Apertus LLM Family Expansion via Distillation and Quantization
For practitioners needing LLMs that fit diverse hardware constraints, this provides a practical recipe for expanding model families, though the approach is incremental.
The authors validate distillation and quantization as cost-effective methods to expand LLM families to new sizes and hardware formats, producing Apertus-v1.1 models up to 4B parameters trained on 1.7T tokens with strong accuracy.
The wide adoption of LLMs has led to their use in great variety of applications and scenarios, such as chatbot assistants and data annotation, creating the need for the models to satisfy certain budget and hardware constraints. This has led to the trend of LLMs being released in batches consisting of similar models of various sizes for the family of models to adhere to as wide of a range of constraints as possible. In this paper, we validate distillation and quantization as a cost-effective way to expand model families to new sizes and hardware formats. Based on the open-recipe Apertus 8B LLM, we produce Apertus-v1.1 - a distilled family of models with up to 4B parameters trained on 1.7T permissive license tokens. We demonstrate cost-efficiency and strong accuracy performance of our approach for covering large ranges of hardware and systems requirements.