Section 1 – Information for all providers of general-purpose AI models:
1. General description:
(a) intended tasks and AI systems it can integrate into
(b) acceptable use policies
(c) release date and distribution methods
(d) architecture and number of parameters
(e) input/output modality and format
(f) licence
2. Detailed description:
(a) technical means for integration (instructions, infrastructure, tools)
(b) design specifications, training methodologies, key design choices
(c) data information: type, provenance, curation methodologies, number of data points
(d) computational resources used for training
(e) known or estimated energy consumption
Section 2 – Additional information for providers of GPAI models with systemic risk:
1. Detailed description of evaluation strategies and results
2. Description of adversarial testing measures (red teaming, alignment, fine-tuning)
3. Description of system architecture