Change8

b9143

📦 llama-cppView on GitHub →
🐛 1 fixes

Summary

This release primarily addresses a numerical stability issue by ensuring intermediate calculations use float casting to avoid operator ambiguity. It also provides numerous pre-compiled binaries for various operating systems and hardware configurations.

🐛 Bug Fixes

  • Fixed issue #22974 by casting intermediate results to float before addition and then casting the final result to the destination type, resolving half+half operator ambiguity.