[Feature]: Kimi K3 Performance Optimization

The feature, motivation and pitch

Tasks

• https://github.com/vllm-project/vllm/pull/50383 @jeejeelee
• https://github.com/vllm-project/vllm/pull/51146 @jeejeelee
• https://github.com/vllm-project/vllm/pull/50478 @robertgshaw2-redhat
• https://github.com/vllm-project/vllm/pull/50484 @GirasoleY
• https://github.com/vllm-project/vllm/pull/50510 @huangzhilin-hzl
• https://github.com/vllm-project/vllm/pull/50585 @yewentao256
• https://github.com/vllm-project/vllm/pull/50656 @tlrmchlsmth
• https://github.com/vllm-project/vllm/pull/50912 @yewentao256
• https://github.com/vllm-project/vllm/pull/51070 @yewentao256
• https://github.com/vllm-project/vllm/pull/51311 @yewentao256
• https://github.com/vllm-project/vllm/pull/51540 @njhill
• https://github.com/vllm-project/vllm/pull/51725 + https://github.com/vllm-project/vllm/pull/51726 @yewentao256
• https://github.com/vllm-project/vllm/pull/51772 @zyongye
• https://github.com/vllm-project/vllm/pull/52388 @yewentao256
• https://github.com/vllm-project/vllm/pull/52458 @BabyDrangoner
• https://github.com/vllm-project/vllm/pull/52789 @yewentao256 @ZeldaHuang + https://github.com/vllm-project/vllm/pull/53614 @ZeldaHuang
• https://github.com/vllm-project/vllm/pull/52993 @hnover-nv + https://github.com/vllm-project/vllm/pull/56082 @gcanlin
• https://github.com/vllm-project/vllm/pull/53053 + https://github.com/vllm-project/vllm/pull/52079 @gau-nernst
• https://github.com/vllm-project/vllm/pull/53152 @yewentao256 + https://github.com/vllm-project/vllm/pull/53327 @zyongye
• https://github.com/vllm-project/vllm/pull/53168 @gcanlin
• https://github.com/vllm-project/vllm/pull/53396 @gcanlin
• https://github.com/vllm-project/vllm/pull/53524 @mingg26
• https://github.com/vllm-project/vllm/pull/53525 @mingg26
• https://github.com/vllm-project/vllm/pull/53534 @gcanlin
• https://github.com/vllm-project/vllm/pull/53556 @gcanlin
• https://github.com/vllm-project/vllm/pull/53794 @wzhao18
• https://github.com/vllm-project/vllm/pull/53942 @yewentao256
• https://github.com/vllm-project/vllm/pull/54088 @yewentao256
• https://github.com/vllm-project/vllm/pull/54015 @gau-nernst
• https://github.com/vllm-project/vllm/pull/54151 @gau-nernst
• https://github.com/vllm-project/vllm/pull/54168 @zyongye
• https://github.com/vllm-project/vllm/pull/54254 @rebklee
• https://github.com/vllm-project/vllm/pull/54255 @djmmoss
• https://github.com/vllm-project/vllm/pull/54261 @zyongye
• https://github.com/vllm-project/vllm/pull/54347 @gcanlin
• https://github.com/vllm-project/vllm/pull/54447 @gcanlin
• https://github.com/vllm-project/vllm/pull/54565 @yewentao256
• https://github.com/vllm-project/vllm/pull/54606 @BolinSNLHM
• https://github.com/vllm-project/vllm/pull/54697 @zyongye
• https://github.com/vllm-project/vllm/pull/54896 @zyongye
• https://github.com/vllm-project/vllm/pull/55242 @wzhao18
• https://github.com/vllm-project/vllm/pull/55356 @yewentao256
• https://github.com/vllm-project/vllm/pull/55364 @wzhao18
• https://github.com/vllm-project/vllm/pull/56159 @yewentao256
• https://github.com/vllm-project/vllm/pull/56510 @gcanlin

For Rocm: https://github.com/vllm-project/vllm/issues/50682

Alternatives

No response

Additional context

I don't have enough GPU resources to run an end-to-end benchmark currently. If you have available GPU capacity and are willing to share access, I would greatly appreciate it!

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.