As we closed phase 1, and published draft v.0.0.6, we can start moving into phase 2. That is, we need to review the systems we analyzed, find the availability of the required components and make a list of the legal documents they use. For example, the template would be a table like the one below, at least for the four systems we looked at (Pythia, Llama2, OpenCV, BLOOM) but we can add more if necessary to have a wider array of legal documents.
The idea is to start testing how the draft Definition works in practice in order to complete it by the end of May.
Model name
| Required component | Link to resource | Legal framework |
|---|---|---|
| Data pre-processing code | URL | OSI-approved license |
| Training, validation and testing code | URL | … |
| Inference code | URL | … |
| Supporting libraries and tools | URL | … |
| Model architecture | URL | … |
| Model parameters | URL | ??? |
| Data transparency requirements | URL | ??? |
As a reminder, this is the process with highlighted the phase we’re starting:
