Error while running the demo : invalid load key, '<'

[dimwael]

I am running the 13B version on a sagemaker instance.

!python demo.py --cfg-path eval_configs/minigpt4_eval.yaml --gpu-id 0

Initializing Chat Loading VIT Loading VIT Done Loading Q-Former Loading Q-Former Done Loading LLAMA

===================================BUG REPORT=================================== Welcome to bitsandbytes. For bug reports, please submit your error trace to: https://github.com/TimDettmers/bitsandbytes/issues

Loading checkpoint shards: 100%|██████████████████| 3/3 [02:46<00:00, 55.49s/it] Loading LLAMA Done Load 4 training prompts Prompt Example

Human: Please provide a detailed description of the picture. ###Assistant:

Load BLIP2-LLM Checkpoint: pretrained_minigpt4.pth ╭───────────────────── Traceback (most recent call last) ──────────────────────╮ │ /home/ec2-user/SageMaker/MiniGPT-4/demo.py:60 in │ │ │ │ 57 model_config = cfg.model_cfg │ │ 58 model_config.device_8bit = args.gpu_id │ │ 59 model_cls = registry.get_model_class(model_config.arch) │ │ ❱ 60 model = model_cls.from_config(model_config).to('cuda:{}'.format(args.g │ │ 61 │ │ 62 vis_processor_cfg = cfg.datasets_cfg.cc_sbu_align.vis_processor.train │ │ 63 vis_processor = registry.get_processor_class(vis_processor_cfg.name).f │ │ │ │ /home/ec2-user/SageMaker/MiniGPT-4/minigpt4/models/mini_gpt4.py:265 in │ │ from_config │ │ │ │ 262 │ │ ckpt_path = cfg.get("ckpt", "") # load weights of MiniGPT-4 │ │ 263 │ │ if ckpt_path: │ │ 264 │ │ │ print("Load BLIP2-LLM Checkpoint: {}".format(ckpt_path)) │ │ ❱ 265 │ │ │ ckpt = torch.load(ckpt_path, map_location="cpu") │ │ 266 │ │ │ msg = model.load_state_dict(ckpt['model'], strict=False) │ │ 267 │ │ │ │ 268 │ │ return model │ │ │ │ /home/ec2-user/anaconda3/envs/pytorch_p39/lib/python3.9/site-packages/torch/ │ │ serialization.py:795 in load │ │ │ │ 792 │ │ │ │ return _legacy_load(opened_file, map_location, _weigh │ │ 793 │ │ │ except RuntimeError as e: │ │ 794 │ │ │ │ raise pickle.UnpicklingError(UNSAFE_MESSAGE + str(e)) │ │ ❱ 795 │ │ return _legacy_load(opened_file, map_location, pickle_module, │ │ 796 │ │ 797 │ │ 798 # Register pickling support for layout instances such as │ │ │ │ /home/ec2-user/anaconda3/envs/pytorch_p39/lib/python3.9/site-packages/torch/ │ │ serialization.py:1002 in _legacy_load │ │ │ │ 999 │ │ │ f"Received object of type \"{type(f)}\". Please update to │ │ 1000 │ │ │ "functionality.") │ │ 1001 │ │ │ ❱ 1002 │ magic_number = pickle_module.load(f, pickle_load_args) │ │ 1003 │ if magic_number != MAGIC_NUMBER: │ │ 1004 │ │ raise RuntimeError("Invalid magic number; corrupt file?") │ │ 1005 │ protocol_version = pickle_module.load(f, pickle_load_args) │ ╰──────────────────────────────────────────────────────────────────────────────╯ UnpicklingError: invalid load key, '<'.

[Amnah1100]

I experienced the same issue! Were you able to fix it?

[dimwael]

I fixed it by deleting the pretrained_minigpt4.pth and download it again directly from the drive. For that you may use this approach : pip install gdown Finally, download the file using gdown and the intended ID:

gdown --id 1a4zLvaiDBr-36pasffmgpvH5P7CKmpze I hope that helps !

[Amnah1100]

It worked! Thank you very much ....