System Info

• transformers version: 4.47.0.dev0
• Platform: Windows-11-10.0.26100-SP0
• Python version: 3.12.5
• Huggingface_hub version: 0.26.1
• Safetensors version: 0.4.5
• Accelerate version: 1.0.1
• Accelerate config: not found
• PyTorch version (GPU?): 2.5.0+cu124 (True)
• 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?:
• Using GPU in script?:
• GPU type: NVIDIA GeForce GTX 1660 Ti

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

1. Install wandb package
2. Run the below script with any model (model initialization is not added here for brevity)

default_train_args = TrainingArguments(output_dir="tmp")

default_train_args.report_to = "none"

trainer = Trainer(model=model, args=default_train_args)

Expected behavior

The script will run without any integration when report_to value is assigned with none value.

Problem

The check for the report_to variable is done in the post_init function of the TrainingArguments. If you don't specify it in the constructor then the code set to all and fetch all the integrations. If you assing a value to report_to variable after initialization constructor, the below issue is happening.

In the Trainer constructor, with get_reporting_integration_callbacks function the class is fetching the integration callbacks but the function is not checking if string or array is the type of the report_to value. Because of that the functions raises ValueError if you assign string to the value because it is trying to check with char rather than string. This behaviour is same for the none and all string values.

• The responsible code parts for the issue can be found below:
  • [code] TrainingArguments
  • [code] Trainer
  • [code] Integration Utils

Solution

Adding below code to the beginning of the get_reporting_integration_callbacks function will fix the issue:

if report_to is None:
     return []

if isinstance(report_to, str):
    if "none" == report_to:
        return []
    elif "all" == report_to:
        report_to = get_available_reporting_integrations()
    else:
        report_to = [report_to]

• If the issue is approved, I will open PR.