push_to_hub from local model

[mmg10]

System Info

• transformers version: 4.43.2
• Platform: Linux-6.10.0-1-cachyos-x86_64-with-glibc2.40
• Python version: 3.12.4
• Huggingface_hub version: 0.24.2
• Safetensors version: 0.4.3
• Accelerate version: 0.33.0
• Accelerate config: not found
• PyTorch version (GPU?): 2.3.1+cpu (False)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using distributed or parallel set-up in script?:

Who can help?

@muellerzr @SunMarc

Information

• The official example scripts
• 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

In https://github.com/huggingface/trl/blob/main/examples/scripts/sft.py, I pass the following arguments.

    --model_name_or_path /home/ubuntu/work/Meta-Llama-3.1-8B \
    --dataset_name="HuggingFaceH4/no_robots" \
    --output_dir /home/ubuntu/work/Meta-Llama-3.1-8B-SFT \
    --report_to="wandb" \
    --push_to_hub true \
    --push_to_hub_model_id "llama3.1-8b-sft" \

I understand that I am using the SFTTrainer, but it is using super().push_to_hub method, hence I created the issue in this library.

The error is:

Traceback (most recent call last):
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 3761, in create_commit
    hf_raise_for_status(response)
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/utils/_errors.py", line 358, in hf_raise_for_status
    raise BadRequestError(message, response=response) from e
huggingface_hub.utils._errors.BadRequestError:  (Request ID: Root=1-66a1dfab-4f8aaadd66a3afc164a238e4;a70493b5-d4ba-4022-978f-9947dcf083c0)

Bad request:
"base_model" with value "/home/ubuntu/work/Meta-Llama-3.1-8B" is not valid. Use a model id from https://hf.co/models.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/ubuntu/work/sft.py", line 151, in <module>
    trainer.save_model(training_args.output_dir)
  File "/opt/conda/lib/python3.11/site-packages/transformers/trainer.py", line 3458, in save_model
    self.push_to_hub(commit_message="Model save")
  File "/opt/conda/lib/python3.11/site-packages/trl/trainer/sft_trainer.py", line 475, in push_to_hub
    return super().push_to_hub(commit_message=commit_message, blocking=blocking, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/transformers/trainer.py", line 4349, in push_to_hub
    return upload_folder(
           ^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 1398, in _inner
    return fn(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 4857, in upload_folder
    commit_info = self.create_commit(
                  ^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 1398, in _inner
    return fn(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/huggingface_hub/hf_api.py", line 3765, in create_commit
    raise ValueError(f"Invalid metadata in README.md.\n{message}") from e
ValueError: Invalid metadata in README.md.
- "base_model" with value "/home/ubuntu/work/Meta-Llama-3.1-8B" is not valid. Use a model id from https://hf.co/models.
wandb: | 0.037 MB of 0.037 MB uploadedded

Expected behavior

The wandb logging and training completes sucessfully. But the push fails. This is because I am loading a model from the local directory /home/ubuntu/work/Meta-Llama-3.1-8B and not from an HF repo. However, the push should work fine!

[LysandreJik]

I don't think there is support for push_to_hub_model_id in that script in sft.py.

Moving your issue to TRL, cc @kashif maybe :)

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

[qgallouedec]

Use hub_model_id instead, see the TrainingArgumentsdoc

(max_steps 100 for demo)

python examples/scripts/sft.py \
    --model_name_or_path facebook/opt-350m \
    --dataset_name timdettmers/openassistant-guanaco \
    --output_dir opt-350m-sft \
    --dataset_text_field text \
    --push_to_hub \
    --max_steps 100 \
    --hub_model_id my_hub_model_id

https://huggingface.co/qgallouedec/my_hub_model_id

[valayDave]

I am still facing the same error EVEN when I pass the hub_model_id to the TrainingArgs. I am not using the sft.py script but I am rather directly using the SFTTrainer.

My Model is loaded directly from local machine and then I run some FT on it. After the trainer.train() is done, I call trainer.push_to_hub() and the code crashes with the exact same error. What is a workaround here.

Versions of Libraries:

        "transformers": "4.44.2",
        "peft": "0.12.0",
        "trl": "0.10.1",