MaartenGr/KeyBERT: v0.9

Version 0.9.0

Release date: 05 Februari, 2025

Model2Vec

You can use Model2Vec for blazingly fast embeddings as follows:

from keybert import KeyBERT
from model2vec import StaticModel

embedding_model = StaticModel.from_pretrained("minishlab/potion-base-8M")
kw_model = KeyBERT(embedding_model)

Light-weight KeyBERT

You can now install a light-weight KeyBERT with:

pip install keybert --no-deps scikit-learn model2vec

Fixes

• Add Model2Vec & light-weight installation in #253
• Add Text Generation Inference with JSON output by @joaomsimoes in #235
• Update pre-commit hooks @afuetterer in #237
• Set up lint job using pre-commit/action @afuetterer in #238

Version 0.8.5

Release date: 14 June, 2024

• Use batch_size parameter with keybert.backend.SentenceTransformerBackend by @adhadse in #210
• Add system_prompt param to LLMs by @lucafirefox in #214
• Update OpenAI API response by @lucafirefox in #213
• Drop support for python 3.6 and 3.7 by @afuetterer in #230
• Bump github actions versions by @afuetterer in #228
• Switch from setup.py to pyproject.toml by @afuetterer in #231

Version 0.8.4

Release date: 15 Februari, 2024

• Update default Cohere model to command by @sam-frampton in #194
• Fix KeyLLM fails when no GPU is available by @igor-pechersky in #201
• Fix AttributeError: 'tuple' object has no attribute 'page_content' in LangChain in #199

Version 0.8.3

Release date: 29 November, 2023

• Fix support for openai>=1

You can now use it as follows:

import openai
from keybert.llm import OpenAI
from keybert import KeyLLM

# Create your LLM
client = openai.OpenAI(api_key=MY_API_KEY)
llm = OpenAI(client)

# Load it in KeyLLM
kw_model = KeyLLM(llm)

Version 0.8.2

Release date: 29 September, 2023

• Fixed cuda error when using pre-calculated embeddings with KeyBERT + KeyLLM

Version 0.8.1

Release date: 29 September, 2023

• Remove unnecessary print statements