Thanks for your great work! Will you release 340M params/15B tokens DeltaNet models (including both the pure DeltaNet and hybrid architecture)? Since this model is trained on 15B tokens, it will be more convenient for the academia to reproduce the results.
