Tags

Tags give the ability to mark specific points in history as being important
  • v0.1.1

    v0.1.1
    
    - Add Python 3.12, Torch 2.5.1 CPU and CUDA 12.4 base tuples for Audiolla, with separate hash-locked dependencies and image tag shapes.
    - Honor tuple selection in GPU build tests and verify real tensor operations on CPU and CUDA.
    - Document matching downstream Torchaudio installation and the retained Torch checkpoint-loading vulnerability.
    - Grant the reusable code workflow permission to upload security results.
  • v0.1.0

    v0.1.0
    
    Initial release.
    
    - Added a digest-pinned PyTorch base-image matrix with Python 3.12 and 3.13 CPU, CUDA 12.6, and CUDA 13.0 tuples.
    - Added one generic Dockerfile, one validated tuple registry, and generated Docker workflow targets for future Python, CUDA, and Torch combinations.
    - Added locked CPU and CUDA dependencies, non-root runtime images, local CPU and GPU image verification, and Docker Buildx cache export.