●Change8

v2.29.0

Breaking Changes
📦 localaiView on GitHub →
⚠ 3 breaking✨ 6 features🔧 1 symbols

Summary

LocalAI v2.29.0 introduces a major overhaul to container images, slimming down defaults and introducing an \`-extras\` suffix for optional dependencies, alongside new model support and experimental video generation capabilities.

⚠️ Breaking Changes

  • Container images containing extra Python dependencies (like those for `diffusers`) now require the \`-extras\` suffix (e.g., \`latest-gpu-nvidia-cuda-12-extras\`). Default images are now slimmer and do not include these dependencies.
  • The separate \`-ffmpeg\` container image tags have been removed because FFmpeg is now included in all core images. Users should switch to the corresponding base image tag (e.g., \`latest-gpu-hipblas-ffmpeg\` becomes \`latest-gpu-hipblas\`).
  • The AutoGPTQ backend has been dropped because it was discontinued upstream.

Migration Steps

  1. If you relied on extra Python dependencies (like diffusers) in your container, update your image pull command to use the new \`-extras\` suffix (e.g., change \`latest-gpu-nvidia-cuda-12\` to \`latest-gpu-nvidia-cuda-12-extras\` if you need those dependencies).
  2. If you were using an image tag ending in \`-ffmpeg\`, remove \`-ffmpeg\` from the tag (e.g., change \`latest-gpu-hipblas-ffmpeg\` to \`latest-gpu-hipblas\`).
  3. If you were using the AutoGPTQ backend, you must switch to an alternative backend as it has been removed.

✨ New Features

  • Official support for the Qwen3 model family has been integrated.
  • Experimental Auto GPU Offload feature added, attempting to automatically detect GPUs and configure optimal layer offloading for llama.cpp and CLIP.
  • Whisper.cpp now supports GPU acceleration via cuBLAS (NVIDIA) and Vulkan.
  • Introduction of an experimental \`/video/generations\` endpoint.
  • The \`install.sh\` script now includes a \`--uninstall\` flag for easy removal.
  • Expanded Hipblas targets to include: \`gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102\`.

🔧 Affected Symbols

AutoGPTQ backend