About the Organization

The AuroraGPT project engages diverse specialists who are organized into eight working groups, each with a distinct focus and responsibilities. 

""

 

Working groups within the AuroraGPT address key research and development factors within and among the stages of the AuroraGPT FM workflow shown here, with additional working groups for management and planning, communications, and distribution.

Working Groups

Planning

Identifies emerging approaches that may advance the training, evaluation, or usage of scientific foundation models.

Data

Preparation, Curation, Deduplication, and Optimization of Natural and Synthetic Scientific Text and Data.

Models

Core model pre-training and model architecture developments.

Evaluation

Multifaceted evaluation and benchmarking of scientific skill and safety.

Post-Training

Improve model performance by tailoring responses, refining instructions, and optimizing preferences and biases.

Inference

Develops high-performance inference to support fine-tuned versions of AuroraGPT for specialized purposes.

Communications

Communicate goals, organization, and progress; Coordinate partnerships.

Distribution

Empowering collaboration and enabling safe sharing of AuroraGPT outcomes and artifacts.