Repository navigation

[FEAT] Rollout outputs: selected-token logprobs, effective seeds and stop token IDs (part of #70) - #83

Open

MinhaoLi0318 wants to merge 12 commits into

InferMatrixLab:mainfrom

MinhaoLi0318:rl/rollout-outputs

Open

[FEAT] Rollout outputs: selected-token logprobs, effective seeds and stop token IDs (part of #70)#83
MinhaoLi0318 wants to merge 12 commits into
InferMatrixLab:mainfrom
MinhaoLi0318:rl/rollout-outputs

Commits on Oct 1, 2026
- Expose training rollout scores and full policy publication
  0z5a
  committed
  f6cfefe
- Report entropy seeds for reproducible training rollouts
  0z5a
  committed
  e591bda
- Measure selected-score cost and replay semantics on Thor
  0z5a
  committed
  bdce287
- Keep training contract review focused on the public API
  0z5a
  committed
  08d1b8d
Commits on Oct 13, 2026
- Merge main and adapt training contract to shared sampler
  0z5a
  committed
  dd62ef4
Commits on Oct 21, 2026
- Merge main and retain training publication on stage owners
  0z5a
  committed
  0fb1632
Commits on Oct 26, 2026
- Merge main and keep engine passthrough with entry model configuration
  0z5a
  committed
  e6fae1e
Commits on Oct 27, 2026