Environmental impact of AI in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Environmental impact of AI

This pipeline shows how training a large AI model affects the environment through energy use and carbon emissions. It tracks data from raw input to model training and measures energy consumption and emissions at each step.

Data Flow - 6 Stages

1Raw Data Collection

1000000 rows x 10 columns→Gathering text and image data from the internet→1000000 rows x 10 columns

Text samples and images collected for training

↓

2Data Preprocessing

1000000 rows x 10 columns→Cleaning, filtering, and formatting data→950000 rows x 10 columns

Removed duplicates and irrelevant data

↓

3Feature Engineering

950000 rows x 10 columns→Converting raw data into numerical features→950000 rows x 512 features

Text converted to 512-dimensional vectors

↓

4Model Training

950000 rows x 512 features→Training a deep neural network with 12 layers→Model weights updated after 10 epochs

Model learns patterns from features

↓

5Energy Consumption Measurement

Model training process→Measuring electricity used by GPUs→Total energy used: 1500 kWh

Energy meter records power during training

↓

6Carbon Emission Estimation

Energy consumption data→Calculating CO2 emissions based on energy source→Estimated emissions: 750 kg CO2

Using local electricity carbon intensity

Training Trace - Epoch by Epoch

Loss
1.2 |*        
0.9 | *       
0.7 |  *      
0.55|   *     
0.45|    *    
0.38|     *   
0.33|      *  
0.29|       * 
0.26|        *
0.24|         *
     ----------------
      Epochs 1 to 10

Epoch	Loss ↓	Accuracy ↑	Observation
1	1.2	0.45	High loss and low accuracy at start
2	0.9	0.60	Loss decreases, accuracy improves
3	0.7	0.72	Model learns important features
4	0.55	0.80	Training progressing well
5	0.45	0.85	Good balance of loss and accuracy
6	0.38	0.88	Model converging
7	0.33	0.90	Stable improvement
8	0.29	0.92	Approaching best performance
9	0.26	0.93	Small improvements
10	0.24	0.94	Training complete with good accuracy

Prediction Trace - 3 Layers

Layer 1: Input Layer

Layer 2: Hidden Layers (12 layers)

Layer 3: Output Layer with Softmax

Model Quiz - 3 Questions

What happens to the data size after preprocessing?

A It decreases because irrelevant data is removed

B It increases because new features are added

C It stays the same size

D It becomes empty

Key Insight

Training large AI models uses significant energy, which leads to carbon emissions. Monitoring energy and emissions during data processing and training helps understand and reduce AI's environmental impact.