DeepGEMM: DeepSeek's FP8 GEMM implementation

Fields

id

sources/src5-deepgemm-github

URL

https://github.com/deepseek-ai/DeepGEMM

Publisher

DeepSeek

Kind

repo

Published

2025-01-01

Accessed

2026-10-04

notes

DeepGEMM is DeepSeek's implementation of FP8 GEMM optimized for DeepSeek models. Uses CUTLASS and Triton. Key for FP8 inference on Hopper/Blackwell. Demonstrates custom GEMM kernel development for specific model architectures.

Notes

DeepGEMM is DeepSeek's implementation of FP8 GEMM optimized for DeepSeek models. Uses CUTLASS and Triton. Key for FP8 inference on Hopper/Blackwell. Demonstrates custom GEMM kernel development for specific model architectures.

Cross-references

This record references no other records.