The Nadi_Indic466K_Instruct dataset is the world's first coding dataset with 18 Indian language support, 466k rows and 142 Million total tokens. This dataset can be used by developers to build Indian coding language models (LLMs) for various programming languages.
Q-LoRA based SFT/PPO/DPO fine-tuning can be done on the dataset in LLAMA-2 or Mistral or any opens-soure LLM for text generation.
The dataset was carefully curated such that the coding part… See the full description on the dataset page:
https://huggingface.co/datasets/convaiinnovations/Nadi_Indic466k_Instruct.
