Skip to content

Quick Start Guide

This guide will walk you through running your first fine-tuning job with the Fine-Tune Pipeline in just a few minutes.

Prerequisites

Before you begin, make sure you have:

Step 1: Verify Your Setup

First, let's make sure everything is working:

# Navigate to your project directory
cd Fine-Tune-Pipeline

# Sync dependencies
uv sync

# Test the installation
uv run python -c "from app.finetuner import FineTune; print('✅ Setup verified!')"

Step 2: Understanding the Default Configuration

The pipeline comes with a pre-configured setup in config.toml. Let's look at the key settings:

[fine_tuner]
# Base model - A small, efficient model for quick testing
base_model_id = "unsloth/Qwen2.5-0.5B-Instruct-bnb-4bit"

# Training data - Default dataset for question-answering
training_data_id = "rtweera/simple_implicit_n_qa_results_v2"

# Training settings - Optimized for quick runs
epochs = 3
device_train_batch_size = 4
learning_rate = 0.0002

First Run Recommendation

The default configuration is designed for a quick first run. It uses a small model and dataset that should complete training in 10-15 minutes on a modern GPU.

Step 3: Run Your First Fine-Tuning Job

Now let's run the fine-tuner with your API keys:

uv run app/finetuner.py --hf-key "your_hf_token" --wandb-key "your_wandb_key"

What Happens During Training

  1. Model Loading: Downloads and loads the base model (Qwen2.5-0.5B)
  2. Data Processing: Downloads and processes the training dataset
  3. LoRA Setup: Configures Low-Rank Adaptation for efficient fine-tuning
  4. Training: Runs 3 epochs of training with progress tracking
  5. Saving: Saves the model locally and pushes to Hugging Face Hub

Expected Output

You should see output similar to this:

--- ✅ Login to Hugging Face Hub successful. ---
--- ✅ Training dataset loaded: rtweera/simple_implicit_n_qa_results_v2 ---
--- ✅ No validation dataset provided. Skipping validation. ---
--- ✅ Model and tokenizer loaded successfully. ---
--- ✅ Data preprocessing completed. ---
Run name set to: fine-tuned-model-20250629-143022
--- ✅ Weights & Biases setup completed. ---
--- ✅ Trainer initialized successfully. ---
--- ✅ Starting training... ---

Training Progress:
  0%|          | 0/150 [00:00<?, ?it/s]
 10%|█         | 15/150 [00:30<04:30,  2.0s/it]
 20%|██        | 30/150 [01:00<04:00,  2.0s/it]
...

--- ✅ Training completed with stats: {...} ---
--- ✅ Model and tokenizer saved to ./models/fine_tuned locally and to Hugging Face Hub ---
--- ✅ Fine-tuning completed successfully. ---

Step 4: Test Your Fine-Tuned Model

After training, let's test the model with inference:

4.1 Update Configuration for Inference

First, update your config.toml to use your newly trained model:

[inferencer]
# Use your Hugging Face username and the generated model name
hf_user_id = "your-hf-username"
# The run_name from training (or leave as "null" to use the latest)
run_name = "null"

# Test dataset
testing_data_id = "rtweera/user_centric_results_v2"

4.2 Run Inference

uv run app/inferencer.py --hf-key "your_hf_token"

This will generate predictions and save them to inferencer_output.jsonl.

Step 5: Evaluate Your Model

Finally, let's evaluate how well your model performed:

uv run app/evaluator.py --openai-key "your_openai_key"

This will generate:

  • evaluator_output_summary.json - Overall performance metrics
  • evaluator_output_detailed.xlsx - Detailed evaluation results

Step 6: Review Results

Weights & Biases Dashboard

  1. Go to wandb.ai
  2. Navigate to your project: fine-tuning-project-ci-cd
  3. View training metrics, loss curves, and system metrics

Local Results

Check the generated files:

# View inference results
head -5 inferencer_output.jsonl

# View evaluation summary
cat evaluator_output_summary.json

# Open detailed results in Excel
start evaluator_output_detailed.xlsx  # Windows
# or
open evaluator_output_detailed.xlsx   # macOS

Next Steps

Congratulations! 🎉 You've successfully run your first fine-tuning pipeline. Here's what you can do next:

Customize Your Training

  1. Use Your Own Data: Replace training_data_id with your dataset
  2. Try Different Models: Experiment with larger models like Llama or Mistral
  3. Adjust Hyperparameters: Modify learning rate, batch size, epochs

Advanced Features

  1. Advanced Configuration - Explore all configuration options
  2. CI/CD Integration - Set up automated training pipelines
  3. API Reference - Deep dive into the codebase

Troubleshooting

If you encounter issues:

  1. Check the Troubleshooting Guide
  2. Verify your API keys are correct
  3. Ensure you have sufficient GPU memory
  4. Check the console output for specific error messages

Common First-Run Issues

Out of Memory

If you get CUDA out of memory errors, reduce the batch size: 

device_train_batch_size = 2  # Reduce from 4
grad_accumulation = 8        # Increase to maintain effective batch size

Dataset Not Found

If the dataset fails to load, check: - Your internet connection - The dataset ID is correct - You have access to the dataset (some require authentication)

Training Too Slow

For faster training on CPU: 

device_train_batch_size = 1
epochs = 1
max_sequence_length = 1024  # Reduce sequence length

Happy fine-tuning! 🚀