Inference with the new trained weight of i2v model?

[zero1778]

Hi,

I actually modified the CogVideoX module and fine-tuned it using the train_zero_i2v.sh script and corresponding configuration. Now I have the model weights in the following structure as per your instructions:

81_HDTF_sft_fp32/
├── model-00001-of-00004.safetensors
├── model-00002-of-00004.safetensors
├── model-00003-of-00004.safetensors
├── model-00004-of-00004.safetensors
├── model.safetensors.index.json
├── pytorch_model-00001-of-00004.bin
├── pytorch_model-00002-of-00004.bin
├── pytorch_model-00003-of-00004.bin
├── pytorch_model-00004-of-00004.bin
├── pytorch_model.bin.index.json

When I try to use these weights for inference with the cli_demo.py script, I encounter the following error at the line:

CogVideoXImageToVideoPipeline.from_pretrained(model_path, torch_dtype=dtype)

The error message is:

OSError: Error no file named model_index.json found in directory /home/CogVideo/output/81_HDTF_sft_fp32

Could you please provide guidance on how to correctly use this weight format for inference? Your help would be greatly appreciated. Thanks!