The retry logic for model loading was duplicated identically across three from_huggingface methods in EasyPipelineForText2Image, EasyPipelineForImage2Image, and EasyPipelineForInpainting.

Changes

• Extracted _load_pipeline_with_retry helper function that encapsulates:

  • Candidate index and failed count tracking
  • HuggingFace search with retry on failure
  • Pipeline loading from single file or pretrained checkpoint
  • Recursive retry logic with configurable max_retries

• Updated all three from_huggingface methods to delegate to the helper, passing only:

  • Pipeline class
  • Model identifier
  • Pipeline mapping (TEXT2IMAGE, IMAGE2IMAGE, or INPAINT)
  • Forwarded kwargs

Impact

Reduces code by ~93 lines. Future changes to retry behavior now require updates in a single location.

# Before: 67 lines of identical retry logic in each method
def from_huggingface(cls, pretrained_model_link_or_path, **kwargs):
    kwargs.update(_status)
    candidate_index = kwargs.pop('_candidate_index', 0)
    # ... 60+ more lines of retry logic

# After: 7 lines delegating to shared helper
def from_huggingface(cls, pretrained_model_link_or_path, **kwargs):
    kwargs.update(_status)
    return _load_pipeline_with_retry(
        cls, pretrained_model_link_or_path,
        SINGLE_FILE_CHECKPOINT_TEXT2IMAGE_PIPELINE_MAPPING, **kwargs
    )

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.