Export Hugging Face models to ONNX
Evaluate code generation with diverse feedback types
Persian Text Embedding Benchmark
Calculate GPU requirements for running LLMs
View and submit LLM benchmark evaluations
View and submit machine learning model evaluations
Display and submit LLM benchmarks
Upload ML model to Hugging Face Hub
Compare model weights and visualize differences
Explore GenAI model efficiency on ML.ENERGY leaderboard
Launch web-based model application
Request model evaluation on COCO val 2017 dataset
Measure BERT model performance using WASM and WebGPU
Export to ONNX is a tool designed to convert Hugging Face models into the ONNX (Open Neural Network Exchange) format. This allows users to export their models for deployment across various platforms, frameworks, and devices. By enabling this conversion, Export to ONNX enhances model portability, cross-framework compatibility, and efficiency in production environments.
• Converts Hugging Face models to ONNX format for broader compatibility
• Supports multiple deep learning frameworks for inference
• Optimizes models for faster inference and reduced latency
• Enables deployment on edge devices and cloud platforms
• Simplifies model sharing and collaboration across teams
• Integrates seamlessly with the Hugging Face ecosystem
pip install onnx
to ensure you have the ONNX library installed.transformers
or torch
library to load your model.
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model_name = "bert-base-uncased"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
torch.onnx.export
function to convert the model.
import torch
dummy_input = "This is a sample input"
inputs = tokenizer(dummy_input, return_tensors="pt")
with torch.no_grad():
torch.onnx.export(
model,
inputs["input_ids"],
"model.onnx",
opset_version=12,
input_names=["input_ids"],
output_names=["output"],
)
What models are supported for export to ONNX?
Export to ONNX supports a wide range of Hugging Face models, including BERT, RoBERTa, and other popular architectures. However, some niche or custom models may require additional configuration.
How do I optimize my ONNX model for inference?
ONNX models can be optimized using tools like ONNX Runtime or TensorRT. These tools can reduce latency and improve performance on target hardware.
What if the model fails to convert to ONNX?
If conversion fails, check for unsupported operations in your model. Some custom layers or operations may not be compatible with ONNX. You may need to modify your model architecture or update your version of ONNX.