5 Releases published by 1 person

• v0.3.15-metal

  published Aug 7, 2025

• v0.3.15-cu124

  published Aug 7, 2025

• v0.3.15-cu121

  published Aug 7, 2025

• v0.3.15-cu123

  published Aug 7, 2025

• v0.3.15-cu122

  published Aug 7, 2025

1 Pull request merged by 1 person

• fix: rename op_offloat to op_offload in llama.py

  #2046 merged Aug 7, 2025

2 Issues opened by 2 people

• add support for MXFP4 quantization to enable use of new gpt-oss models by OpenAI

  #2048 opened Aug 6, 2025

• Build fails on Windows with non-CUDA backends (CLBlast, Vulkan) for versions >= 0.2.78

  #2047 opened Aug 3, 2025

7 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Building and installing llama_cpp from source for RTX 50 Blackwell GPU

  #2028 commented on Aug 2, 2025 • 0 new comments

• Cannot run T5-based models

  #1587 commented on Aug 3, 2025 • 0 new comments

• Windows11：ERROR: ERROR: Failed to build installable wheels for some pyproject.toml based projects (llama-cpp-python)

  #2035 commented on Aug 3, 2025 • 0 new comments

• Can't install with GPU support with Cuda toolkit 12.9 and Cuda 12.9

  #2013 commented on Aug 3, 2025 • 0 new comments

• Pre-built cpu wheel does not work on Ubuntu due to libc.musl dependency

  #1628 commented on Aug 6, 2025 • 0 new comments

• Regression in unified KV cache appears after `llama.cpp` release b5912 in b5913

  #2045 commented on Aug 7, 2025 • 0 new comments

• feat: Add Gemma3 chat handler (#1976)

  #1989 commented on Aug 5, 2025 • 0 new comments