pytorch_model.bin corrupted (truncated ZIP) for CALM-M on HuggingFace

[YEOMJINSEOP]

Thanks for sharing wonderful work CALM!

Problem

The pytorch_model.bin file for CALM-M on HuggingFace is corrupted. CALM-L and CALM-XL seems work fine.

  import torch
  from huggingface_hub import hf_hub_download

  path = hf_hub_download('cccczshao/CALM-M', 'pytorch_model.bin')
  torch.load(path, map_location='cpu')
  # RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

Cause (Maybe)

The file size matches HuggingFace metadata (1,748,107,520 bytes), so it's not a download issue. The ZIP End of Central Directory (EOCD) signature is missing — the file might be truncated during upload.

Coud you please re-upload pytorch_model.bin for CALM-M ?

Best regards.

[shaochenze]

Thank you for letting us know! We have re-uploaded the model, and it should load properly now.

[goodluckcwl]

Thanks for sharing wonderful work CALM!

Problem

The pytorch_model.bin file for CALM-M on HuggingFace is corrupted. CALM-L and CALM-XL seems work fine.

import torch
from huggingface_hub import hf_hub_download

path = hf_hub_download('cccczshao/CALM-M', 'pytorch_model.bin')
torch.load(path, map_location='cpu')

RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

Cause (Maybe)

The file size matches HuggingFace metadata (1,748,107,520 bytes), so it's not a download issue. The ZIP End of Central Directory (EOCD) signature is missing — the file might be truncated during upload.

Coud you please re-upload pytorch_model.bin for CALM-M ?

Best regards.

Could you provide some example generated by CALM-XL or CALM-L?

[shaochenze]

This is a sample of CALM-XL on the test set. The earlier generation is fluent, but some errors occurred in the middle, which led to a quality drop of the later output. I think this is related to the current autoencoder. A context-aware or autoregressive autoencoder should help improve the generation quality.