llama.cpp has released version b11398, which includes support for BF16/FP16/FP32 K tails in tinyBLAS on x86. This update is aimed at developers and researchers working with machine learning models on x86 architectures.
What happened
llama.cpp, an open-source project, has released version b11398. This release adds support for BF16/FP16/FP32 K tails in tinyBLAS on x86, enhancing performance and compatibility for machine learning models. The release also includes vectorization of BF16 K tails in tinyBLAS and skips tinyBLAS when use_ref is enabled for CPU tests. The project aims to provide a versatile and efficient framework for developers and researchers working with machine learning on various hardware platforms.
What to weigh
- The release includes support for BF16/FP16/FP32 K tails in tinyBLAS on x86.
- The release also vectorizes BF16 K tails in tinyBLAS.
Source: llama.cpp