Is the bug of Skip download feature?

Hi, thanks for your great contributions to the community! I’ve been using DiffSynth for a long time.

After upgrading to v2.0.0, I ran into an issue during inference and I suspect there may be a bug in config.py?

I have downloaded the model weights locally and tried to load the model like this:

pipe = FluxImagePipeline.from_pretrained(
        torch_dtype=torch.bfloat16,
        device="cuda",
        model_configs=[
            ModelConfig(model_id="black-forest-labs/FLUX.1-Kontext-dev", origin_file_pattern="flux1-kontext-dev.safetensors", local_model_path=args.model_path, skip_download=True),
            ModelConfig(model_id="black-forest-labs/FLUX.1-Kontext-dev", origin_file_pattern="text_encoder/model.safetensors", local_model_path=args.model_path, skip_download=True),
            ModelConfig(model_id="black-forest-labs/FLUX.1-Kontext-dev", origin_file_pattern="text_encoder_2/model*.safetensors", local_model_path=args.model_path, skip_download=True),
            ModelConfig(model_id="black-forest-labs/FLUX.1-Kontext-dev", origin_file_pattern="ae.safetensors", local_model_path=args.model_path, skip_download=True),
        ],
    )

However, I got an error because self.path doesn’t seem to be passed correctly (it looks like the local path is not being applied as expected when skip_download=True). Could you please take a look, or let me know if I’m missing something? If this is indeed a bug, I’d be happy to submit a PR to fix it :).

The traceback was showed below:

Loading models from: null
Traceback (most recent call last):
  File "xxx/FLUX.1-Kontext-dev-inference.py", line 66, in <module>
    pipe = FluxImagePipeline.from_pretrained(
  File "xxx/diffsynth/pipelines/flux_image.py", line 130, in from_pretrained
    model_pool = pipe.download_and_load_models(model_configs, vram_limit)
  File "xxx/diffsynth/diffusion/base_pipeline.py", line 289, in download_and_load_models
    model_pool.auto_load_model(
  File "xxx/diffsynth/models/model_loader.py", line 66, in auto_load_model
    model_hash = hash_model_file(path)
  File "xxx/diffsynth/core/loader/file.py", line 118, in hash_model_file
    keys_dict = load_keys_dict(path)
  File "xxx/diffsynth/core/loader/file.py", line 71, in load_keys_dict
    if file_path.endswith(".safetensors"):
AttributeError: 'NoneType' object has no attribute 'endswith'

Start in diffsynth/core/loader/config.py and trace how ModelConfig.local_model_path and skip_download are handled into diffsynth/diffusion/base_pipeline.py and diffsynth/models/model_loader.py. Reproduce the provided FluxImagePipeline configuration and inspect the path passed to diffsynth/core/loader/file.py; done means local model files are used with skip_download=True without a None path reaching hash_model_file.