🤗 Optimum Neuron Model Exporter 🏎️

optimum-neuron version: 0.4.1

This Space allows you to automatically export 🤗 transformers to AWS Neuron-optimized format for Inferentia/Trainium acceleration.

Simply provide a model ID from the Hugging Face Hub, and choose your desired output.

✨ Key Features

• 🚀 Create a New Optimized Repo: Automatically converts your model and uploads it to a new repository under your username (e.g., your-username/model-name-neuron).
• 🔗 Link Back to Original: Creates a Pull Request on the original model's repository to add a link to your optimized version, making it easier for the community to discover.
• 🛠️ PR to a Custom Repo: For custom workflows, you can create a Pull Request to add the optimized files directly into an existing repository you own.
• 📦 Contribute to Cache: Contribute the generated compilation artifacts to a centralized cache repository (or your own private cache), helping avoid recompilation of already exported models.

⚙️ How to Use

1. Model ID: Enter the ID of the model you want to export (e.g., bert-base-uncased or stabilityai/stable-diffusion-xl-base-1.0) and choose the corresponding task.
2. Export Options: Select at least one option for where to save the exported model. You can provide your own cache repo ID or use the default (aws-neuron/optimum-neuron-cache).
3. Convert & Upload: Click the button and follow the logs to track progress!