JLiangHe commited on
Commit
09a2e7b
·
verified ·
1 Parent(s): 6a55c50

Publish agentic RL step-20 intermediate evaluations

Browse files

Full trajectories, judgments, native artifacts and frozen configurations. Unresolved scores remain null/provisional; final step-35 results unchanged.

.gitattributes CHANGED
@@ -440,3 +440,7 @@ rl-agentic/intermediate/step-10/scicoding-agent/judge_trajectories.jsonl filter=
440
  rl-agentic/intermediate/step-10/scicoding-agent/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
441
  rl-agentic/intermediate/step-10/terminal-bench-2.1/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
442
  rl-agentic/intermediate/step-10/terminal-bench-2.1/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
440
  rl-agentic/intermediate/step-10/scicoding-agent/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
441
  rl-agentic/intermediate/step-10/terminal-bench-2.1/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
442
  rl-agentic/intermediate/step-10/terminal-bench-2.1/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
443
+ rl-agentic/intermediate/step-20/scicoding-agent/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
444
+ rl-agentic/intermediate/step-20/scicoding-agent/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
445
+ rl-agentic/intermediate/step-20/terminal-bench-2.1/judge_trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
446
+ rl-agentic/intermediate/step-20/terminal-bench-2.1/trajectories.jsonl filter=lfs diff=lfs merge=lfs -text
rl-agentic/intermediate/step-20/README.md ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Agentic RL 12×24 — step 20
2
+
3
+ Checkpoint: `JLiangHe/sci-ckpt-final/rl-agentic-new-12x24/step-20`, revision `a3d690d2356f3a8622f4b8047d2e98d23f3e9338`.
4
+
5
+ | Benchmark | Metric | Result | Unscored |
6
+ |---|---|---:|---:|
7
+ | [scicoding-agent](scicoding-agent/) | avg@2 | 42.00% | 0 |
8
+ | [terminal-bench-2.1](terminal-bench-2.1/) | avg@2 | 26.40% | 0 |
9
+
10
+ All planned attempts are retained. Native verifier success counts after abnormal agent stops. Missing verdicts remain null, not manually assigned zeros. Agentic percentages use every planned attempt; provisional Research means use scored trajectories only.
11
+
12
+ Agentic: Terminus 2, avg@2, temperature 0.7, top-p 0.95, top-k 20; 16 H100 GPUs, TP2, eight engines, two tasks per engine.
13
+
14
+ Reasoning benchmarks were not evaluated for this checkpoint and are not included.
15
+
16
+ Each benchmark includes full trajectories, judge/verifier evidence, frozen configs and provenance. Native agent artifacts and reasoning event histories are compressed. No evaluation was regenerated or rejudged for publication.
rl-agentic/intermediate/step-20/configs.json ADDED
The diff for this file is too large to render. See raw diff
 
rl-agentic/intermediate/step-20/manifest.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"README.md": {"size": 1039, "sha256": "fe5dd1eb497a155b53d629dc4a0009f08d4213e2d4160149b9011ae8693213c2"}, "configs.json": {"size": 216158, "sha256": "781f86bdf7a7b3981a52ea062a16bfb26c9d7e345cd373bbefe7c535dbff2b4a"}, "provenance.json": {"size": 435, "sha256": "d79da89b2cd9b5bb0fa63deddc1aa060df18d69ecc6eda50d0199522179f73c6"}, "scicoding-agent/configs.json": {"size": 216145, "sha256": "18f1b408adcc89864e9af13879e5ccd477feae2aef2b36fb3e7aa1f255eaf9d5"}, "scicoding-agent/judge_trajectories.jsonl": {"size": 814599, "sha256": "2cd918d8c0652be06f6db5c638d967d24e3a181316e9cd86abc385cf6e9f1c00"}, "scicoding-agent/native_artifacts.tar.gz": {"size": 152191785, "sha256": "12ebece53ede5a47a84cf239a03c3ca2c79bc746e9ae21e7e8db210d70e15654"}, "scicoding-agent/provenance.json": {"size": 435, "sha256": "d79da89b2cd9b5bb0fa63deddc1aa060df18d69ecc6eda50d0199522179f73c6"}, "scicoding-agent/summary.json": {"size": 570, "sha256": "a37996896f56423c92c303fbc7cd571e72ef261e44e53776bc083a48eb1c9adb"}, "scicoding-agent/trajectories.jsonl": {"size": 60102140, "sha256": "9cbff8af32bd95bf9ba756e9ca4f56b857fa5c7a176b929291b35dbeea1414c4"}, "summary.json": {"size": 1691, "sha256": "7445f6bd729b8a4d974179530ac22b948a07d158d74cc8e0b5ea9abaced48a40"}, "terminal-bench-2.1/configs.json": {"size": 216145, "sha256": "18f1b408adcc89864e9af13879e5ccd477feae2aef2b36fb3e7aa1f255eaf9d5"}, "terminal-bench-2.1/judge_trajectories.jsonl": {"size": 5562464, "sha256": "e307629fe08c9041c8cae01a95f030bea0a91a1b10bd7765e94ad41e8c9d9cf1"}, "terminal-bench-2.1/native_artifacts.tar.gz": {"size": 1005948626, "sha256": "a3ac5414212ec5bc1442e0e746331648bf34d7d0aa243ab53827a76c60244e76"}, "terminal-bench-2.1/provenance.json": {"size": 435, "sha256": "d79da89b2cd9b5bb0fa63deddc1aa060df18d69ecc6eda50d0199522179f73c6"}, "terminal-bench-2.1/summary.json": {"size": 585, "sha256": "673915412a02d63c425574c3d4cb5626e517b31303a956ee0cf7a217aeb8f516"}, "terminal-bench-2.1/trajectories.jsonl": {"size": 94928260, "sha256": "0b2fe3b3bfab90cd247e181fff64ce9341614bd6800b260793c2fb5b72590456"}}
rl-agentic/intermediate/step-20/provenance.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"identity": {"repo": "JLiangHe/sci-ckpt-final", "revision": "a3d690d2356f3a8622f4b8047d2e98d23f3e9338", "folder": "rl-agentic-new-12x24/step-20"}, "agentic_source": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "generation_or_judgment_rerun": false, "manual_zero_adjudications": false, "exporter_sha256": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15"}
rl-agentic/intermediate/step-20/scicoding-agent/configs.json ADDED
The diff for this file is too large to render. See raw diff
 
