No module named 'torch.tensor'

Hi, I wrote the correct code a few weeks ago and run on Colab, it works well. Now I want to reuse the code. Right now the basic code could not running, is it a PyTorch upgrade or something else

import torch print(torch.hub.list('s3prl/s3prl')) print(torch.hub.load('s3prl/s3prl', 'audio_albert'))

the error is

` [hubconf] can not import upstream.wav2vec2_hug.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.roberta.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.apc.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.spec_augment.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.hubert.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.wav2vec.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.vq_apc.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.mockingjay.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.example.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.tera.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.vq_wav2vec.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.npc.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.audio_albert.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.baseline.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.pase.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.wav2vec2.hubconf: No module named ‘torch.tensor’… Pass. [hubconf] can not import upstream.cpc.hubconf: No module named ‘torch.tensor’… Pass. Using cache found in /root/.cache/torch/hub/s3prl_s3prl_master

RuntimeError Traceback (most recent call last) <ipython-input-6-ba07a714d0f8> in <module>() ----> 1 print(torch.hub.load(‘s3prl/s3prl’, ‘audio_albert’))

2 frames /usr/local/lib/python3.7/dist-packages/torch/hub.py in load(repo_or_dir, model, *args, **kwargs) 362 repo_or_dir = _get_cache_or_reload(repo_or_dir, force_reload, verbose) 363 –> 364 model = _load_local(repo_or_dir, model, *args, **kwargs) 365 return model 366

/usr/local/lib/python3.7/dist-packages/torch/hub.py in _load_local(hubconf_dir, model, *args, **kwargs) 390 hub_module = import_module(MODULE_HUBCONF, hubconf_path) 391 –> 392 entry = _load_entry_from_hubconf(hub_module, model) 393 model = entry(*args, **kwargs) 394

/usr/local/lib/python3.7/dist-packages/torch/hub.py in _load_entry_from_hubconf(m, model) 210 211 if func is None or not callable(func): –> 212 raise RuntimeError(‘Cannot find callable {} in hubconf’.format(model)) 213 214 return func

RuntimeError: Cannot find callable audio_albert in hubconf `

thanks！

[myhrbeu]

Hi, When I saw that the middle three layers had the same value and the dictionary key was named “hidden_state_XX”, I thought this was the weight for the middle layer.

I recommend that writing a document or a demo tells us what the values mean of each key is and how to calculate it, and so far I’ve only found a brief introduction in the introduction. It would be better if you could also let me know how the list features of the previous version were calculated by this version. I lost 2 points of accuracy for the features of the two versions, which was about 4 points higher than that of SOTA before, but now it is 2% higher.

The idea of the features of middle layers contains in a dict that can be used by the users is great！ Thank you and the team for the work！

Thanks！

[leo19941227]

Hey!

This issue is resolved in the latest master. Thanks for your patient!