【Bug已解决】Optimize old_per_token_logps recomputation in GRPOTrainer: per-rollout window check instead
【Bug已解决】Optimize old_per_token_logps recomputation in GRPOTrainer: per-rollout window check instead of static modulo 解决方案.md原始报错:Optimize old_per_token_logps recomputation in GRPOTrainer: per-rollout window check instead of static m…
2026/7/22 7:42:45