How to install Sentence Transformers with Conda

Conda environments are a common way to keep Python machine learning packages separate from the system Python and from other projects. Installing Sentence Transformers from conda-forge keeps the library, PyTorch, Transformers, and numerical dependencies in one solver-managed environment for embedding work.

The conda-forge package provides the base Sentence Transformers text embedding library. Pip-style extras such as image, audio, video, ONNX, and OpenVINO are not Conda extras, so keep this environment focused on the standard package unless the project has a separate extras plan.

Use a fresh environment instead of installing into base. Keeping the solve on conda-forge avoids mixing incompatible package channels, and a small model smoke test confirms the installed package can load a public model and return embeddings.

1. Open a terminal where conda is initialized.
2. Create an isolated conda-forge environment with Sentence Transformers.

   $ conda create --yes --name st-conda --channel conda-forge --override-channels python=3.11 sentence-transformers

   --override-channels keeps this solve on conda-forge for the new environment. Use another supported Python version only when the project already standardizes on it.

3. Activate the new environment.

   $ conda activate st-conda

4. Confirm that Python imports the installed package.

   $ python -c "import sentence_transformers; print(sentence_transformers.__version__)"
   5.6.0

   The exact version changes as conda-forge publishes new builds. The import must run from the activated environment.

5. Run a small embedding smoke test.

   $ python - <<'PY'
   from sentence_transformers import SentenceTransformer
   
   model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
   embeddings = model.encode([
       "Conda installed Sentence Transformers",
       "Embeddings are ready",
   ])
   print(embeddings.shape)
   PY
   ##### snipped
   (2, 384)

   The first model load may download files from Hugging Face. The shape shows two input texts encoded with the 384-dimensional all-MiniLM-L6-v2 model.
   Related: Set the Sentence Transformers cache directory