vLLM(@vllm_project):Training and rollout logprobs matched bit for bit on ROCm. The @RLKernel team integrated RL-Align/RL-Kernel with vllm-project/vime. A 200-step Qwen3-8B GRPO run on 8× @AMD MI300X recorded zero logprob mismatches between Megatron training and vLLM rollout. The strict path aligns reduction order, intermediate precision, rounding points, and math primitives across both sides. Deep dive:

vLLM

@vllm_project

A high-throughput and memory-efficient inference and serving engine for LLMs. Join to discuss together with the community!

加入 March 2024

36 正在關注 50.2K 粉絲

vLLM@vllm_project

2026.09.24 15:06

Training and rollout logprobs matched bit for bit on ROCm. The @RLKernel team integrated RL-Align/RL-Kernel with vllm-project/vime. A 200-step Qwen3-8B GRPO run on 8× @AMD MI300X recorded zero logprob mismatches between Megatron training and vLLM rollout. The strict path aligns reduction order, intermediate precision, rounding points, and math primitives across both sides. Deep dive: