llama.cpp
18eb236a
- Restrict (fused) MMVF to 2*ts alignment, and cublas FP8 fo 16-byte
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
17 hours ago
Restrict (fused) MMVF to 2*ts alignment, and cublas FP8 fo 16-byte
References
#28413 - CUDA: Add OCP FP8 support
#28898 - ggml: support quant scales for fp8 and nvfp4
#29791 - vulkan: add fp8 and scaled matmul support
Author
ORippler
Committer
ORippler
Parents
2217433b
Loading