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