[BUG] wandb logging syncs project name as run name

[laelhalawani]

Prerequisites

• I have read the documentation.
• I have checked other issues for similar problems.

Backend

Local

Interface Used

CLI

CLI Command

model = "meta-llama/Llama-3.2-1B-Instruct"


llm_params = LLMTrainingParams(
    model="meta-llama/Llama-3.2-1B-Instruct",
    data_path=hf_dataset,
    chat_template=None,
    prompt_text_column="input_text",
    text_column="output_text",
    train_split="train",
    trainer="sft", # "sft" or "default"
    epochs=24,
    batch_size=1,
    lr=5e-5,
    peft=True,
    quantization=None, # "int8", "int4" or None
    target_modules="all-linear",
    padding="right",
    optimizer="paged_adamw_8bit",
    scheduler="cosine",
    gradient_accumulation=8,
    mixed_precision="fp16", # "bf16", "fp16" or None
    merge_adapter=True,
    project_name="autotrain-llama32-1b-finetune",
    log="wandb",
    push_to_hub=False,
    username=os.environ.get("HF_USERNAME"),
    token=os.environ.get("HF_TOKEN"),
    eval_steps=100,
    warmup_ratio=0.01,
)



backend = "local"
project = AutoTrainProject(params=llm_params, backend=backend, process=True)
project.create()

UI Screenshots & Parameters

No response

Error Logs

There are no error logs for this, just a screenshot confirming it's setting project name as generic "huggingface" and run name as whatever was set as project name.

Additional Information

When using "wandb" as parameter for log in LLMTrainingParams, the project name on wandb.ai doesn't reflect the project name set in auto trainer. Instead the project name is synced as run name.

How it should work:

1. Sync project name as project name
2. Allow specifying run name in a separate parameter.

Thank you!

[github-actions]

This issue is stale because it has been open for 30 days with no activity.

[laelhalawani]

bump

[abhishekkrthakur]

please create a PR if possible :)