py": "c91336fdf26c8ef513eb86e865aec74d6feb707357456ea04c456cd5765a091c", "benchmarks/hle-verified/scripts/prepare_suite.py": "2b0a9338a0f227856ff3ec7832a124931eff4b0c2db76fd28c55cd97f44c6263", "benchmarks/scicode-verified/scripts/install_helpers.py": "5a34e6e0f27a6e38ad342c975c0cc74e788b6f31833386b8c1b31312c5e5d858", "benchmarks/scicode-verified/scripts/regrade.py": "5b363c3fd8dca09f6b5142e72499ff02009493cbf2556f7ea2909648cbe84309", "benchmarks/scicode-verified/scripts/render_pod.py": "0a557dd8b71a2149b30a45ca5c85680ea2b4205e32c20b4c72df0031b08b3a3b", "benchmarks/scicode-verified/scripts/run.py": "b4570a429f1ed781fd8d93c616cd1c11391abca8b45ecee9fbcafb3da920d79b", "benchmarks/scicode-verified/scripts/smoke_grader.py": "1e163e2d2c34700e4e909b433695afd03b093c9916aeeb49b54269f5eeb7558e", "benchmarks/scicode-verified/tests/test_runner.py": "c5dc73854c075280bc72ab0b6ced4cea4ea7fe10c73c5831eb730927cb0ebeb2", "benchmarks/scicode-verified/vendor/compare/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "benchmarks/scicode-verified/vendor/compare/cmp.py": "8b3e790cc29dc87ef1f0f1b9bfd68c877728c78c40b541a69d10015221e9e235", "benchmarks/scicoding-agent/harness/__init__.py": "4ccaffbb46e9585d7fa8a5cd810082b855a7c66a01e7574e5e31c9a04b8da266", "benchmarks/scicoding-agent/harness/constants.py": "718bfe6e812c6f47cb6190f5476403e048095390232c2368ffcf3142c2a4f4c1", "benchmarks/scicoding-agent/harness/facet_runtime.py": "b1c5c5af8ed9a165194b74191a5a9d76a740d5dee33abfb3de2f75622e992617", "benchmarks/scicoding-agent/harness/reasoning_budget.py": "7309e3c11b3e271d0023e379ee3edab0f7d713edc621c37543fde5575f67418c", "benchmarks/scicoding-agent/harness/terminus_distillation.py": "37e61ea9fce55a24b0d5f2d5eb04555546dcf0e5d410ab4c61015aa38eaa9965", "benchmarks/scicoding-agent/scripts/prepare.py": "9cc53bc0c6e4b090336e5bd5124641fcbe2e7488910b2e4642aff063373c0e18", "benchmarks/scicoding-agent/scripts/run_trial.py": "ebb89ad022b3ecf172774cbe66b8ae62746c9e2f54121e25eb5701f9a4922c63", "benchmarks/terminal-bench-2.1/scripts/run_shard.py": "1ea3f7b2267c2cfc6077c17afc32f22999c10eeecec38eabcea0832bab0d29fe", "benchmarks/terminal-bench-2.1/scripts/run_shared_queue.py": "0ae05be22ecdf31ddd316ac52ebcf575cf6c34f016ccc7295d2e3ea4062d5a41", "benchmarks/terminal-bench-science/scripts/harbor_runner_pod.py": "c86667a49088853d4d262426353bfbd6a6aca0df8cc9bc94c4711d58c2c89a75", "benchmarks/terminal-bench-science/scripts/prepare_shared_queue.py": "16c15adafb57a1b0b7e904e4a743ea3ef16eae8026fb5c4c27e9c24872ad8816", "benchmarks/terminal-bench-science/scripts/request_worker.py": "e95313feaf39a4703882da1bd8ee20acf42e9b68af72802914f61febfd57909c", "benchmarks/terminal-bench-science/scripts/submit.py": "781ff39ee1c6c044f49a5a872286d7b281b644103baced1e8047d8021efc917d", "benchmarks/terminal-bench-science/tests/test_integration.py": "9418422f3841d6ab9f72739acea6b39289f98589a4e8dab799a5d73839cd88a3", "serving/sglang_external_models/sitecustomize.py": "f3a3bcaefe09508be16d8a08a552625ca3065a68c77ec613ce101dfd56cb6000", "serving/sglang_external_models/gemma4_compat/__init__.py": "632612ae71e30ebfe66eeaaaf201aa687d1ab20cc9a36b8b337d6f95e505a0a9", "serving/sglang_external_models/gemma4_compat/model.py": "c73f576392dda11d1b9adb76233282f65dbc620d66c9f5e7e5d1aaf0037abeba", "serving/sglang_external_models/sci_qwen35_text_model/__init__.py": "e70dc084d24abfa4b9578f47e0e224d8e91771b449c9479449938831055503eb", "serving/sglang_external_models/sci_qwen35_text_model/qwen35_text.py": "a7232c4a5b097909a1ce2939eb2d5e143961d76e2abd02501720dbe033d07465", "serving/sglang_external_models/sci_qwen35_text_model/sglang_config.py": "ba306790d7eba6b770d239027226e456bb55526cbba4025902bb8314bbaa7276", "vendor/src/scieval/__init__.py": "19d667e1beb23f97356ee5fd5bffbbf28929d1580cdd407ee8fe25c7c2fe4510", "vendor/src/scieval/agentic.py": "358744588fe07c21c277557bd9122ea56fb9d9992f7a01868cf7220fd5f032da", "vendor/src/scieval/cli.py": "5e0cb762bc7428418433876b14b648b152314a55e7faf9e15a86ebdca3a2b709", "vendor/src/scieval/config.py": "95043b5cd443615ede853d5a925583470ed989361bdaf3b4761c17e3e61907ea", "vendor/src/scieval/judge_defaults.py": "25e0b9065ae231d78382b5b4f07619c78ca95cf6e7e142747d6461e7165c9939", "vendor/src/scieval/native.py": "f51fc840c23a438799ed98a3e17183ee61df9fdc859580d414dd9e110a7eb917", "vendor/src/scieval/polar_evaluator.py