Bitwise Consistent On-Policy Reinforcement Learning with VLLM and TorchTitan

1 points | by brrrrrm 7 hours ago