45m ago · llama.cpp releases →
ggml-backend : start a new split at an incompatible weight in a…
Includes: b12378: mtmd : avoid padding-only Whisper audio chunks (#30403), b12379: jinja : reduce template rendering overhead (#32023), b12380: mtmd : allow custom video decoder (#32188), llama.cpp b12381, llama.cpp b12383