Transformers now runs GGUF at llama.cpp speed via GGML kernels

LysandreJik · x · 2026-09-22

Hugging Face Transformers has supported loading GGUF files for a few years, but only by unquantizing them, which hurt performance. Thanks to work by @marcsun, Transformers can now run GGUF models through GGML kernels via the kernels library, matching llama.cpp performance. Lysandre credits the @ggmlorg team for building these kernels.

Related event: Transformers Now Calls GGML Kernels Directly, Matching llama.cpp Performance for GGUF(6 posts)→

Original post →

More from coding & agent

coding & agent channel →