🚀 Day 3 of #OpenSourceWeek: DeepGEMM Introducing DeepGEMM - an FP8 GEMM library that supports both dense and MoE GEMMs, powering V3/R1 training and inference. ⚡ Up to 1350+ FP8 TFLOPS on Hopper GPUs ✅ No heavy dependency, as clean as a tutorial ✅ Fully Just-In-Time compiled ✅ Core logic at ~300 lines - yet outperforms expert-tuned kernels across most matrix sizes ✅ Supports dense layout and two MoE layouts 🔗 GitHub: t.co/cxJ55w61pT
- Views
- 978,294
- Likes
- 6,322
- Replies
- 465
- Reposts
- 851
- Quotes
- 181
- Bookmarks
- 863
🚀 Day 3 of : DeepGEMM
The post announces DeepGEMM during OpenSourceWeek. It highlights the library’s FP8 performance, low dependency footprint, compilation method, compact codebase, and support for dense and MoE layouts, then links to GitHub.
- Views
- 978.3K
- Likes
- 6.3K
- Replies
- 465
- Reposts
- 851
- Quotes
- 181
- Bookmarks
- 863
- Topics
Analysis
The post announces DeepGEMM during OpenSourceWeek. It highlights the library’s FP8 performance, low dependency footprint, compilation method, compact codebase, and support for dense and MoE layouts, then links to GitHub.
Sign in to see the full analysis.
Sign in🚀 Day 3 of #OpenSourceWeek: DeepGEMM Introducing DeepGEMM - an FP8 GEMM library that supports both dense and MoE GEMMs, powering V3/R1 training and inference. ⚡ Up to 1350+ FP8 TFLOPS on Hopper GPUs ✅ No heavy dependency, as clean as a tutorial ✅ Fully Just-In-Time compiled ✅ Core logic at ~300 lines - yet outperforms expert-tuned kernels across most matrix sizes ✅ Supports dense layout and two MoE layouts 🔗 GitHub: https://t.co/cxJ55w61pT
🚀 Day 3 of #OpenSourceWeek: DeepGEMM
- 01The Event Hook
Announce the day or stage of an event, then reveal the featured project or release.
- 02The Introduction
Name the release and explain what it is used for.
- 03The Feature List
Give a fast list of measurable benefits, implementation details, and supported options.
- 04The Link CTA
Point readers to the place where they can explore or access the release.