Zhewen Li comments

Repositories
Issues
Comments

Results 3 comments of


                                            Zhewen Li

[Core] Integrate Fastsafetensor loader for loading model weights

@manish-sethi / @njhill This PR needs to skip tests on AMD, fixing in https://github.com/vllm-project/vllm/pull/27612

[Bug fix] ROCm FlashAttention: add missing `full_scales` argument to Triton wrapper

fixed by https://github.com/vllm-project/vllm/commit/8e4b351a0c9e414b0c56c32cbdef51a21d1ea1be?fbclid=IwZXh0bgNhZW0CMTEAYnJpZBExT0JVeTF6emF0SWRaNUU0WAEegqoGNk7T1CxCHX02o88944Fc9floNtJCcbPwnqQXqkmhllh4IHT8cXbdL7w_aem_Jn0FOH8jEbpGlaw77nBNkw

[Kernel][Moe Configs] llama4 maverick fp8 moe config tp8 on mi325

@yeqcharlotte just copied from mi300 since they share a same architecture