Change8

b9891

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

Summary

This release introduces the col2im_1d operation to the Metal backend across multiple precisions and refines the Metal backend's operator predicate checks for COL2IM_1D.

✨ New Features

  • Added col2im_1d operation support for Metal backend supporting f32, f16, and bf16 data types.

🐛 Bug Fixes

  • Aligned the GGML_OP_COL2IM_1D predicate on Metal with CPU, CUDA, and Vulkan backends by requiring a contiguous destination buffer and matching source/destination types.

Affected Symbols