Retrieval Agumented Generation for Privacy and Efficiency

Vector Database:

• FAISS (Facebook AI Similarity Search)

$$ Euclidean\ Distance(q, x1) = \sqrt{(q_1 - x_i1)^2+(q_2 - x_i2)^2} $$

Embeddings Models:

Model Name: all-MiniLM-L6-v2
Embedding dimension: 384
Parameter: 22.7Million
Model Size: 0.008GB
Max Tokens: 512

Model Name: stella-base-en-v2
Embedding dimension: 768
Parameter: 55Million
Model Size: 0.2GB
Max Tokens: 512

Generator Models:

Name: Deepseek-r1
Parameter: 1.5billion
Size: 1.1GB
Year: 2024

Name: Gemma
Parameter: 2billion
Size: 1.7GB
Year: 2024

Name: Llama3.2
Parameter: 1billion
Size: 1.3GB
Year: 2024