Change8

b8787

📦 llama-cppView on GitHub →
1 features🐛 4 fixes🔧 1 symbols

Summary

This release focuses on updates to the ggml-webgpu backend, specifically improving matmul accumulation precision and fixing several related bugs across different platforms.

✨ New Features

  • Updated ggml-webgpu register tiling matmul to use f32 accumulation.

🐛 Bug Fixes

  • Fixed profiling code related to ggml-webgpu.
  • Fixed register tiling matmul for Chrome (attributed to dawn issues).
  • Fixed use of new load function in ggml-webgpu context.
  • Applied compile fixes.

Affected Symbols