rl-agentic/intermediate/step-20/scicoding-agent/judge_trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2cd918d8c0652be06f6db5c638d967d24e3a181316e9cd86abc385cf6e9f1c00
3
+ size 814599
rl-agentic/intermediate/step-20/scicoding-agent/native_artifacts.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12ebece53ede5a47a84cf239a03c3ca2c79bc746e9ae21e7e8db210d70e15654
3
+ size 152191785
rl-agentic/intermediate/step-20/scicoding-agent/provenance.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"identity": {"repo": "JLiangHe/sci-ckpt-final", "revision": "a3d690d2356f3a8622f4b8047d2e98d23f3e9338", "folder": "rl-agentic-new-12x24/step-20"}, "agentic_source": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "generation_or_judgment_rerun": false, "manual_zero_adjudications": false, "exporter_sha256": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15"}
rl-agentic/intermediate/step-20/scicoding-agent/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 42, "avg_at_2": 0.42, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 17, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "provisional": false, "grading_complete": true, "generation_finished": true, "hardware": "H100", "tasks_per_engine": 2, "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}}
rl-agentic/intermediate/step-20/scicoding-agent/trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cbff8af32bd95bf9ba756e9ca4f56b857fa5c7a176b929291b35dbeea1414c4
3
+ size 60102140
rl-agentic/intermediate/step-20/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"checkpoint": "/workspace/data/raw_model/sci-ckpt-final/a3d690d2356f3a8622f4b8047d2e98d23f3e9338/rl-agentic-new-12x24/step-20", "identity": {"repo": "JLiangHe/sci-ckpt-final", "revision": "a3d690d2356f3a8622f4b8047d2e98d23f3e9338", "folder": "rl-agentic-new-12x24/step-20"}, "benchmarks": {"scicoding-agent": {"tasks": 50, "attempts": 100, "attempts_per_task": 2, "metric": "avg@2", "passed": 42, "avg_at_2": 0.42, "unscored": 0, "agent_timeouts": 0, "reward_positive_with_exception": 17, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "provisional": false, "grading_complete": true, "generation_finished": true, "hardware": "H100", "tasks_per_engine": 2, "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}}, "terminal-bench-2.1": {"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 47, "avg_at_2": 0.2640449438202247, "unscored": 0, "agent_timeouts": 112, "reward_positive_with_exception": 6, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "provisional": false, "grading_complete": true, "generation_finished": true, "hardware": "H100", "tasks_per_engine": 2, "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}}}, "unscored": {"scicoding-agent": 0, "terminal-bench-2.1": 0}, "generation_finished": true, "grading_complete": true, "provisional": false, "full_suite_complete": false, "evaluated_groups": ["agentic"]}
rl-agentic/intermediate/step-20/terminal-bench-2.1/configs.json ADDED
The diff for this file is too large to render. See raw diff
 
rl-agentic/intermediate/step-20/terminal-bench-2.1/judge_trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e307629fe08c9041c8cae01a95f030bea0a91a1b10bd7765e94ad41e8c9d9cf1
3
+ size 5562464
rl-agentic/intermediate/step-20/terminal-bench-2.1/native_artifacts.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a3ac5414212ec5bc1442e0e746331648bf34d7d0aa243ab53827a76c60244e76
3
+ size 1005948626
rl-agentic/intermediate/step-20/terminal-bench-2.1/provenance.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"identity": {"repo": "JLiangHe/sci-ckpt-final", "revision": "a3d690d2356f3a8622f4b8047d2e98d23f3e9338", "folder": "rl-agentic-new-12x24/step-20"}, "agentic_source": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "generation_or_judgment_rerun": false, "manual_zero_adjudications": false, "exporter_sha256": "3aec4e729f9b4b7d63fd9c183c07226a5e3350a7e5991a7e0c58738f3d4c0d15"}
rl-agentic/intermediate/step-20/terminal-bench-2.1/summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"tasks": 89, "attempts": 178, "attempts_per_task": 2, "metric": "avg@2", "passed": 47, "avg_at_2": 0.2640449438202247, "unscored": 0, "agent_timeouts": 112, "reward_positive_with_exception": 6, "scoring": "native_verifier", "scoring_policy": "native-verifier-after-agent-stop-v1", "source_run": "/workspace/data/sci-llm-training/runtime/agentic-12x24-intermediate-20261117/02-rl-agentic-12x24-step20", "provisional": false, "grading_complete": true, "generation_finished": true, "hardware": "H100", "tasks_per_engine": 2, "sampling": {"temperature": 0.7, "top_p": 0.95, "top_k": 20}}
rl-agentic/intermediate/step-20/terminal-bench-2.1/trajectories.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b2fe3b3bfab90cd247e181fff64ce9341614bd6800b260793c2fb5b72590456
3
+ size 94928260