": "1bca84bf1e3ca1eb34ce50f66ad094ca560e5215d5d2667a7c75ece493d433eb", "vendor/src/scieval/polar_harness.py": "db21c93ddd0d58982abfd2ca421635551005b73ca462f85a2b18dea813b0f92f", "vendor/src/scieval/polar_harness_command.py": "69fefbc78137a52dca374768e72ddad2cba3f93f8b3a8196896a588a793e018c", "vendor/src/scieval/polar_harness_v2.py": "78fbf1f8891b3b6947057b2135d6bf6d50cf2dfb9ef2e0021e7ecbdee7f9015c", "vendor/src/scieval/runner.py": "f4d95b14f29779cde6db604051a401c09e8f2fa9b584999cbb2752e4984e9d8a", "vendor/src/scieval/suite.py": "e8b3b4996edd481896167334db704b8ee77620c8f488c50bc34c9a42b52ecadf", "vendor/src/scieval/token_cap.py": "b177220aace70b8b60d75f2928b7d1f669f15e0d988853bb649b484d49bc996b", "vendor/src/scieval/types.py": "41eda562665e33e5fa07f3cb6be1a4f933110c2d89d3cb2139cbe73165a45120", "vendor/src/scieval/_vendor/__init__.py": "686340d0be9cf7d81f553cba5d6fce041444a17a46d92d8cc2bdfcc640710ed9", "vendor/src/scieval/backends/__init__.py": "01430eaff184aca2a3e4a5099bf360770ed92d3b35f7aafe2da70191135d8454", "vendor/src/scieval/backends/base.py": "108e00432b8ed8baaf2c5da94b6cd9f979aecd46a2d356760cd086a84119b059", "vendor/src/scieval/backends/bigbang_tools.py": "ca56cc65ac79d8fd91fc25670bb095f0eb00b7f26ca167685b591dd949538ece", "vendor/src/scieval/backends/kubernetes_code_exec.py": "6548ea6964ce38777e7a274938b73e31ae18dae4aff045ec6d9783513654dadc", "vendor/src/scieval/backends/openai_compatible.py": "0409964b292218e13b2953d6a063bf81221b3e2b432db8c5b75596ad96e4f368", "vendor/src/scieval/backends/registry.py": "30cc109218351ba08a6631e98daf05d7cd3a840cb8f5e84494a73e46895ff05c", "vendor/src/scieval/backends/usage_recording.py": "ed2610f830ce7f19aaa01a50fa1631af8e85580fb89492934589ae4d6fbc65d0", "vendor/src/scieval/benchmarks/__init__.py": "a805a2a219c5c420c70606f54f2a3bb7c4dc7b414bcdc08c22fa235909d6645f", "vendor/src/scieval/benchmarks/base.py": "6de50922aaa9c8c83a4b4f566fdddc63e5df4b0e1f8f2da00c649be880b4b5dc", "vendor/src/scieval/benchmarks/boxed_math.py": "b635652f5b443909ac9bea2469ef1968a60a7e0ae037e10014fadfd1ca004767", "vendor/src/scieval/benchmarks/data.py": "3cbe22d485592eaddcacca92b84621e155ace93e53974c76992a7068addcfd6a", "vendor/src/scieval/benchmarks/frontierscience.py": "d1acab1dd5d53f2a4e1c64d8106e2ada3bdeb614734d90ddc3c7157e668a9c3a", "vendor/src/scieval/benchmarks/hle.py": "7fe7ad72875a12c392860630210010a1e18c65840c52b53e5796d632bc8cf35e", "vendor/src/scieval/benchmarks/matharena_arxivmath.py": "65113884207d1694b026bd9aaea49c906359bbd51f15f5cbbca9f77d0d0471d1", "vendor/src/scieval/benchmarks/registry.py": "10879afe7f8f83e521ff5b225959b53606af7d062531f5d82f44187f3879c1a3", "vendor/src/scieval/benchmarks/scoring.py": "7c2af930faf2b0cdab62253e9c1c443b89c66a97e3ca04d2e5a6866616e73ca5", "vendor/src/scieval/_vendor/phybench_eed/EED.py": "ccabaa6dd7793f7fb1295a9436c365abb3806c8c42b486e192a59c8b49c9c955", "vendor/src/scieval/_vendor/phybench_eed/__init__.py": "4d1e465078b502aaba5c7730bcb95eb20e6c5897afbed5753c04179e7998905f", "vendor/src/scieval/_vendor/phybench_eed/extended_zss.py": "7fbfbd623f49d2bb3ebd4fd73150f335aa9237ce4f3a3f13832e8bace4ee6624", "vendor/src/scieval/_vendor/phybench_eed/latex_pre_process.py": "b8766b59bdadcee1a995e9e582e88877ee11fec83b8b829054e599d018e598a5"}
|
|
|
|
|
|
gpt-oss-20b/recovery-provenance/combined-summary.json
DELETED
|
@@ -1 +0,0 @@
|
|
| 1 |
-
{"time": 1793375731.7584558, "policy": "replace exactly 69 authorized HTTP-400 attempts; retain every original attempt in denominator", "benchmarks": {"scicoding-agent": {"expected": 100, "passed": 22, "unscored": 0, "exceptions": 31, "avg_at_2": 0.22, "grading_complete": true}, "terminal-bench-2.1": {"expected": 178, "passed": 17, "unscored": 2, "exceptions": 39, "avg_at_2": 0.09550561797752809, "grading_complete": false}}, "originals_preserved": true, "canonical_native_files_modified": false}
|
|
|
|
|
|
gpt-oss-20b/recovery-provenance/results-overlay.json
DELETED
|
@@ -1 +0,0 @@
|
|
| 1 |
-
