jgrusewski ba8fb8eedd feat(ppo): wire symlog, adaptive entropy, percentile scaling into training loop ...

Integrate all PPO improvement modules into the core training paths:
- Symlog value predictions in compute_value_loss (MLP + LSTM)
- Adaptive entropy auto-tuning replaces fixed entropy_coeff
- Percentile P5/P95 advantage scaling for heavy-tailed returns
- DAPO clip_epsilon_high wired in all 7 PPOConfig construction sites
- Shape mismatch fix in adaptive_entropy (unsqueeze scalar)

2726 tests pass, 0 clippy errors.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-03 20:12:00 +01:00

..

infra

feat(infra): GPU training launcher with local/cloud routing

2026-02-23 11:19:58 +01:00

plans

feat(ppo): wire symlog, adaptive entropy, percentile scaling into training loop

2026-03-03 20:12:00 +01:00