google

/

gemma-2b

like 904

Follow

Google 3,935

Running into the same issue, bypass was to get through Kaggle.

Running into the same issue, the bypass was to get through Kaggle.

Were you able to deploy it here? I could run it on Kaggle and also on my local PC, but not here on HuggingFace.

Hi @asif00 , sorry you're facing this issue. Could you create a token in your user settings (https://huggingface.co/settings/tokens) and set it as HF_TOKEN secret in your Space. No need to use login(...) anymore when doing so. If you have yourself access to https://huggingface.co/google/gemma-2b/tree/main from the browser (which seems to be the case), it should work in the Space as well. Please let me know if this doesn't solve your issue.

Also failed with Cannot access gated repo for url https://huggingface.co/google/gemma-2b/resolve/main/config.json. Repo model google/gemma-2b is gated. You must be authenticated to access it.

running local and my code as follow:

from transformers import AutoTokenizer, AutoModelForCausalLM

access_token='hf_******'

tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b")
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", token=access_token)

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer(input_text, return_tensors="pt")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

@chuanfengx64 Could you check this Space I've just created: https://huggingface.co/spaces/Wauplin/tmp_for_gemma_discussion_11/blob/main/app.py.

Instead of hardcoding my user token, I set it as HF_TOKEN in the Space secrets (e.g. in https://huggingface.co/spaces/Wauplin/tmp_for_gemma_discussion_11/settings) which a valid token that I generated for it. And then everything works fine and the files are downloaded correctly.

Hi @Wauplin , thanks for your reply, I can't access your Space with 404.

Oops sorry @chuanfengx64 , it was private while I was debugging it. Should be good now!

Also failed with Cannot access gated repo for url https://huggingface.co/google/gemma-2b/resolve/main/config.json. Repo model google/gemma-2b is gated. You must be authenticated to access it.

running local and my code as follow:

from transformers import AutoTokenizer, AutoModelForCausalLM

access_token='hf_******'

tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b")
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", token=access_token)

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer(input_text, return_tensors="pt")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Oh! tokenizer should be also with token!!

Hi @Wauplin and thanks but I can't access your Space with 403.

@Yhnf How do you try to access it? If I go to https://huggingface.co/spaces/Wauplin/tmp_for_gemma_discussion_11 with incognito mode (e.g. not logged in), it works for me. This Space is not really interesting though, it was just to demonstrate how to load gemma weights in transformers in a Space by handling authentication with a HF_TOKEN secret.

Oh! tokenizer should be also with token!!

Yes @garyox64 !But note that if the token is set from an environment variable (or a Space secret), you won't have to think about it anymore -e.g. won't have to know which methods require authentication-.

Oh ok thanks for the reply @Wauplin , actually the url without /settings works fine, my bad for not have checked that

No worries, glad it's figured out :)

Hi @asif00 , sorry you're facing this issue. Could you create a token in your user settings (https://huggingface.co/settings/tokens) and set it as HF_TOKEN secret in your Space. No need to use login(...) anymore when doing so. If you have yourself access to https://huggingface.co/google/gemma-2b/tree/main from the browser (which seems to be the case), it should work in the Space as well. Please let me know if this doesn't solve your issue.

@Wauplin Thank you so much for your kind response. It worked like a charm!