Popular repositories Loading

1. qwen3.5-fp8-cuda qwen3.5-fp8-cuda Public

   Qwen3.5-4B FP8 推理引擎 · 手写 CUDA 融合算子 · 54 tok/s on RTX 5060

   Python

2. ai-inference-lab ai-inference-lab Public

   LLM inference engineering on a single 8 GB GPU: FP8/INT4-AWQ quantization, a 200-question fairness-controlled quality eval, vLLM tuning, a FastAPI gateway, and a verified 7.49 GiB offline bundle.

   Python