pytei-client 0.1.0

Minimal wrapper for Hugging Face's Text Embeddings Inference.

PyTEI

PyTEI is a minimal python interface for Hugging Face's Text Embeddings Inference.

PyTEI supports in-memory and persistent caching for text embeddings.

Installation

First, clone the git repository by running:

git clone https://github.com/daniel-gomm/PyTEI.git

Next, install this repository as python package using pip by running the following command from the root directory of this repository:

pip install .

Add the -e-flag in case you want to modify the code.

Usage

Prerequisite for using PyTEI is a running Text Embeddings Inference instance, for example a local docker container running TEI. Such a docker contain can be spun-up by running:

docker run --gpus all -p 8080:80 \
  -v $PWD/data:/data \
  --pull always ghcr.io/huggingface/text-embeddings-inference:1.6 \
  --model-id Alibaba-NLP/gte-Qwen2-1.5B-instruct

TEI Client

For more details check out the Documentation.

Establish a connection to TEI through a TEIClient. The client gives you access to the text-embedding API of the TEI instance:

from pytei import TEIClient

client = TEIClient(url="127.0.0.1:8080/embed")

text_embedding = client.embed("Lorem Ipsum")

The default configuration uses in-memory caching of embeddings. For persistent caching use the DuckDBDataStore or implement your own caching solution by extending the DataStore base-class.

from pytei import TEIClient
from pytei.store import DuckDBEmbeddingStore

persistent_data_store = DuckDBEmbeddingStore(db_path="data/embedding_database.duckdb")
client = TEIClient(embedding_store=persistent_data_store, url="127.0.0.1:8080/embed")

text_embedding = client.embed("Lorem Ipsum")

For a more detailed description and the full description of the API check out the Documentation