{"terminal-bench-2.1--break-filter-js-from-html--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--break-filter-js-from-html--attempt-002/break-filter-js-from-html__7bDmnkX/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--break-filter-js-from-html--attempt-002/break-filter-js-from-html__r4Wigzz/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--build-pmars--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--build-pmars--attempt-001/build-pmars__ZgivZ5k/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--build-pmars--attempt-001/build-pmars__CtXU4qe/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--build-pmars--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--build-pmars--attempt-002/build-pmars__4B9bTrA/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--build-pmars--attempt-002/build-pmars__qJdMUUh/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--build-pov-ray--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--build-pov-ray--attempt-001/build-pov-ray__482nShJ/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--build-pov-ray--attempt-001/build-pov-ray__h9xA5Yx/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--build-pov-ray--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--build-pov-ray--attempt-002/build-pov-ray__oJ68K3D/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--build-pov-ray--attempt-002/build-pov-ray__GwqYy3R/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--caffe-cifar-10--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--caffe-cifar-10--attempt-001/caffe-cifar-10__su5aAZt/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--caffe-cifar-10--attempt-001/caffe-cifar-10__oRnHFnw/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--caffe-cifar-10--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--caffe-cifar-10--attempt-002/caffe-cifar-10__5fWZvyx/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--caffe-cifar-10--attempt-002/caffe-cifar-10__ZMz7AMM/result.json", "reward": 0.0, "exception": "ConsecutiveCapHitsExceeded", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--chess-best-move--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--chess-best-move--attempt-001/chess-best-move__JzQ8R58/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--chess-best-move--attempt-001/chess-best-move__cqWphAj/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--circuit-fibsqrt--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--circuit-fibsqrt--attempt-001/circuit-fibsqrt__UEZtjNb/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--circuit-fibsqrt--attempt-001/circuit-fibsqrt__Lffrzbd/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--circuit-fibsqrt--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--circuit-fibsqrt--attempt-002/circuit-fibsqrt__4UcwtpE/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--circuit-fibsqrt--attempt-002/circuit-fibsqrt__imZsK79/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--cobol-modernization--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--cobol-modernization--attempt-001/cobol-modernization__UxYuLaZ/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--cobol-modernization--attempt-001/cobol-modernization__Nvdtv3Y/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--cobol-modernization--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--cobol-modernization--attempt-002/cobol-modernization__jgW8Gg4/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--cobol-modernization--attempt-002/cobol-modernization__729TCD9/result.json", "reward": 1.0, "exception": "ConsecutiveCapHitsExceeded", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--code-from-image--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--code-from-image--attempt-001/code-from-image__6shzUWv/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--code-from-image--attempt-001/code-from-image__mq4a6UC/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--constraints-scheduling--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--constraints-scheduling--attempt-002/constraints-scheduling__BFhqW5c/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--constraints-scheduling--attempt-002/constraints-scheduling__soRdcsR/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--crack-7z-hash--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--crack-7z-hash--attempt-002/crack-7z-hash__7BmN6Ga/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--crack-7z-hash--attempt-002/crack-7z-hash__iR6vdfB/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--custom-memory-heap-crash--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--custom-memory-heap-crash--attempt-001/custom-memory-heap-crash__wmMBNNz/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--custom-memory-heap-crash--attempt-001/custom-memory-heap-crash__p8HUKLs/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--custom-memory-heap-crash--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--custom-memory-heap-crash--attempt-002/custom-memory-heap-crash__nYZ8HZr/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--custom-memory-heap-crash--attempt-002/custom-memory-heap-crash__5cfeMrr/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--distribution-search--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--distribution-search--attempt-002/distribution-search__d4iv3FT/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--distribution-search--attempt-002/distribution-search__3cmjRHH/result.json", "reward": 0.0, "exception": "ConsecutiveCapHitsExceeded", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--dna-assembly--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--dna-assembly--attempt-002/dna-assembly__Fs63B5u/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--dna-assembly--attempt-002/dna-assembly__DRqo8Mv/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--dna-insert--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--dna-insert--attempt-001/dna-insert__mbfbc45/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--dna-insert--attempt-001/dna-insert__Yg3R6J6/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--dna-insert--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--dna-insert--attempt-002/dna-insert__hiAKVH6/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--dna-insert--attempt-002/dna-insert__dbshFpC/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--extract-elf--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--extract-elf--attempt-001/extract-elf__JhrnviS/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--extract-elf--attempt-001/extract-elf__32dkPpa/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--feal-differential-cryptanalysis--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--feal-differential-cryptanalysis--attempt-001/feal-differential-cryptanalysis__fsGiM7v/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--feal-differential-cryptanalysis--attempt-001/feal-differential-cryptanalysis__Dfkb6zd/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002/feal-differential-cryptanalysis__FCDUdug/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--feal-differential-cryptanalysis--attempt-002/feal-differential-cryptanalysis__RSWhaEc/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--feal-linear-cryptanalysis--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--feal-linear-cryptanalysis--attempt-001/feal-linear-cryptanalysis__YWVf2y6/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--feal-linear-cryptanalysis--attempt-001/feal-linear-cryptanalysis__b9ZKzqq/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--feal-linear-cryptanalysis--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--feal-linear-cryptanalysis--attempt-002/feal-linear-cryptanalysis__o6iAnBL/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--feal-linear-cryptanalysis--attempt-002/feal-linear-cryptanalysis__DRv66SQ/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--financial-document-processor--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--financial-document-processor--attempt-002/financial-document-processor__ZPKd88X/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--financial-document-processor--attempt-002/financial-document-processor__ijMK6mQ/result.json", "reward": 0.0, "exception": "ConsecutiveCapHitsExceeded", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--fix-code-vulnerability--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--fix-code-vulnerability--attempt-001/fix-code-vulnerability__n8zacAu/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--fix-code-vulnerability--attempt-001/fix-code-vulnerability__iAoBkeR/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--fix-code-vulnerability--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--fix-code-vulnerability--attempt-002/fix-code-vulnerability__iq77wEv/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--fix-code-vulnerability--attempt-002/fix-code-vulnerability__w69CmeU/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