IsADirectoryError: [Errno 21] Is a directory: 'my-company/my-llm'

gventuri commented 9 months ago

System Info

transformers version: 4.37.0.dev0
Platform: Linux-5.15.0-89-generic-x86_64-with-glibc2.31
Python version: 3.10.11
Huggingface_hub version: 0.19.4
Safetensors version: 0.4.1
Accelerate version: 0.27.0.dev0
Accelerate config: not found
PyTorch version (GPU?): 2.1.0+cu121 (True)
Tensorflow version (GPU?): not installed (NA)
Flax version (CPU?/GPU?/TPU?): not installed (NA)
Jax version: not installed
JaxLib version: not installed
Using GPU in script?: yes
Using distributed or parallel set-up in script?: no

Who can help?

@ArthurZucker @younesbelkada

Information

[ ] The official example scripts
[x] My own modified scripts

Tasks

[ ] An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
[ ] My own task or dataset (give details below)

Reproduction

Run the following code:


from transformers import AutoModelForCausalLM
import torch

m = AutoModelForCausalLM.from_pretrained( "pretrained-model", return_dict=True, torch_dtype=torch.bfloat16, device_map="auto" )

m.push_to_hub("my-company/my-llm")


- I get the following error

IsADirectoryError Traceback (most recent call last) Cell In[5], line 1 ----> 1 m.push_to_hub("my-company/my-llm")

File /opt/conda/lib/python3.10/site-packages/transformers/modeling_utils.py:2530, in PreTrainedModel.push_to_hub(self, *args, *kwargs) 2528 if tags: 2529 kwargs["tags"] = tags -> 2530 return super().push_to_hub(args, **kwargs)

File /opt/conda/lib/python3.10/site-packages/transformers/utils/hub.py:865, in PushToHubMixin.push_to_hub(self, repo_id, use_temp_dir, commit_message, private, token, max_shard_size, create_pr, safe_serialization, revision, commit_description, tags, **deprecated_kwargs) 860 repo_id = self._create_repo( 861 repo_id, private=private, token=token, repo_url=repo_url, organization=organization 862 ) 864 # Create a new empty model card and eventually tag it --> 865 model_card = create_and_tag_model_card( 866 repo_id, tags, token=token, ignore_metadata_errors=ignore_metadata_errors 867 ) 869 if use_temp_dir is None: 870 use_temp_dir = not os.path.isdir(working_dir)

File /opt/conda/lib/python3.10/site-packages/transformers/utils/hub.py:1120, in create_and_tag_model_card(repo_id, tags, token, ignore_metadata_errors) 1104 """ 1105 Creates or loads an existing model card and tags it. 1106 (...) 1116 the process. Use it at your own risk. 1117 """ 1118 try: 1119 # Check if the model card is present on the remote repo -> 1120 model_card = ModelCard.load(repo_id, token=token, ignore_metadata_errors=ignore_metadata_errors) 1121 except EntryNotFoundError: 1122 # Otherwise create a simple model card from template 1123 model_description = "This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated."

File /opt/conda/lib/python3.10/site-packages/huggingface_hub/repocard.py:185, in RepoCard.load(cls, repo_id_or_path, repo_type, token, ignore_metadata_errors) 182 raise ValueError(f"Cannot load RepoCard: path not found on disk ({repo_id_or_path}).") 184 # Preserve newlines in the existing file. --> 185 with card_path.open(mode="r", newline="", encoding="utf-8") as f: 186 return cls(f.read(), ignore_metadata_errors=ignore_metadata_errors)

File /opt/conda/lib/python3.10/pathlib.py:1119, in Path.open(self, mode, buffering, encoding, errors, newline) 1117 if "b" not in mode: 1118 encoding = io.text_encoding(encoding) -> 1119 return self._accessor.open(self, mode, buffering, encoding, errors, 1120 newline)

IsADirectoryError: [Errno 21] Is a directory: 'my-company/my-llm'



This happen every time I run `push_to_hub`, even with other configurations.

### Expected behavior

I would expect this code to push the pretrained model to hugging face.

ArthurZucker commented 9 months ago

Hey! Make sure you are logged in to the hub on my-company . For me this works:

>>> from transformers import AutoModel
>>> model = AutoModel.from_pretrained("bert-base-uncase")
>>> model.push_to_hub("ArthurZ/my-llm")
CommitInfo(commit_url='https://huggingface.co/ArthurZ/my-llm/commit/53cb603f0cdd1d6284f21c541f06a3b8a0ddd2d1', commit_message='Upload model', commit_description='', oid='53cb603f0cdd1d6284f21c541f06a3b8a0ddd2d1', pr_url=None, pr_revision=None, pr_num=None)

gventuri commented 9 months ago

@ArthurZucker I tried to login both with

!huggingface-cli login --token $hf_token

and with

huggingface_hub.login

In both cases I provided my personal token (I'm admin in the company). If I try to login again, it says that I'm already connected.

ArthurZucker commented 9 months ago

cc @Wauplin you might know better than me what this represents!

Wauplin commented 9 months ago

@gventuri Can it be that you have a local directory called "my-company/my-llm" and that it conflicts with the push_to_hub?

(to confirm that you are indeed logged in, you can do huggingface-cli whoami but it really doesn't seem to be the problem here)

gventuri commented 9 months ago

@Wauplin thanks a lot, it was conflicting with the local folder, it's now working!

Wauplin commented 9 months ago

Great to know your problem's solved! :hugs:

claytoncohn commented 4 months ago

I'm having the same error. I checked, and I also have a local directory that is the same name as my HuggingFace path (which I am assuming is causing the conflict). Do you have any guidance for remedying this?

I initially tried changing the name of the local path that it saved to, but then I got a different error. It looks like whatever path I pass in to model.push_to_hub it tries to access that same path locally, even though I defined them differently. For reference, I am using QLoRA + ORPO:

base_model = "meta-llama/Meta-Llama-3-8B"
new_model = "XXX/TEST_ORPO_LLaMA3"
local_path = "models"

...
trainer.save_model(local_path)

del trainer, model
gc.collect()
torch.cuda.empty_cache()

tokenizer = AutoTokenizer.from_pretrained(base_model)
fp16_model = AutoModelForCausalLM.from_pretrained(
    base_model,
    low_cpu_mem_usage=True,
    return_dict=True,
    torch_dtype=torch.float16,
    device_map="auto",
)
fp16_model, tokenizer = setup_chat_format(fp16_model, tokenizer)

model = PeftModel.from_pretrained(fp16_model, local_path)
model = model.merge_and_unload()

model.push_to_hub(new_model, use_temp_dir=False)
tokenizer.push_to_hub(new_model, use_temp_dir=False)

FileNotFoundError: [Errno 2] No such file or directory: 'TEST_ORPO_LLaMA3'

EDIT: It seems that even though the base model is merged with the adapters in the Colab environment, the call to model.push_to_hub attempts to retrieve the model locally (i.e., from the folder w/ the saved trained adapters) from the same path as the repo that I am trying to push to on HuggingFace Hub.

FotieMConstant commented 2 months ago

hey, just stumbled on this issue, any solution to this please?

claytoncohn commented 2 months ago

Answered here:

https://github.com/mlabonne/llm-course/issues/84

Needed temp folder parameter set.

huggingface / transformers