GitLab

The numa node location would be calculated based on the total number
of worker threads.
So we should always use the actual number of threads instead of using a min() op.

📝 update benchmark.md

[update] Update doc.

⚡ release v0.2.2rc1

[release]  Release 0.2.2rc.

[update] Update readme.

 Update README.md

📝 add benchmark.md

⚡ update git ignore add docker dev container

Feat absorb for long prefill

[feat] Support fp8 linear kernel;

musa: support bf16

Ensure backward compatibility with PyTorch 2.2

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

fix KExpertsMarlin on GPU with out CUDA Graph

support moonlight, use ktransformers/optimize/optimize_rules/Moonlight-16B-A3B.yaml