--fix-ocaml-gc--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--fix-ocaml-gc--attempt-001/fix-ocaml-gc__u5JnS5m/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--fix-ocaml-gc--attempt-001/fix-ocaml-gc__AQRGuFh/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--fix-ocaml-gc--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--fix-ocaml-gc--attempt-002/fix-ocaml-gc__7Szfjsv/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--fix-ocaml-gc--attempt-002/fix-ocaml-gc__s4nKzdk/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--gpt2-codegolf--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--gpt2-codegolf--attempt-002/gpt2-codegolf__prKjXEb/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--gpt2-codegolf--attempt-002/gpt2-codegolf__QZewKdW/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--headless-terminal--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--headless-terminal--attempt-001/headless-terminal__T3c9wJe/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--headless-terminal--attempt-001/headless-terminal__kAGpUhd/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--largest-eigenval--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--largest-eigenval--attempt-002/largest-eigenval__LSxhaHh/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--largest-eigenval--attempt-002/largest-eigenval__jPRaCfA/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--llm-inference-batching-scheduler--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--llm-inference-batching-scheduler--attempt-001/llm-inference-batching-scheduler__wt3goBd/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--llm-inference-batching-scheduler--attempt-001/llm-inference-batching-scheduler__3naaYTp/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--mailman--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--mailman--attempt-001/mailman__V6kofkk/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--mailman--attempt-001/mailman__7vBQ3bp/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--make-doom-for-mips--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--make-doom-for-mips--attempt-002/make-doom-for-mips__bhuF8YE/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--make-doom-for-mips--attempt-002/make-doom-for-mips__9qSq4K2/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--make-mips-interpreter--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--make-mips-interpreter--attempt-001/make-mips-interpreter__o3ja5VE/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--make-mips-interpreter--attempt-001/make-mips-interpreter__PvRzTdR/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--make-mips-interpreter--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--make-mips-interpreter--attempt-002/make-mips-interpreter__DT279bf/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--make-mips-interpreter--attempt-002/make-mips-interpreter__Be4oj6i/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--mcmc-sampling-stan--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--mcmc-sampling-stan--attempt-002/mcmc-sampling-stan__hBHx4XN/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--mcmc-sampling-stan--attempt-002/mcmc-sampling-stan__raf5HAe/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--merge-diff-arc-agi-task--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--merge-diff-arc-agi-task--attempt-002/merge-diff-arc-agi-task__o5nsj68/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--merge-diff-arc-agi-task--attempt-002/merge-diff-arc-agi-task__xTUBHkf/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--overfull-hbox--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--overfull-hbox--attempt-001/overfull-hbox__L8rUBZs/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--overfull-hbox--attempt-001/overfull-hbox__yGzWzJp/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--password-recovery--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--password-recovery--attempt-002/password-recovery__Wb8769n/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--password-recovery--attempt-002/password-recovery__yBJwbM3/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--path-tracing--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--path-tracing--attempt-001/path