Registry / ai-ml / simpletransformers

simpletransformers

library0.70.8pypypiunverified

Simple Transformers is an easy-to-use wrapper library for the Transformers library, providing a unified API for training and using transformer models for NLP tasks such as classification, NER, QA, LM, etc. Current version is 0.70.8, released in 2022. Development is in maintenance mode with infrequent updates.

ai-ml

pip install simpletransformers

github ↗homepage ↗

INSTALL

IMPORT

SIG · SIMPLETRANSFORMERS

simpletransformers

ai-mlpythonv0.70.8

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

ClassificationModel

✓ from simpletransformers import ClassificationModel

✗ from simpletransformers.classification import ClassificationModel

Minimal example: train a DistilBERT classifier and predict. Set use_cuda=False if no GPU.

from simpletransformers.classification import ClassificationModel import pandas as pd # Prepare data train_data = [ ["I love this product!", 1], ["This is terrible.", 0], ] train_df = pd.DataFrame(train_data, columns=["text", "labels"]) # Initialize model (uses distilbert by default) model = ClassificationModel( "distilbert", "distilbert-base-uncased", use_cuda=False, args={"num_train_epochs": 1, "no_save": True}, ) # Train model.train_model(train_df) # Predict predictions, raw_outputs = model.predict(["Great product!"],) print(predictions)

Debug

Known issues

breakingImport paths changed from top-level to submodule imports in v0.65+. Direct imports like `from simpletransformers import ClassificationModel` no longer work.

fix

Use from simpletransformers.classification import ClassificationModel.

affects: >=0.65.0

deprecatedThe argument `num_labels` is deprecated. Use `labels_list` in args to specify label mapping.

fix

Set args={'labels_list': [0, 1]} instead of passing num_labels to model constructor.

affects: >=0.65.0

gotchaWhen using a custom model path, set `model_type` correctly (e.g., 'bert', 'roberta') otherwise model loading silently fails or uses wrong architecture.

fix

Check HuggingFace model type and pass matching `model_type` string.

affects: all

gotchaTraining may hang or crash if `use_cuda` is set to True but no GPU is available. The library does not always fall back to CPU gracefully.

fix

Set `use_cuda=False` explicitly in non-GPU environments, or detect GPU availability.

affects: all

breakingThe `wandb_project` argument no longer works; wandb must be configured externally using `wandb.init()`.

fix

Initialize wandb before creating the model: import wandb; wandb.init(project='my_project').

affects: >=0.65.0

gotchaData format changed: classification models now expect a pandas DataFrame with columns 'text' and 'labels'. Using lists or dictionaries may raise errors.

fix

Convert your data to a DataFrame: pd.DataFrame({'text': [...], 'labels': [...]}).

affects: >=0.65.0

Upgrade

Version history

0.70.8latest on PyPI · released May 31, 2026

Audit

Dependencies

transformersrequiredCore dependency for all transformer models.

torchoptionalRequired for PyTorch backend.

tensorflowoptionalOptional TensorFlow backend.

seqevaloptionalRequired for NER evaluation metrics.

scikit-learnrequiredUsed for classification metrics.

Agent activity

18 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/ThilinaRajapakse/simpletransformers/ ↗

homepagehttps://github.com/ThilinaRajapakse/simpletransformers/ ↗