tsellm 0.1.0a11

Use LLMs in SQLite and DuckDB

tsellm: Use LLMs in SQLite

tsellm is the easiest way to access LLMs through your SQLite database.

pip install tsellm

usage: tsellm [-h] [--sqlite | --duckdb] [-v] [filename] [sql]

Use LLMs in SQLite and DuckDB

positional arguments:
  filename       SQLite/DuckDB database to open (defaults to SQLite ':memory:').
                 A new database is created if the file does not previously exist.
  sql            An SQL query to execute. Any returned rows are printed to
                 stdout.

options:
  -h, --help     show this help message and exit
  --sqlite       SQLite mode
  --duckdb       DuckDB mode
  -v, --version  Print tsellm version

Behind the scenes, tsellm is based on the beautiful llm library, so you can use any of its plugins:

Generative

For example, to access gpt4all models

llm install llm-gpt4all
# Then pick any gpt4all (it will be downloaded automatically the first time you use any model
tsellm :memory: "select prompt('What is the capital of Greece?', 'orca-mini-3b-gguf2-q4_0')"
tsellm :memory: "select prompt('What is the capital of Greece?', 'orca-2-7b')"

Embeddings

llm install llm-sentence-transformers
llm sentence-transformers register all-MiniLM-L12-v2
tsellm :memory: "select embed('Hello', 'sentence-transformers/all-MiniLM-L12-v2')"

Embeddings for binary (BLOB) columns

wget https://tselai.com/img/flo.jpg
sqlite3 images.db <<EOF
CREATE TABLE images(name TEXT, type TEXT, img BLOB);
INSERT INTO images(name,type,img) VALUES('flo','jpg',readfile('flo.jpg'));
EOF

llm install llm-clip
tsellm images.db "select embed(img, 'clip') from images"

Examples

Things get more interesting if you combine models in your standard queries.

First, create a db with some data. You can easily toggle between SQLite and DuckDB, and tsellm will pick this up automatically.

SQLite

sqlite3 prompts.db <<EOF
CREATE TABLE prompts (
   p TEXT
);
INSERT INTO prompts VALUES('hello world!');
INSERT INTO prompts VALUES('how are you?');
INSERT INTO prompts VALUES('is this real life?');
INSERT INTO prompts VALUES('1+1=?');
EOF

With a single query you can access get prompt responses from different LLMs:

tsellm prompts.db "
        select p,
        prompt(p, 'orca-2-7b'),
        prompt(p, 'orca-mini-3b-gguf2-q4_0'),
        embed(p, 'sentence-transformers/all-MiniLM-L12-v2') 
        from prompts"

DuckDB

duckdb prompts.duckdb <<EOF
CREATE TABLE prompts (
   p TEXT
);
INSERT INTO prompts VALUES('hello world!');
INSERT INTO prompts VALUES('how are you?');
INSERT INTO prompts VALUES('is this real life?');
INSERT INTO prompts VALUES('1+1=?');
EOF

With a single query you can access get prompt responses from different LLMs:

tsellm prompts.duckdb "
        select p,
        prompt(p, 'orca-2-7b'),
        prompt(p, 'orca-mini-3b-gguf2-q4_0'),
        embed(p, 'sentence-transformers/all-MiniLM-L12-v2') 
        from prompts"

Interactive Shell

If you don't provide an SQL query, you'll enter an interactive shell instead.

tsellm prompts.db

Installation

pip install tsellm

How

tsellm relies on the following facts:

• SQLite is bundled with the standard Python library (import sqlite3)
• Python 3.12 ships with a SQLite interactive shell
• one can create Python-written user-defined functions to be used in SQLite queries (see create_function)
• Simon Willison has gone through the process of creating the beautiful llm Python library and CLI

Development

pip install -e '.[test]'
pytest