huggingface/optimum

Health v2 · maintenance only

Adoption not part of the score

3469 stars · 673 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Optimum is a Hugging Face library that extends Transformers, Diffusers, timm, and Sentence Transformers with hardware-specific optimization tools for faster training and inference. It provides drop-in model classes and export/quantization utilities targeting accelerators like ONNX Runtime, OpenVINO, TensorRT-LLM, AMD, AWS Neuron, and Habana Gaudi.

Use cases

• speed up huggingface transformer inference
• convert pytorch model to onnx
• quantize a model to int8 for faster inference
• run transformers on intel openvino
• fine-tune models on habana gaudi or aws trainium
• accelerate stable diffusion inference on gpu
• export transformers model to tensorrt-llm

When to choose

• you use the Hugging Face ecosystem and want faster inference or training on specific hardware
• you need model export to ONNX, OpenVINO, or TensorRT-LLM with minimal code changes
• you want post-training quantization or pruning of Transformers/Diffusers models
• you deploy on partner accelerators like Habana Gaudi, AWS Inferentia/Trainium, or AMD NPUs

When to avoid

• you only run small models on CPU where optimization overhead isn't worth it
• you need a runtime-agnostic serving solution rather than a Python optimization library
• your hardware has no Optimum integration and you'd rather use the vendor's native tooling directly

Facets

library · maturity active

machine-learning llm-inference llm-training gpu-computing cli machine-learning deep-learning large-language-models developer-tools gpu-computing python cross-platform huggingface transformers onnx onnxruntime openvino quantization model-optimization tensorrt-llm habana-gaudi aws-neuron diffusers hardware-acceleration gpu docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="huggingface/optimum")