huggingface/transformers · error · RuntimeError

return_tensors set to 'pt' but PyTorch can't be imported

Error message

return_tensors set to 'pt' but PyTorch can't be imported

What it means

The featurizer can package examples as a torch TensorDataset when return_tensors='pt', but transformers is installed without PyTorch in the current environment (is_torch_available() is False). The library raises RuntimeError instead of attempting the import so you get a clear message rather than a ModuleNotFoundError for torch.

Solutions

1. Install PyTorch in the active environment (pip install torch), then retry.
2. If you deliberately run without torch, call the featurizer with return_tensors=None and consume the list of InputFeatures.
3. Verify the environment with python -c "import torch" and check you are in the interpreter/venv you think you are.

Example fix

# before
features = featurizer.get_features(texts, return_tensors="pt")  # no torch installed

# after (option A): install torch
# pip install torch
# after (option B): stay framework-free
features = featurizer.get_features(texts, return_tensors=None)

from transformers.utils import is_torch_available
if return_tensors == "pt" and not is_torch_available():
    raise RuntimeError("torch required for return_tensors='pt'; pip install torch or use return_tensors=None")

try:
    ds = featurizer.get_features(texts, return_tensors="pt")
except RuntimeError as e:
    if "PyTorch" in str(e):
        features = featurizer.get_features(texts, return_tensors=None)  # graceful degrade
    else:
        raise

When it happens

Trigger: Calling get_features(..., return_tensors='pt') in an environment where torch is not installed; running on a CPU-only slim install of transformers (pip install transformers without torch); a venv/container where torch was uninstalled or never installed.

Common situations: Using transformers only for tokenizers/ONNX/JAX and then trying the PyTorch tensor path; CI images that omit torch to save space; switching conda envs mid-project.

Related errors

• Missing requirements in your local environment for
• PyTorch must be installed to return a PyTorch dataset.
• This modeling file requires the following packages that…
• To use , please install the following dependencies
• requires newer versions of

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/5fbb70826e938ae3. Report an issue: GitHub.

Appendix: source

Thrown at src/transformers/data/processors/utils.py:316

            elif self.mode == "regression":
                label = float(example.label)
            else:
                raise ValueError(self.mode)

            if ex_index < 5 and self.verbose:
                logger.info("*** Example ***")
                logger.info(f"guid: {example.guid}")
                logger.info(f"input_ids: {' '.join([str(x) for x in input_ids])}")
                logger.info(f"attention_mask: {' '.join([str(x) for x in attention_mask])}")
                logger.info(f"label: {example.label} (id = {label})")

            features.append(InputFeatures(input_ids=input_ids, attention_mask=attention_mask, label=label))

        if return_tensors is None:
            return features
        elif return_tensors == "pt":
            if not is_torch_available():
                raise RuntimeError("return_tensors set to 'pt' but PyTorch can't be imported")
            import torch
            from torch.utils.data import TensorDataset

            all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long)
            all_attention_mask = torch.tensor([f.attention_mask for f in features], dtype=torch.long)
            if self.mode == "classification":
                all_labels = torch.tensor([f.label for f in features], dtype=torch.long)
            elif self.mode == "regression":
                all_labels = torch.tensor([f.label for f in features], dtype=torch.float)

            dataset = TensorDataset(all_input_ids, all_attention_mask, all_labels)
            return dataset
        else:
            raise ValueError("return_tensors should be `'pt'` or `None`")

View on GitHub (pinned to a597f97485)