llama.cppProduct update

llama.cpp · v0.6.0

llama.cpp v0.6.0 introduces llama_batch_ext, an extended batch API with llama_process() that mixes tokens and embeddings in one batch and carries per-token state embeddings for MTP and deepstack models. Newly supported models include GLM-5.3-Flash (GLM5-Next), a 320B text and vision hybrid, and the Clef decision model, while Qwen4Exp gains MTP speculative decoding and the server adds a /v1/systemone endpoint for decision models. On Metal there is a tensor API flash attention kernel for F16 KV, and ggml moves to v0.26.0.

GitHub ReleasesOct 6, 202617 min read