Generate mistral-embed vectors and compare them in Python

What you get

1024-dimension text embeddings from mistral-embed, plus a cosine-similarity check you can use for semantic search or dedupe.

Prerequisites

A Mistral API key in MISTRAL_API_KEY
Python with the mistralai package

Steps

Install the SDK and set your key.

Shell

pip install mistralai
export MISTRAL_API_KEY="your-api-key"

Save as embed.py: embed a batch of strings with client.embeddings.create.

Python

import os
import math
from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

resp = client.embeddings.create(
    model="mistral-embed",
    inputs=["How do I reset my password?", "I forgot my login credentials."],
)

a = resp.data[0].embedding
b = resp.data[1].embedding
print(len(a))  # 1024

def cosine(x, y):
    dot = sum(i * j for i, j in zip(x, y))
    return dot / (math.sqrt(sum(i * i for i in x)) * math.sqrt(sum(j * j for j in y)))

print(round(cosine(a, b), 3))

Test it

Shell

python embed.py

It prints 1024 and a similarity score; the two related sentences should score noticeably higher than an unrelated pair.