New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[DO NOT MERGE] patch wvSpltK_fused_moe from https://github.com/amd-hhashemi/vllm/tre… #126

Open

carlushuang wants to merge 2 commits into main from wvSpltK_fused_moe

carlushuang commented Aug 10, 2024

This PR originally from https://github.com/amd-hhashemi/vllm/tree/wvSpltK_fused_moe, commit: 2f842d4

# this PR can pass below test
from vllm import LLM, SamplingParams
sampling_params = SamplingParams(temperature=0, top_k=1)
llm = LLM("/mnt/nvme0n1/models/AI-ModelScope/Mixtral-8x7B-Instruct-v0___1", tensor_parallel_size=1,dtype="float16")
print(llm.generate("What is batch inference?", sampling_params))

valarLip added 2 commits

August 10, 2024 22:27


          patch wvSpltK_fused_moe from https://github.com/amd-hhashemi/vllm/tre…

9e956fb

…e/wvSpltK_fused_moe


          add VLLM_MOE_MFMASWIZZLE_M_THRSHLD env, default 32, means only M less…

39fec6f

… equals 32 will run mega_fused_moe_kernel

carlushuang mentioned this pull request

[DO NOT MERGE] Vinayak/moe final hashem #127

Open

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet