llama.cpp
09b4eab8 - Wire activation scales and disallow broadcasting over K

Commit

2 days ago

Wire activation scales and disallow broadcasting over K

References

0cc4m/fp8-scaled-mm

#28898 - ggml: support quant scales for fp8 and nvfp4

#29791 - vulkan: add fp8 and scaled matmul support

Author

ORippler

ORippler

Committer

ORippler

ORippler

Parents

Loading