deepseek-ai/deepgemm v2.1.1.post3 is the latest release of a CUDA kernel library that compiles and runs FP8, FP4, and BF16 matrix multiplication and MoE operations at runtime on NVIDIA GPUs.
deepseek-ai/deepgemm v2.1.1.post3 is the latest release of a CUDA kernel library that compiles and runs FP8, FP4, and BF16 matrix multiplication and MoE operations at runtime on NVIDIA GPUs.