Tool usage (function calling) in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Tool usage (function calling)

This pipeline shows how a machine learning model uses a tool (function) during prediction to improve its output. The model calls a function to get extra information, then combines it with its own prediction.

Data Flow - 5 Stages

1Input data

1 sample x 3 features→Receive raw input features→1 sample x 3 features

[5.1, 3.5, 1.4]

↓

2Preprocessing

1 sample x 3 features→Normalize features to range 0-1→1 sample x 3 features

[0.51, 0.7, 0.14]

↓

3Model prediction

1 sample x 3 features→Feed features into neural network→1 sample x 2 outputs

[0.6, 0.4]

↓

4Function call (tool usage)

1 sample x 3 features→Call external function to get extra info→1 sample x 1 feature

[0.8]

↓

5Combine outputs

1 sample x 2 outputs + 1 feature→Combine model output and function output→1 sample x 2 outputs

[0.68, 0.32]

Training Trace - Epoch by Epoch

Loss
0.7 |****
0.6 |*** 
0.5 |**  
0.4 |*   
0.3 |    
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.65	0.60	Model starts learning, loss high, accuracy low
2	0.50	0.72	Loss decreases, accuracy improves
3	0.40	0.80	Model learns important patterns
4	0.32	0.85	Loss continues to drop, accuracy rises
5	0.28	0.88	Model converges with good accuracy

Prediction Trace - 5 Layers

Layer 1: Input features

Layer 2: Normalization

Layer 3: Neural network prediction

Layer 4: Function call (tool usage)

Layer 5: Combine outputs

Model Quiz - 3 Questions

What does the function call (tool usage) provide in the pipeline?

A Raw input data

B Extra information to improve prediction

C Final prediction output

D Training loss value

Key Insight

Using a tool (function call) during prediction can provide extra useful information that improves the model's final output beyond what the model alone predicts.