Seeking Recommendations for Efficient Models for Conversational Agents in Regional Indic Languages

I am in the process of developing conversational agents tailored for regional Indic languages, with an emphasis on enhancing natural language understanding in low-resource linguistic contexts. The primary obstacle has been identifying a Large Language Model (LLM) that supports my target languages and is also viable for chatbot creation. While GPT might seem like an obvious recommendation, my project is constrained by a lack of funding for commercial GPT models, and unfortunately, the performance of the open-source GPT-2 model falls short of our needs. I looked into BLOOM but could not find enough projects where it is used in building a chatbot. Is there any model that I can use for this purpose?

I also am looking into the translation-based approach, utilizing Indic BART for translating prompts from the target language into English, processing these through a conversational model, and then translating responses back to the original language. Although this method introduces added complexity and latency, it aligns with my goal of advancing language understanding within my target demographic.

I am currently considering LLAMA 2.0 for the conversational aspect of this project but am open to suggestions. Given the resource-intensive nature of models like LLAMA 2.0, I am actively seeking advice on alternative models that are less demanding in terms of computational resources yet effective for my use case.