Vector similarity metrics in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Vector similarity metrics

This pipeline shows how vector similarity metrics compare two sets of numbers to find how alike they are. It helps machines understand if two things are close or far in meaning or features.

Data Flow - 3 Stages

1Input vectors

2 vectors x 5 dimensions→Receive two numeric vectors representing features→2 vectors x 5 dimensions

[0.1, 0.3, 0.5, 0.7, 0.9] and [0.2, 0.4, 0.6, 0.8, 1.0]

↓

2Normalize vectors (optional)

2 vectors x 5 dimensions→Scale vectors to unit length for cosine similarity→2 vectors x 5 dimensions (normalized)

[0.1, 0.3, 0.5, 0.7, 0.9] -> [0.08, 0.23, 0.39, 0.54, 0.69]

↓

3Calculate similarity

2 vectors x 5 dimensions→Compute similarity score using chosen metric (cosine, Euclidean, or dot product)→1 similarity score (scalar)

Cosine similarity = 0.998

Training Trace - Epoch by Epoch

Loss
0.5 |***************
0.4 |************
0.3 |*********
0.2 |*****
0.1 |**
0.0 +----------------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.45	0.60	Initial similarity predictions are rough with moderate error.
2	0.30	0.75	Model learns to better distinguish similar vectors.
3	0.18	0.85	Similarity scores become more accurate and consistent.
4	0.10	0.92	Model converges with low loss and high accuracy.
5	0.07	0.95	Final fine-tuning improves similarity precision.

Prediction Trace - 3 Layers

Layer 1: Input vectors

Layer 2: Normalize vectors

Layer 3: Calculate cosine similarity

Model Quiz - 3 Questions

What does a cosine similarity score close to 1 indicate?

A Vectors point in almost the same direction

B Vectors are orthogonal (at 90 degrees)

C Vectors point in opposite directions

D Vectors have zero length

Key Insight

Vector similarity metrics help machines measure how close two things are by comparing their features as numbers. Normalizing vectors and using metrics like cosine similarity make this comparison meaningful and consistent.