llama.cpp 1.0 Drops 35 Deprecated APIs, Adopts Semantic Versioning
llama.cpp 1.0.0 removes 35 deprecated functions from its public C API, adopts semantic versioning where only breaking API changes trigger major versions, adds FP8 KV cache support, MoE CPU offloading, new model integrations including Kimi-K3.1-Pro and drama-3, and provides pre-built binaries for 150+ architecture variants across major platforms.
On November 13, 2024 (UTC+8), llama.cpp released version 1.0.0 , and its underlying tensor library ggml also reached 1.0.0 . The release notes state that the public header llama.h had 35 deprecated functions removed , including llama_set_warmup and a defragmentation switch.
According to the Hugging Face announcement, both libraries now follow semantic versioning : the major version increments only when the public C API becomes incompatible; minor versions add compatible features, backends, new model support, and new APIs; patch versions contain fixes that do not change the interface. This policy exists because downstream projects already compile against these headers.
A version-rule table (shown in the release notes) maps change types to version bumps. The major version is determined solely by the public C headers under llama.cpp/include/ (for libllama) and ggml/include/ (for libggml). The project has been in development for 3.5 years.
Key API and Feature Changes
The speculative-decoding NextN interface moved from an extension header into llama.h. llama_set_embd_state_src was renamed to llama_set_ctx_state_src.
KV cache can now be stored as FP8 (type f8_e4m3), controlled via the -ctk and -ctv flags.
Mixture-of-experts (MoE) cache can allocate a portion on CPU; cache misses are fetched on demand while hits are served immediately without waiting.
New Model Integrations
Text models: Kimi-K3.1-Pro, Ripple-65B-Preview, xing4_0.
Voice model: drama-3.
Server: Added OpenAI-compatible /v1/audio/speech endpoint.
Vision models: MageVL-Cowork and Kimi-K3.1 vision.
Web conversations can now be grouped.
Distribution and Platform Support
Pre-release packages cover Linux, macOS, Windows, Android, and iOS. Distribution channels include Docker, Debian, Homebrew, conda-forge, MacPorts, and Winget. The project currently supports over 150 architecture variants and includes mainstream speculative-decoding methods. Source code is hosted at https://github.com/ggml-org/llama.cpp.
Sources: llama.cpp v1.0.0 release notes, Nov 13: https://github.com/ggml-org/llama.cpp/releases/tag/v1.0.0 Hugging Face announcement, Nov 13: https://huggingface.co/blog/ggml-org/llama-cpp-v1
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.