Common file types and technical specifications in AI.

Model Files

• .safetensors - Secure format for storing AI model weights
• .ckpt - Checkpoint files that save model training progress
• .pth - PyTorch model files containing weights and architecture
• .pb - TensorFlow's protobuf format for saved models
• .h5 - HDF5 format used for Keras and other frameworks

Optimized Formats

• .gguf - Efficient format for running models on consumer hardware
• .ggml - Earlier version of GGUF for CPU inference
• ONNX - Cross-platform format that works with different frameworks
• TensorRT - NVIDIA's optimized format for GPU inference

Data Files

• .jsonl - JSON Lines format for storing training data
• .parquet - Columnar format efficient for large datasets
• .arrow - Memory format for fast data processing
• .tfrecord - TensorFlow's format for training data

Configuration

• .yaml - Human-readable format for configuration files
• .toml - Simple format for configuration and metadata
• model cards - Documentation files describing AI models
• .env - Files storing environment variables and API keys

Archive Formats

• .tar.gz - Compressed archives common in ML repositories
• .zip - Standard compressed format for model distributions
• Git LFS - System for storing large model files in Git

← Back to AI Glossary