b10000

ggml-org/llama.cppb10000Jul 14, 2026by github-actions[bot]same day as b9999

What changed since b9999

low risk

b10000 refactors ggml im2col destination-type casting across all convolution ops and fixes im2col_f16 kernel-type handling.

Improvements

  • im2col destination-type casting is now uniform across all convolution operations, improving internal consistency.
  • im2col_f16 now accepts any kernel type instead of a restricted subset.

Bug fixes

  • Corrected im2col casting logic so f16 im2col no longer rejects valid kernel types.

Upgrade notes

This is a small ggml-internal change with no config or API surface changes reported; a drop-in binary swap should suffice. Rebuild if you compile from source, since the change touches ggml/src/ggml.c. No migration steps are required.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Uniformizes the im2col destination type casting logic across all convolution operations to improve consistency and fixes casting logic to allow any kernel type.

Key Highlights

  • Uniformized im2col casting logic across all convolution operations.
  • Fixed im2col_f16 to accept any kernel type.
  • Released standard build binaries for multiple platforms.

Release Notes (excerpt)

<details open> ggml: uniformize im2col dst_type for all conv ops (#23660) * ggml: uniformize im2col dst_type for all conv ops * Update ggml/src/ggml.c Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * ggml : uniformize im2col casting logic across all conv ops * fix : allow im2col_f16 to accept any kernel type --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> </details> **macOS/iOS:** - macOS Apple Silicon (arm64)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories