on-policy-self-distillation

gpt-5.6-sol

on-policy-self-distillation — graded by the task's verifier against a measured baseline.

backend: modal
variant: codex azure_ai/gpt-5.6-sol
metric: AIME24 avg@12 accuracy
actions: 428
duration: 9h 21m
recorded: 2026-08-02

run idopsd-aime24-4gpus__eNDoWrD

Everything gpt-5.6-sol did on this task, in order.

download JSON ↗

codexgpt-5.6-sol444 turns16 narration428 reasoning428 action428 observation1 final

#151

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "62",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 62
Wall time 61.0 seconds
Output:

#152

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "62",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script completed
Wall time 54.6 seconds
Output:
Warning: truncated output (original token count: 384)
Total output lines: 30

ok/s
vLLM generation done - elapsed time: 6.39s, prompts: 4, total tokens: 4346, avg length: 1086.5, speed: 680.3 tok/s

cated…: 4, total tokens: 7156, avg length: 1789.0, speed: 1050.0 tok/s

 29%|…284 tokens truncated… 64%|██████▍   | 64/100 [16:19<09:16, 15.46s/it]
ens truncated…000000003e-06, 'on_policy_loss': -0.0077, 'epoch': 0.07}

 68%|██████▊   | 68/100 [17:22<08:18, 15.58s/it]

#153

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolexec

args

{
  "input": "let out=\"\";for(let i=0;i<9;i++){const r=await tools.write_stdin({session_id:71719,chars:\"\",yield_time_ms:60000,max_output_tokens:80});out+=r.output.slice(-150)+\"\\n\";if(r.exit_code!==undefined){out+=`EXIT${r.exit_code}`;break;}}text(out);\n"
}

exec result

Script running with cell ID 63
Wall time 11.0 seconds
Output:

#154

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#155

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#156

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#157

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#158

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#159

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#160

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 63
Wall time 61.0 seconds
Output:

#161

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "63",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script completed
Wall time 8.6 seconds
Output:
Warning: truncated output (original token count: 362)
Total output lines: 24

cated… - elapsed time: 6.77s, prompts: 4, total tokens: 6462, avg length: 1615.5, speed: 954.5 tok/s

 73%|███████▎  | 73/100 [18:40<07:00, 15.59s/it]
ngth: 1724.8, speed: 1014.4 tok…262 tokens truncated…███| 100/100 [25:41<00:00, 15.60s/it]
 see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
TRAIN_CKPT=/workspace/runs/long2048/capped_seed42/checkpoint-100

EXIT0

#162

narration

The long-horizon run completed cleanly. Its average rollout grew from roughly 850 tokens to 1,300–1,800 on the formerly truncated batches, so it is genuinely learning from later reasoning and final-answer regions rather than just padding the same trajectories. I’m evaluating it on the full benchmark; no partial result will be used for this decision.

#163

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolexec

args

{
  "input": "const r = await tools.exec_command({cmd:\"bash /workspace/dev_eval.sh /workspace/runs/long2048/capped_seed42/checkpoint-100\",\"workdir\":\"/workspace\",\"yield_time_ms\":1000,\"max_output_tokens\":500});\ntext(r.output); if(r.session_id) text(`SESSION_ID=${r.session_id}`);"
}

exec result

Script completed
Wall time 1.2 seconds
Output:
SESSION_ID=34388

#164

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolexec

args

{
  "input": "let out=\"\";for(let i=0;i<12;i++){const r=await tools.write_stdin({session_id:34388,chars:\"\",yield_time_ms:60000,max_output_tokens:50});out+=r.output.slice(-100)+\"\\n\";if(r.exit_code!==undefined){out+=`EXIT${r.exit_code}`;break;}}text(out);\n"
}

exec result

Script running with cell ID 65
Wall time 11.0 seconds
Output:

#165

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#166

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#167

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#168

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#169

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#170

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#171

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#172

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#173

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#174

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script running with cell ID 65
Wall time 61.0 seconds
Output:

#175

reasoning (redacted)

encrypted chain-of-thought — recorded, not exposed

toolwait

args

{
  "cell_id": "65",
  "yield_time_ms": 60000,
  "max_tokens": 100
}

wait result

Script completed
Wall time 55.8 seconds
Output:
Warning: truncated output (original token count: 241)
Total output lines: 12

 prompts:   0%|          | 0/96 [00:00<?, ?it/s, est. speed input: 0.00 toks/s, output: 0.00 toks/s]


12%|█▎        | 12/96 [03:19<23:17, 16.63s/it, est. speed input: 9.38 toks/s, output: 434.92 …141 tokens truncated…▋    | 48/84 [10:14<06:17, 10.48s/it, est. speed input: 6.58 toks/s, output: 1202.44 toks/s]
%|█▍        | 12/84 [10:45<1:04:32, 53.78s/it, est. speed input: 1.56 toks/s, output: 228.99 toks/s]