Datasets:
Use in continual pretraining mix of Apache 2.0 model
#2
by
droussis
- opened
Hi and congrats for the great work!
I'm a bit confused about NVIDIA Open Data License Agreement.
Specifically, I am interested in using parts of the dataset (those not inheriting other model licenses) in the training mix for continual pretraining of an Apache 2.0 model.
The model will be evaluated on several benchmarks and this particular part of the NVIDIA Data Agreement for Model Training confused me:
2.2 Restrictions. Company may not:
...
2.2.5 Disclose any evaluation, test or benchmarking results regarding the Datasets without NVIDIA’s prior written consent documented as an amendment to this Agreement [...]
So, is it possible to use the dataset in the pretraining mix of a model that will have an Apache 2.0 license? Can this model be evaluated without prior consent if it doesn't feature any specific ablations for NVIDIA's datasets under this agreement?
Thank you very much in advance!