vllm-qwen by hec-ovi

vLLM Qwen 3.6-27B (AWQ-INT4) + DFlash speculative decoding on AMD Strix Halo (gfx1151 iGPU, 128 GB UMA, ROCm 7.13). 24.8 t/s single-stream, vision, tool calling, 256K context, OpenAI-compatible, Docker. Matches DGX Spark FP8+DFlash+MTP at a third of the cost. No CUDA. Maintained by hec-ovi on GitHub, where it has earned 50 stars from the community.

DGX Spark / GB10 vLLM Docker stack for large-model serving, presets, patches, and validation notes. Maintained by bjk110 on GitHub, where it has earned 54 stars from the community.

A "standard library" of Triton kernels. Maintained by stackav-oss on GitHub, where it has earned 26 stars from the community.