Emerging trends (smaller models, edge AI) in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Emerging trends (smaller models, edge AI)

This pipeline shows how smaller AI models are trained and deployed on edge devices. It highlights data preparation, model training with fewer parameters, and running predictions locally on devices like phones or sensors.

Data Flow - 6 Stages

1Raw Data Collection

10000 rows x 20 columns→Collect sensor readings and user inputs→10000 rows x 20 columns

Temperature, humidity, motion, and user activity logs

↓

2Data Preprocessing

10000 rows x 20 columns→Clean missing values, normalize features→10000 rows x 20 columns

Normalized temperature values between 0 and 1

↓

3Feature Engineering

10000 rows x 20 columns→Select important features, reduce dimensions→10000 rows x 10 columns

Selected key features like motion intensity and temperature trend

↓

4Model Training

8000 rows x 10 columns→Train small neural network with 2 layers and 5000 parameters→Model with 5000 parameters

Model learns to classify user activity as walking, sitting, or running

↓

5Model Deployment to Edge

Model with 5000 parameters→Compress and deploy model to mobile device→Model running on device with limited memory

Model size reduced to 2MB for phone deployment

↓

6Local Prediction

1 sample x 10 features→Run prediction on device without internet→1 prediction output

Predicted activity: walking with 85% confidence

Training Trace - Epoch by Epoch

Loss
0.8 |****
0.6 |*** 
0.4 |**  
0.2 |*   
0.0 +----
     1 5 10 15 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.75	0.55	Model starts learning basic patterns
5	0.45	0.75	Accuracy improves as model learns features
10	0.30	0.85	Model converges with good accuracy
15	0.25	0.88	Slight improvement, training stabilizes

Prediction Trace - 4 Layers

Layer 1: Input Layer

Layer 2: Hidden Layer 1 (ReLU)

Layer 3: Output Layer (Softmax)

Layer 4: Prediction

Model Quiz - 3 Questions

Why do smaller models work well on edge devices?

A They require more internet bandwidth

B They have more layers than big models

C They use less memory and compute power

D They always have higher accuracy

Key Insight

Smaller AI models trained with careful feature selection can run efficiently on edge devices, enabling fast local predictions without internet. This trend helps bring AI closer to users with limited resources.