Popular repositories Loading

1. MegaQwen MegaQwen Public

   🚀 Achieve faster Qwen3-0.6B inference with the MegaQwen CUDA megakernel, delivering 531 tok/s decode on RTX 3090—3.9x faster than HuggingFace.

   Cuda 1

2. testgit testgit Public

3. pogud.github.io pogud.github.io Public

   🚀 Accelerate Qwen3-0.6B inference with MegaQwen, a custom CUDA megakernel achieving 531 tok/s on RTX 3090, 3.9x faster than existing frameworks.