Hugging Face fine-tuning in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Hugging Face fine-tuning

This pipeline shows how a pre-trained language model from Hugging Face is fine-tuned on a new text dataset to improve its performance on a specific task, like sentiment analysis or question answering.

Data Flow - 5 Stages

1Data Loading

1000 rows x 2 columns→Load raw text data and labels from CSV→1000 rows x 2 columns

Row: {text: 'I love this movie!', label: 'positive'}

↓

2Preprocessing

1000 rows x 2 columns→Tokenize text into input IDs and attention masks→1000 rows x 128 tokens

Input IDs: [101, 1045, 2293, 2023, 3185, 999, 102]

↓

3Train/Test Split

1000 rows x 128 tokens→Split data into 800 training and 200 testing samples→Train: 800 rows x 128 tokens, Test: 200 rows x 128 tokens

Train sample input IDs: [101, 1045, 2293, 2023, 3185, 999, 102]

↓

4Model Fine-tuning

800 rows x 128 tokens→Train pre-trained model on training data for 3 epochs→Fine-tuned model weights

Model updates weights to better predict sentiment

↓

5Evaluation

200 rows x 128 tokens→Predict labels on test data and compute accuracy→Accuracy score (e.g., 0.88)

Predicted labels vs true labels compared

Training Trace - Epoch by Epoch

Loss
0.7 |*       
0.6 |**      
0.5 |***     
0.4 |****    
0.3 |*****   
0.2 |        
    +--------
     1 2 3 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.65	0.70	Model starts learning, loss decreases from initial high value
2	0.42	0.82	Loss decreases further, accuracy improves significantly
3	0.30	0.88	Model converges with low loss and high accuracy

Prediction Trace - 5 Layers

Layer 1: Tokenization

Layer 2: Embedding Layer

Layer 3: Transformer Layers

Layer 4: Classification Head

Layer 5: Prediction

Model Quiz - 3 Questions

What happens to the loss value during fine-tuning?

A It stays the same

B It increases steadily over epochs

C It decreases steadily over epochs

D It randomly jumps up and down

Key Insight

Fine-tuning a pre-trained Hugging Face model adapts it to a new task by updating weights with new data, improving accuracy while starting from a strong base. The loss decreases and accuracy increases as the model learns.