A white paper describes eight Open ELM models, four are pretrained with the Corenet library and four are fine-tuned with instructions. Apple uses a strategy where they scale layers to increase both accuracy and efficiency, writes the site Macrumors.
Apple shares not only the final trained model but also code and training logs. The researchers behind the project hope this will lead to faster progress and more reliable results in natural language AI.
“Unlike before when we only shared model weights and code for inference, and trained on private datasets, our release includes a complete framework for training and evaluating the language model with publicly available datasets, training logs, multiple checkpoints and pre-training configurations,” write the developers on the Hugging site Face Hub.
Apple says it is releasing the OpenELM models to “strengthen and enrich the open research community.” Sharing open source code gives researchers the opportunity to investigate risks as well as investigate data and model bias. Developers and companies can use the models as is or make changes.
