JoJo-Lorray

Megatron training adaptation issue, obtaining CUDA_SOME as None

1

### 🐛 Describe the bug Megatron training adaptation issue, obtaining CUDA_SOME as None ```python # running megatron training, loading fused kernels reports an error, the core code is as follows...

Fix: add model parameter and improve response parsing for vllm v1/completions

1

Tested with vllm 0.7.3 and confirmed compatibility with both new (choices-based) and legacy (top-level text) response formats. Fix: add model parameter and improve response parsing for v1/completions # Description Add...