How to load a pre-trained model from local file? thanks

[lukc12138]

When I load the lightning checkpoint from my local file, I have a problem like this and I don't know how to fix it.

hydra.errors.InstantiationException: Error in call to target 'lightning.pytorch.core.module.LightningModule.load_from_checkpoint':
UnpicklingError("invalid load key, '`'.")
full_key: model

Can you give me some advice? Thanks a lot.

[gorold]

Hi @lukc12138, I'm unable to provide any further assistance without substantially more details. Alternatively, this seems to be an error due to the hydra config files, which you can find more details about here: https://hydra.cc/docs/intro/

[lukc12138]

Thanks for your reply. The problem occurred when I tried to load .safetensors from local file and it reported 'UnpicklingError("invalid load key, '`'.")'.

[lukc12138]

I changed my code is like this:

model = MoiraiForecast.load_from_checkpoint(
    module_kwargs={
        'distr_output': {
            '_target_': 'uni2ts.distribution.MixtureOutput',
            'components':{
                '_target_': 'uni2ts.distribution.StudentTOutput',
                '_target_': 'uni2ts.distribution.NormalFixedScaleOutput',
                '_target_': 'uni2ts.distribution.NegativeBinomialOutput',
                '_target_': 'uni2ts.distribution.LogNormalOutput'}
        },
        'd_model': 384,
        'num_layers': 6,
        'patch_sizes': 32,
        'max_seq_len': 512,
        'attn_dropout_p': 0.0,
        'dropout_p': 0.0,
        'scaling': True
    },
    prediction_length=PDT,
    context_length=CTX,
    patch_size=PSZ,
    num_samples=100,
    target_dim=1,
    feat_dynamic_real_dim=ds.num_feat_dynamic_real,
    past_feat_dynamic_real_dim=ds.num_past_feat_dynamic_real,
    checkpoint_path="/moirai/uni2ts-main/pretrained_model/moirai_small.safetensors",
)

The error report is like this:

Traceback (most recent call last):
  File "/moirai/uni2ts-main/start.py", line 38, in <module>
    model = MoiraiForecast.load_from_checkpoint(
  File "/miniconda3/envs/moirai/lib/python3.10/site-packages/lightning/pytorch/core/module.py", line 1532, in load_from_checkpoint
    loaded = _load_from_checkpoint(
  File "/miniconda3/envs/moirai/lib/python3.10/site-packages/lightning/pytorch/core/saving.py", line 62, in _load_from_checkpoint
    checkpoint = pl_load(checkpoint_path, map_location=map_location)
  File "/miniconda3/envs/moirai/lib/python3.10/site-packages/lightning/fabric/utilities/cloud_io.py", line 51, in _load
    return torch.load(f, map_location=map_location)  # type: ignore[arg-type]
  File "/miniconda3/envs/moirai/lib/python3.10/site-packages/torch/serialization.py", line 1028, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/miniconda3/envs/moirai/lib/python3.10/site-packages/torch/serialization.py", line 1246, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, '`'.

[lukc12138]

I solved the problem by the following steps:

1. check the version of lightning version.
2. check the folder that contains the .safetensors, a config.json should be put there.

[HALF111]

Hi, I encountered the same problem as you, so what version should be lightning? My torch==2.2.2, lightning==2.2.2, but it does not work :(

[lukc12138]

Hi, I encountered the same problem as you, so what version should be lightning? My torch==2.2.2, lightning==2.2.2, but it does not work :(

l use the lightning==2.1.3

[xcf11111]

Hi, I encountered the same problem as you, so what version should be lightning? My torch==2.2.2, lightning==2.2.2, but it does not work :(

l use the lightning==2.1.3

What version of PyTorch are you using?