-tracing__XA2feWD/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--path-tracing--attempt-001/path-tracing__pvifniy/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--path-tracing--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--path-tracing--attempt-002/path-tracing__gME7UH2/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--path-tracing--attempt-002/path-tracing__nma5FwX/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--path-tracing-reverse--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--path-tracing-reverse--attempt-001/path-tracing-reverse__igWvJxU/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--path-tracing-reverse--attempt-001/path-tracing-reverse__JunWm6C/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--portfolio-optimization--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--portfolio-optimization--attempt-001/portfolio-optimization__7GensuX/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--portfolio-optimization--attempt-001/portfolio-optimization__a2PaJN9/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--protein-assembly--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--protein-assembly--attempt-001/protein-assembly__p9yEoJ2/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--protein-assembly--attempt-001/protein-assembly__mbkm4b3/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--prove-plus-comm--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--prove-plus-comm--attempt-001/prove-plus-comm__XmnqReF/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--prove-plus-comm--attempt-001/prove-plus-comm__PXVVT5n/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--prove-plus-comm--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--prove-plus-comm--attempt-002/prove-plus-comm__GYsTtwv/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--prove-plus-comm--attempt-002/prove-plus-comm__YmNMdWL/result.json", "reward": 1.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--pytorch-model-cli--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--pytorch-model-cli--attempt-002/pytorch-model-cli__XMQM57w/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--pytorch-model-cli--attempt-002/pytorch-model-cli__yLFWyR8/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--pytorch-model-recovery--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--pytorch-model-recovery--attempt-001/pytorch-model-recovery__kmXRRwh/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--pytorch-model-recovery--attempt-001/pytorch-model-recovery__MDMYB6U/result.json", "reward": null, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--pytorch-model-recovery--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--pytorch-model-recovery--attempt-002/pytorch-model-recovery__MuZvqgm/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--pytorch-model-recovery--attempt-002/pytorch-model-recovery__9yuq7dz/result.json", "reward": null, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--regex-chess--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--regex-chess--attempt-001/regex-chess__XKySyHR/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--regex-chess--attempt-001/regex-chess__oB9xbqt/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--regex-chess--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--regex-chess--attempt-002/regex-chess__hbE4c4e/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--regex-chess--attempt-002/regex-chess__KZiAbZm/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--rstan-to-pystan--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--rstan-to-pystan--attempt-001/rstan-to-pystan__TMPvyhQ/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--rstan-to-pystan--attempt-001/rstan-to-pystan__VtQRPHh/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--rstan-to-pystan--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--rstan-to-pystan--attempt-002/rstan-to-pystan__yfAoDe5/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--rstan-to-pystan--attempt-002/rstan-to-pystan__A33838P/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--sam-cell-seg--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--sam-cell-seg--attempt-001/sam-cell-seg__7XY8Xjb/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--sam-cell-seg--attempt-001/sam-cell-