Skip to content

Instella

Open language models trained on AMD Instinct GPUs

Visit official site

What Instella is

The project is described as a series of open language models from the AMD GenAI team, trained on AMD Instinct MI300X GPU for natural language processing tasks.

What you can do with it

  • Use the high performance language generation capability
  • Customise the open model weights and training code
  • Run pretrained or instruction tuned variants
  • Train across multiple nodes

Getting started

  • Install PyTorch and dependencies with AMD GPU support
  • Clone the repository and add Flash-Attention
  • Load a pretrained model with the Hugging Face API
  • Run inference or fine tuning as needed

Application scenarios

  • Researchers explore new natural language processing methods
  • Enterprises build a customer service or content tool
  • Developers adapt the open weights for a business need

Pros & cons

✓ What we like

  • Multi-node distributed training is listed
  • Both pretrained and instruction tuned models are provided

! What to watch out for

  • Training is described on AMD Instinct MI300X GPU
  • Loading uses the Hugging Face API

FAQ

Who developed the models?

The material names the AMD GenAI team as the developer.

What is included in the release?

Model weights, training code, training data, pretrained models and instruction tuned models are listed.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review