Installation#

Synora supports multiple installation methods depending on your use case.

From PyPI#

For stable releases:

# Core dependencies (torch, torchvision, torchaudio, gym, gymnasium, etc.)
pip install synora

# With specific extras
pip install synora[gym]       # Additional gym environments (huggingface-hub, pygame, autorom)
pip install synora[ml-agents] # Unity ML-Agents support
pip install synora[ml]        # TensorBoard, Weights & Biases, logging tools
pip install synora[viz]       # FastAPI, Uvicorn, documentation tools
pip install synora[docs]      # Sphinx and documentation tools
pip install synora[dev]       # Testing and development tools (pytest, mypy, pre-commit)

# Install multiple extras
pip install synora[gym,ml-agents,dev]

Available Extras#

Extra	Description
`gym`	Additional Gym environment dependencies (huggingface-hub, pygame, autorom)
`ml-agents`	Unity ML-Agents support
`ml`	TensorBoard, Weights & Biases, and logging tools
`viz`	Latent-space visualization (opencv-python, umap-learn, scikit-learn, plotly)
`docs`	Sphinx and documentation building tools
`dev`	pytest, mypy, pre-commit for development

Core Dependencies#

The minimal installation includes only: torch, torchvision, click, einops, tqdm, and pyyaml. Everything else — environment backends, logging, and visualization — comes from the extras above.

From Source#

For the latest development version:

git clone https://github.com/ParamThakkar123/synora.git
cd synora

# Core dependencies
pip install -e .

# With extras
pip install -e ".[gym,ml-agents,ml,viz,dev,docs]"

PyTorch Build Selection#

Synora does not pin a single PyTorch wheel index. Install the PyTorch build that matches your platform (CPU, macOS, CUDA, ROCm, etc.) using the index recommended by the PyTorch installation selector.

# Example: CUDA 12.1 wheels. Replace the index for CPU, ROCm, or other CUDA versions.
uv add torch torchvision torchaudio --index https://download.pytorch.org/whl/cu121

# Or using pip.
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Docker#

Build the default CPU image and run the Synora CLI:

# Build
docker build -t synora .

# Show the CLI help
docker run --rm synora

# Run a specific command
docker run --rm synora models list

The Dockerfile installs PyTorch explicitly before installing Synora so the wheel source is controlled by the PYTORCH_INDEX_URL build argument. The default uses CPU wheels. To build against a CUDA wheel index, pass the matching PyTorch index and run the container with the NVIDIA runtime:

docker build \
  --build-arg PYTORCH_INDEX_URL=https://download.pytorch.org/whl/cu121 \
  -t synora:cu121 .

docker run --rm --gpus all synora:cu121 models list

Additional optional dependency groups can be installed at build time with SYNORA_EXTRAS, for example --build-arg SYNORA_EXTRAS=viz,ml. Runtime data is stored under /data/synora, which you can persist with a bind mount or volume.

Verification#

Verify your installation:

import torch
import synora

print(f"PyTorch: {torch.__version__}")
print(f"CUDA available: {torch.cuda.is_available()}")
print("Synora imported successfully!")