Skip to main content
To help you track, visualize, and optimize your model training process in detail, you can integrate Weights & Biases (W&B). During the fine-tuning process, W&B will automatically log key metrics, hyperparameters, loss curve, gradient decay etc.
  • Real-Time Monitoring: Track your training’s progress through W&B’s live dashboards.
  • Experiment Tracking: View and compare multiple fine-tuning runs to choose the best-performing model.
  • Visualizations & Alerts: Leverage W&B’s advanced tools for visualizing training performance and setting up alerts for key events.

​
Connecting to Weights & Biases

  • Bring your own W&B account and connect it during training. We’ll automatically upload all your training metrics to your account for seamless tracking.

​
Integration Steps

  1. Connect Your W&B Account: You connect your W&B Account by adding your W&B API Key in the Advanced Configuration while initializing the training job.
{
        "quantization": {
          "load_in_4bit": true,
          "load_in_8bit": false,
          "bnb_4bit_compute_dtype": "float16",
          "bnb_4bit_quant_type": "nf4",
          "bnb_4bit_use_double_quant": true,
          "bnb_4bit_quant_storage": "float16"
        },
        "trainer": {
          "type": "finetune",
          "learning_rate": 0.0001,
          "batch_size": 1,
          "epochs": 1,
          "gradient_accumulation_steps": 16,
          "logging_steps": 1,
          "freeze_vit": true,
          "freeze_llm": false,
          "logging_backend": "wandb"
        },
        "image_column": "url",
        "wandb_api_key": "<YOUR API HERE>"
}
  1. Start Fine-Tuning: Once the fine-tuning job begins, we will automatically send the training metrics to your W&B account.
  2. Monitoring: All metrics from your training job will be logged to a project named Simplismart in your W&B account.

​
W&B Metrics Overview