Change8

b10201

📦 llama-cppView on GitHub →
1 features🐛 3 fixes🔧 2 symbols

Summary

This release includes improvements to the ggml-webgpu backend, specifically enhancing the flash_attn_vec function for quantized KV caches at long contexts. Several bugs and build issues have also been addressed.

✨ New Features

  • Improved flash_attn_vec for quantized KV cache at long contexts.

🐛 Bug Fixes

  • Fixed some bugs and comments related to flash_attn_vec.
  • Fixed v type check and some comments.
  • Fixed build error caused by rebasing.

Affected Symbols