Pruna AI open sources its AI model optimization framework

Pruna AI, a European startup focused on AI model compression, is open-sourcing its optimization framework, which integrates methods like caching, pruning, quantization, and distillation. This framework standardizes the process of saving, loading, and evaluating compressed models, similar to how Hugging Face standardized transformers. Pruna AI's tool is designed to combine various efficiency methods, making it easier for developers to optimize models without significant quality loss. The company is currently focusing on image and video generation models and offers an enterprise version with advanced features, including a compression agent that automates model optimization based on user-defined criteria.

Pruna AI has raised $6.5 million in seed funding from investors such as EQT Ventures and Daphni. The startup aims to provide a cost-effective solution for AI infrastructure by significantly reducing model sizes and inference costs. The company charges for its pro version on an hourly basis, akin to renting a GPU on cloud services. Existing users include Scenario and PhotoRoom, and Pruna AI hopes its framework will be seen as a valuable investment.

Key takeaways:

Pruna AI is open-sourcing its optimization framework for AI model compression, which includes methods like caching, pruning, quantization, and distillation.
The framework can evaluate the quality loss and performance gains after model compression, similar to how Hugging Face standardized transformers and diffusers.
Pruna AI's enterprise offering includes an optimization agent that automatically optimizes models based on user-defined criteria, charging by the hour for its pro version.
Pruna AI recently raised a $6.5 million seed funding round, with investors including EQT Ventures, Daphni, Motier Ventures, and Kima Ventures.

Pruna AI open sources its AI model optimization framework | TechCrunch

Key takeaways:

Comments (0)

Newsletter