Lu Fang

Results 87 comments of Lu Fang

support of __float2bfloat162_rn

CUDA provided `__float2bfloat162_rn`, however, hip doesn't have the corresponding definition. So when hipifying the code, it cannot compile.

[Core] Support Local Chunked Attention for Hybrid KV Cache

Please don't forget to fill the task description :-)

[Core] Support Local Chunked Attention for Hybrid KV Cache

Rerunning the test and see if the problem is transient

[torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass

Hi @zou3519 , could you also help review on the torch.compile pass part? Thanks.

[custom_op][vllm-plugin] update custom_op class to use op_registry

@simon-mo is out this week, maybe @youkaichao or @WoosukKwon ?

[Bug]: 0.8.5 部署qwen-vl模型报错，降级0.8.4没问题

Try install the nightly build? Something like ``` pip install vllm --pre --extra-index-url https://wheels.vllm.ai/nightly ```

[v1] Re-add fp32 support to v1 engine through FlexAttention

Please fix the linter