Pinned Loading

1. SGLang-RadixMoE SGLang-RadixMoE Public

   Forked from sgl-project/sglang

   RadixMoE extends SGLang's Radix Cache to store expert activation patterns alongside KV cache entries, enabling zero-compute routing prediction for Mixture-of-Experts models.

   Python 1

2. vLLM-FlashMLA vLLM-FlashMLA Public

   Forked from vllm-project/vllm

   FlashMLA is a high-performance kernel library specifically optimized for Multi-Head Latent Attention (MLA) architectures and integrated in vLLM.

   Python 1