Genesis-Embodied-AI/genesis-world · critical · ImportError

'torch' module not available. Please install pytorch manuall

Error message

'torch' module not available. Please install pytorch manually: https://pytorch.org/get-started/locally/

What it means

Genesis hard-depends on PyTorch for its tensor pipeline, so genesis/__init__.py imports torch right after quadrants at package import time. If the import raises ImportError, it is re-raised with an actionable message pointing at PyTorch's install wizard. Hitting it means the active Python environment simply has no usable torch install.

Source

Thrown at genesis/__init__.py:23

import logging as _logging
import traceback
import weakref
from typing import Callable
from warnings import warn
from contextlib import redirect_stdout

# Import quadrants while collecting its output without printing directly
_qd_outputs = io.StringIO()

os.environ.setdefault("QD_ENABLE_PYBUF", "0" if sys.stdout is sys.__stdout__ else "1")

with redirect_stdout(_qd_outputs):
    import quadrants as qd

try:
    import torch
except ImportError as e:
    raise ImportError(
        "'torch' module not available. Please install pytorch manually: https://pytorch.org/get-started/locally/"
    ) from e
import numpy as np

from .constants import backend as _gs_backend
from .logging import Logger
from .version import __version__
from .utils import redirect_libc_stderr, set_random_seed, get_device


_IS_OLD_TORCH = tuple(map(int, torch.__version__.split(".")[:2])) < (2, 8)
# FIXME: qd.Field does not support zero-copy on Metal for 'torch<=2.9.1'.
# See: https://github.com/pytorch/pytorch/pull/168193
_TORCH_MPS_SUPPORT_DLPACK_FIELD = torch.torch_version.TorchVersion(torch.__version__.split("+", 1)[0]) > (2, 9, 1)
if _IS_OLD_TORCH:
    warn("'torch<2.8.0' is not supported. Please upgrade pytorch manually: https://pytorch.org/get-started/locally/")

View on GitHub (pinned to 56e4aa5d82)

Solutions

1. Install PyTorch for your platform: `pip install torch` (CPU) or the CUDA wheel from https://pytorch.org/get-started/locally/.
2. Verify you are in the same interpreter Genesis uses: `python -c "import torch; print(torch.__version__)"`.
3. If torch is installed but fails to import, reinstall the wheel matching your CUDA/Python version (`pip install --force-reinstall torch`).

Example fix

# before
import genesis  # ImportError: 'torch' module not available

# after
pip install torch
import genesis  # OK

import importlib.util
if importlib.util.find_spec("torch") is None:
    raise SystemExit("PyTorch is required. Install it from https://pytorch.org/get-started/locally/")
import genesis

try:
    import genesis
except ImportError as e:
    if "torch" in str(e):
        print("Install pytorch first: pip install torch")
    raise

When it happens

Trigger: `import genesis` (or anything importing it) in a Python environment where `pip install torch` was never run, where torch is installed for a different interpreter/conda env, or where a broken torch install raises ImportError on load.

Common situations: Fresh clone of Genesis used with an incomplete `pip install -e .` that skipped torch; wrong conda env activated; system pip vs virtualenv mismatch; CUDA-incompatible torch wheel failing to import on machines without the right CUDA runtime.

Related errors

• Python module 'uipc' is required by IPCCoupler but is not in
• `morph` in hybrid entity should be either URDF or Mesh
• Cannot handle soft material {material_soft}
• Compounding joints of types 'FREE' or 'FIXED' with any other
• This method has been removed. Please use 'get_AABB()' instea

AI-assisted analysis of Genesis-Embodied-AI/genesis-world@56e4aa5d82 (2026-08-28). Data as JSON: /api/errors/27ccb05103fab44f. Report an issue: GitHub.