Bug NeuralModel ?

[aagostinelli86]

Discussed in https://github.com/MAIF/melusine/discussions/147

^{Originally posted by **aagostinelli86** January 3, 2023} Hello @TFA-MAIF I have an issue with the tutorial notebook n 13, cell 33 (Camembert model): "OSError: Model name 'jplu/tf-camembert-base' was not found in tokenizers model name list (camembert-base). We assumed 'jplu/tf-camembert-base' was a path, a model identifier, or url to a directory containing vocabulary files named ['sentencepiece.bpe.model'] but couldn't find such vocabulary files at this path or url." I think it's due to the proxy, so I downloaded locally the pretrained model (config.json + .tf_model.h5) and I specified its path in your NeuralModel wrapper. The issue holds with this different message: "OSError: Model name 'C:\Users\vgkj536\git_projects\melusine\tutorial\jplu\tf-camembert-base' was not found in tokenizers model name list (camembert-base). We assumed 'C:\Users\vgkj536\git_projects\melusine\tutorial\jplu\tf-camembert-base' was a path, a model identifier, or url to a directory containing vocabulary files named ['sentencepiece.bpe.model'] but couldn't find such vocabulary files at this path or url." On the opposite, if I call directly the TFCamembertModel.from_pretrained() with this local path, it works well. Is this due to a bug ? 

[TFA-MAIF]

Hello,

Calling directly jplu's model (available on Huggingface) is not possible if you have proxy. Indeed if you want it to work, you need to have your bert model locally and call your local path. I will let my colleagues in charge of support (ex: @hugo-quantmetry) assists you to dive into your bug.

Best regards, Tiphaine

[HugoPerrier]

New melusine v3.0.0 is out. Closing legacy issues