- Ambarish Gurjar (Indiana University)
- Ameya Naik
- Andreas Fehlner
- Arthit Suriyawongkul
- Arun Prabhakar
- Gopi Krishnan Rajabhadur
- Helen Oakley
- Jay White
- Martin
- Stefana Simion
- Brainstorming additional fields in 3.1
- CISA SBOM Tiger Team AIBOM sign-up: https://docs.google.com/document/d/11UU_Wiaemi7zBs3sE-MgovieyPx1XJEOaju2EM5btts/edit#heading=h.f0icpudrrs5j
- Discussion on 3.1 fields
- Prompt-related properties and relationships
- System prompt
- System prompt provenance
- Size of the system prompt
- Prompt size (in tokens, etc), or input/output size in general
- Max size supported for the system prompt
- Prompt technique
- Foundational model-runtime configuration (dict)
- Computational resource consumption
- In addition to the existing energy consumption
- Training time/resource consumption (FLOPs / to decide on unit(s))
- Inference time/resource (GPU)
- Deployment time/resource (GPU)
- Seperate system prompt and user prompt
- May need a Prompt Profile?
- We need to think about RAG (Retrieval-Augmented Generation)
- Relationships
- fintunedOn - Relationship that links a dataset profile to the foundational model (FM) and describes what dataset the FM was finetuned on
- See the living document at: https://docs.google.com/document/d/1u3XG726xfTxRoa6yQ50YDm-liI-m7ASs0klblrwJOk4/edit
- Also diagrams from last week: https://docs.google.com/presentation/d/1E7iAcHlrXizBUePUTzE5hDD3rA_J-fm7VDp2kW1joVE/edit#slide=id.g2726d5e68b4_0_9
- Prompt-related properties and relationships