Export Hugging Face models to ONNX
Submit models for evaluation and view leaderboard
Evaluate adversarial robustness using generative models
Create and upload a Hugging Face model card
Evaluate reward models for math reasoning
Persian Text Embedding Benchmark
Browse and submit evaluations for CaselawQA benchmarks
Evaluate and submit AI model results for Frugal AI Challenge
View and submit LLM benchmark evaluations
Upload ML model to Hugging Face Hub
Measure execution times of BERT models using WebGPU and WASM
Merge Lora adapters with a base model
Display and submit LLM benchmarks
Export to ONNX is a tool designed to convert machine learning models from the Hugging Face ecosystem into the Open Neural Network Exchange (ONNX) format. ONNX is an open standard that allows models to be exported and used across different frameworks and platforms, enabling interoperability and deployment in various environments. This tool simplifies the process of transitioning models for inference or further development in frameworks that support ONNX.
• Cross-Framework Compatibility: Convert models from Hugging Face to ONNX format for use in frameworks like PyTorch, TensorFlow, or Microsoft Cognitive Toolkit (CNTK).
• Optimization for Inference: ONNX models are often optimized for inference, making them suitable for production environments.
• Simplified Export Process: Streamlined workflow for converting models with minimal effort.
• Scalability: Supports a wide range of model architectures, including popular transformers and other deep learning models.
transformers and torch-onnx.What models are supported for export?
• Most Hugging Face models, including popular transformer-based architectures, are supported for export to ONNX.
Why should I convert my model to ONNX?
• Converting to ONNX allows for better interoperability and optimization, making it easier to deploy models in production environments.
How do I handle complex or custom models?
• For complex or custom models, ensure all operations are supported in ONNX. You may need to modify the model or use additional tools to handle unsupported layers.