Build b12541 sped up creation of IQ2 quantized weights by caching intermediate work. The change applies when preparing low-bit models for distribution or local use.
Tool / RuntimeLLMCross-Modal
llama.cpp
Also known as: llamacpp · llama-server
Plain English
llama.cpp runs language models on local machines. It can be used from the command line, embedded in another program, or run as an API server.
Chronological Entries (3)
#3Release
AI-Assisted
#2Release
AI-Assisted
The llama.cpp maintainers announced native MXFP4 support for OpenAI gpt-oss. Supported CPU and GPU backends could run the model with its original low-precision expert weights in GGUF.
#1Milestone
AI-Assisted
The llama.cpp project merged a public C API for its multimodal library. Applications embedding the runtime gained a common interface for preparing and evaluating image inputs alongside text on supported models.
Direct Relations (2)
- Part ofInference Engines
- UsesGGUF