1 resource in Models.
Fine-tuned on 1.4M Odia instruction pairs. Outperforms base Llama-3 on Odia benchmarks by 38%.
