How to create a Conda environment for TensorFlow

Conda environments give TensorFlow projects their own Python interpreter, pip entry point, and package directories. That separation matters when notebooks, training jobs, and other machine-learning projects need different versions of TensorFlow, Keras, NumPy, or supporting libraries.

TensorFlow is still installed from the official PyPI package in this workflow. Conda creates and activates the environment, while python -m pip install tensorflow installs the TensorFlow wheel inside that active environment instead of resolving the framework from Conda channels.

Python 3.12 is a conservative target for a new TensorFlow environment because it stays inside TensorFlow's supported wheel range and is widely available through Conda. Use Linux or WSL2 for current NVIDIA GPU work; native Windows GPU support ended with TensorFlow 2.10, and official macOS installs are CPU-only.

1. Open a terminal where Conda is already available.
2. Create a dedicated Conda environment with Python 3.12 and pip.

   $ conda create --yes --name tf python=3.12 pip
   Collecting package metadata (repodata.json): done
   Solving environment: done
   ##### snipped #####
   Preparing transaction: done
   Verifying transaction: done
   Executing transaction: done

   The --yes option accepts Conda's package plan without an interactive prompt. Omit it when you want to review the package list before the environment is created.

3. Activate the new environment before installing TensorFlow.

   $ conda activate tf
   (tf) $

4. Confirm the environment is using the expected Python version.

   (tf) $ python --version
   Python 3.12.13

5. Confirm pip points inside the Conda environment.

   (tf) $ python -m pip --version
   pip 26.1.2 from /home/user/miniforge3/envs/tf/lib/python3.12/site-packages/pip (python 3.12)

   The path should include /envs/tf/ before the site-packages directory. If it points outside the Conda environment, reactivate the environment before continuing.

6. Upgrade pip inside the active environment.

   (tf) $ python -m pip install --upgrade pip
   Requirement already satisfied: pip in /home/user/miniforge3/envs/tf/lib/python3.12/site-packages (26.1.2)

   This upgrades only the active tf environment, not the base Conda environment or the system Python installation.

7. Install TensorFlow from PyPI with the active environment's pip.

   (tf) $ python -m pip install tensorflow
   Collecting tensorflow
   ##### snipped #####
   Successfully installed tensorflow-2.21.0

   TensorFlow's install docs recommend pip for the tensorflow package because TensorFlow is officially released to PyPI, not as the latest stable Conda package. For current Linux or WSL2 GPU systems, install tensorflow[and-cuda] instead of tensorflow after the NVIDIA driver is working.
   Related: How to enable GPU acceleration in TensorFlow

8. Import TensorFlow and run a small tensor operation.

   (tf) $ python -c "import tensorflow as tf; print(tf.__version__); print(tf.reduce_sum(tf.constant([1, 2, 3])).numpy())"
   2.21.0
   6

   A printed TensorFlow version plus the numeric tensor result confirms that the package imports and executes inside the active Conda environment.
   Related: How to check the TensorFlow version