seg__WgxzCvV/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--schemelike-metacircular-eval--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--schemelike-metacircular-eval--attempt-001/schemelike-metacircular-eval__MKP7Wpc/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--schemelike-metacircular-eval--attempt-001/schemelike-metacircular-eval__adDdAzN/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--schemelike-metacircular-eval--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--schemelike-metacircular-eval--attempt-002/schemelike-metacircular-eval__dFj58Hy/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--schemelike-metacircular-eval--attempt-002/schemelike-metacircular-eval__Pu3nTz2/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--sparql-university--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--sparql-university--attempt-001/sparql-university__GztFfgW/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--sparql-university--attempt-001/sparql-university__hKePYxK/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--sqlite-db-truncate--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--sqlite-db-truncate--attempt-001/sqlite-db-truncate__GoPtu3m/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--sqlite-db-truncate--attempt-001/sqlite-db-truncate__zZvHtwX/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--sqlite-with-gcov--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--sqlite-with-gcov--attempt-002/sqlite-with-gcov__FYtR3qU/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--sqlite-with-gcov--attempt-002/sqlite-with-gcov__GQ2wzmU/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--train-fasttext--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--train-fasttext--attempt-002/train-fasttext__VWLSWLX/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--train-fasttext--attempt-002/train-fasttext__55WRcYz/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--tune-mjcf--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--tune-mjcf--attempt-001/tune-mjcf__kQiZQFA/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--tune-mjcf--attempt-001/tune-mjcf__Sgf8sfn/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--tune-mjcf--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--tune-mjcf--attempt-002/tune-mjcf__sJjFa35/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--tune-mjcf--attempt-002/tune-mjcf__Pq58BXc/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--vulnerable-secret--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--vulnerable-secret--attempt-001/vulnerable-secret__8JXDQTQ/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--vulnerable-secret--attempt-001/vulnerable-secret__C9WQe3R/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--vulnerable-secret--attempt-002": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--vulnerable-secret--attempt-002/vulnerable-secret__UMN7SrY/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--vulnerable-secret--attempt-002/vulnerable-secret__dDkGcCo/result.json", "reward": 0.0, "exception": null, "selection": "retry replaces original, not best-of"}, "terminal-bench-2.1--write-compressor--attempt-001": {"original_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/run/agentic/jobs/task-terminal-bench-2.1--write-compressor--attempt-001/write-compressor__fFAsA7u/result.json", "retry_result": "/workspace/data/sci-llm-training/runtime/gpt-oss-20b-full-suite-20261030/operations/harmony-recovery-20261030/agentic/jobs/task-terminal-bench-2.1--write-compressor--attempt-001/write-compressor__59CgE49/result.json", "reward": 0.0, "exception": "AgentTimeoutError", "selection": "retry replaces original, not best-of"}}
|
|
|
|
|
|
gpt-oss-20b/recovery-provenance/superseded-http400-attempts/summary.json
DELETED
|
@@ -1 +0,0 @@
|
|
| 1 |
-
{"tasks": 49, "attempts": 69, "attempts_per_task": null, "metric": "pass@1", "passed": 1, "pass_at_1": 0.014492753623188406, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 1, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1"}
|
|
|
|
|
|
gpt-oss-20b/recovery-provenance/superseded-http400-attempts/trajectories.jsonl
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:42a78ca139195a06f2737de65bf0526d538ce4b9c354d70f23e2634c0822bdb2
|
| 3 |
-
size 12341402
|
|
|
|
|
|
|
|
|
|
|
|