lyhue1991/torchkeras

Health v2 · maintenance only

Adoption not part of the score

2009 stars · 250 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

torchkeras is a lightweight PyTorch model training template library that brings Keras-style compile/fit/evaluate APIs to PyTorch. Its core is only ~200 lines of modular code, with built-in progress bars, metrics, early stopping, checkpointing, and dynamic visualization in Jupyter notebooks.

Use cases

• train pytorch models without writing training loops
• keras-style fit API for pytorch
• add early stopping and metrics to pytorch training
• visualize training curves in jupyter notebook
• finetune models with a simple trainer
• replace messy custom trainers in pytorch libraries

When to choose

• you want Keras-like simplicity for PyTorch training
• you need a short, hackable training template you can modify
• you train in Jupyter and want live plots and pretty logs

When to avoid

• you need distributed training at scale or advanced Trainer features
• you already use PyTorch Lightning or Hugging Face Trainer workflows

Facets

library · maturity active

machine-learning deep-learning llm-training data-visualization logging deep-learning machine-learning developer-tools python cross-platform pytorch keras-api training-loop model-training jupyter early-stopping tensorboard wandb

Member repositories

For agents

markdown · JSON · MCP: product_card(name="lyhue1991/torchkeras")