Unbabel

/

wmt22-comet-da

like 24

Follow

Unbabel 85

It usually uses HF cache to cache those models. In any case, you can always clone this repo and then use the "load_from_checkpoint" method to load the checkpoint path

See my point is I am using comet model path....but it is downloading models--xlm-roberta-large but using huggingface url....but in my production environment that url is blocked now how to use local path of roberta large

ahh I see I misunderstood what you were saying. But that should not happen because for models that are trained the encoder models are not downloaded. We prevent that here and here.

I would have to investigate that further... But should not happen

The only thing that might be happening is that you still need to download some things like the tokenizer and the configs of the model... maybe then the best thing for you is to modify the functions I presented above to load directly from a local xlm-roberta file

its probably that... its attempting to download the configs and tokenizer

Can't load tokenizer for 'xlm-roberta-large'. If you were trying to load it from '
https://huggingface.co/models
', make sure you don't have a local directory with the same name. Otherwise, make sure 'xlm-roberta-large' is the correct path to a directory containing all relevant files for a XLMRobertaTokenizerFast tokenizer

This is the error I am facing in production environment because https://huggingface.co/models is blocked....if roberta large model is present in cache it works perfectly if not it dowloads

Hey, facing the same issue as Atulad, while what i did was load the tokenizer of 'xlm-roberta-large' seperately but still faced the same error. Right now i have all the relevant files for the tokenizer but it seems the directory is incorrect. May I ask the syntax or the preset local directory for the tokenizer so it will not trigger this error:

Can't load tokenizer for 'xlm-roberta-large'. If you were trying to load it from '
https://huggingface.co/models
', make sure you don't have a local directory with the same name. Otherwise, make sure 'xlm-roberta-large' is the correct path to a directory containing all relevant files for a XLMRobertaTokenizerFast tokenizer

Hi I solved this issue, while pushing the code in gitlab/github do push the directory of xlm-roberta-large size(15mb) and create a docker file and copy the directory in your root/home directory.
RUN mkdir -p $WORKING_DIR/.cache/huggingface/hub
ADD models--xlm-roberta-large $WORKING_DIR/.cache/huggingface/hub/models--xlm-roberta-large

Hi I solved this issue, while pushing the code in gitlab/github do push the directory of xlm-roberta-large size(15mb) and create a docker file and copy the directory in your root/home directory.
RUN mkdir -p $WORKING_DIR/.cache/huggingface/hub
ADD models--xlm-roberta-large $WORKING_DIR/.cache/huggingface/hub/models--xlm-roberta-large

Hey Atulad, I fixed the issue by just moving my directory up a level. Easy fix!

Hi I solved this issue, while pushing the code in gitlab/github do push the directory of xlm-roberta-large size(15mb) and create a docker file and copy the directory in your root/home directory.
RUN mkdir -p $WORKING_DIR/.cache/huggingface/hub
ADD models--xlm-roberta-large $WORKING_DIR/.cache/huggingface/hub/models--xlm-roberta-large

Hey Atulad, I fixed the issue by just moving my directory up a level. Easy fix!

Hey esesjay, I am facing the same problem by using model = load_from_checkpoint(checkpoint_path="/model/wmt20-comet-da/checkpoints/model.ckpt"). I wonder how you fix the issue by moving your directory up a level. Many thanks!