Datasets:
License:
[OSAID discussion] the Pangea suite of models: open multimodal data, open weights, LLM backbone initialized from Qwen
#3
by
yjernite
HF staff
- opened
The Pangea suite of models is a pretty interesting case.
Most of the components are a clear fit for the OSI OSAID:
- open multimodal data for pre-training and fine-tuning
- open-source training and evaluation code
- open model weights with an open source software license
The main part that is harder to fit within the OSAID is the use of a Qwen model for part of the initialization, since Qwen provides very limited information on its training data. Would love to hear other's perspectives on this!