Finetuning LLaMA 3.2 model (3B weights)

Finetuning LLaMA 3.2 model using QLoRA to predict product price based on description. Data preparation and model evaluation against business objectives.

Key components

• Finetuning datasets generation
• Base model evaluation for baseline (without finetuning)
• Quantization parameters (quantization is only for the base model)
• Lora Config with key param such as Alpha, Dropout, r and target modules
• SFT Trainer using key hyperparameters and the Lora config
• Setting up Weight & Biases for live metrics collection and visualization
• Selecting the most performing model from the different checkpoints
• Run Inference on the QLoRA finetuned model

Training Performance: