Change8

b10246

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

Summary

This release includes an optimization for OpenCL routing of large q6_K lm_head weights to improve performance. It also provides pre-compiled binaries for various platforms and hardware configurations.

🐛 Bug Fixes

  • Optimized OpenCL routing for large q6_K lm_head weights by adding a direct size condition, improving performance for models like gemma-4 E2B that have large dimensions but don't meet the original dimension condition for gemv_noshuffle.

Affected Symbols