Popular repositories Loading

1. mi210-llm-stack mi210-llm-stack Public

   AMD MI210 (gfx90a/CDNA2) LLM inference optimization — TurboQuant, KIVI, per-layer KV types, FlashAttention, MoE expert caching

   Shell 15

2. aiter-cdna2 aiter-cdna2 Public

   Run AMD AITER's hand-written ASM kernels on CDNA2 / gfx90a (MI210, MI250). 242 of 1,422 kernels translated, with the tests and benchmarks to prove which actually run.

   Python 11 1

3. vllm-expert-cache vllm-expert-cache Public

   vLLM plugin: device-side LRU/LFU expert cache for MoE offload. Keep N experts resident in VRAM, stream the rest from host RAM.

   Python 8 1

4. r9700-lru-expert-cache r9700-lru-expert-cache Public

   Device-side LRU expert cache + kernel-count patches: Qwen3.8-Flash-Next decode on 2x AMD R9700 (gfx1201), ROCm 10, tcclaviger vLLM fork

   Python 6 2

5. mi210-vllm mi210-vllm Public

   Pinned, gate-verified vLLM deployment stack for AMD MI210 (gfx90a/CDNA2)

   Python 5

6. cve-af-alg-block cve-af-alg-block Public

   Shell 3