cached-sentence-transformer 0.1.0

PostgreSQL-backed embedding cache for SentenceTransformers

Cached Sentence Transformer

PostgreSQL-backed embedding cache for SentenceTransformers.

This package provides a small wrapper that caches computed sentence embeddings in a Postgres table keyed by a stable hash of (model, normalize flag, text), so repeated runs can reuse stored vectors instead of recomputing them.

Installation

pip install cached-sentence-transformer

Quickstart

from cached_sentence_transformer import CachedSentenceTransformer

st = CachedSentenceTransformer(
    model_name_or_path="sentence-transformers/all-MiniLM-L6-v2",
    pg_dsn="host=localhost port=5432 dbname=mydb user=myuser password=mypassword",
)

emb = st.encode(["hello", "world"], normalize_embeddings=True)
st.close()

Environment-based DSN

If you do not pass pg_dsn, the wrapper will attempt to build it from environment variables (auto loads environment variables in the .env file in the current working directory) and will fail fast if any are missing:

• PSQL_HOST_NAME
• PSQL_PORT
• PSQL_DBNAME
• PSQL_USER
• PSQL_PASSWORD