Instella: New Open 3B Language Models

AMD has introduced Instella, a family of open-source 3-billion-parameter language models trained on AMD Instinct MI300X GPUs. These models surpass existing open models of comparable size and rival leading open-weight models like Llama-3.2-3B and Qwen-2.5-3B. Trained on up to 4.15 trillion tokens using advanced techniques, Instella's release includes pre-trained and instruction-tuned versions, along with all model weights, configurations, datasets, and code to promote AI research collaboration.