The original basemodel is
Qwen/Qwen1.5-7B.
This pretraining has then been continued on Odia language data by the
OdiaGenAI organization.
Finally, the model has been instruct-finetuned for 6 epochs on 5 Odia-language instruct datasets translated or produced by the OdiaGenAI organization,
and 1 English instruction dataset.
The instruction tuning stage is documented in detail in
this tutorial.
