Sample Discussion — This is example content to show how discussions work.Start a real discussion to engage with the community!
Best practices for fine-tuning LLMs for Indian languages?
NLP Engineer at Stealth · Mar 13, 2026, 6:44 PM
Working on a customer support AI that needs to handle Hindi, Tamil, and Telugu. Current approach: - Base model: Llama 3 70B - Fine-tuning on translated ShareGPT + custom Indian language data - Using LoRA with r=16 Issues I'm facing: 1. Code-mixing (Hinglish) is tough 2. Limited evaluation benchmarks for Indian languages 3. Transliteration (Hindi in Roman script) Has anyone tackled this? Would love to connect with others working on Indic NLP. Also - is Sarvam's base model worth checking out for this use case?
0 Replies
No replies yet. Be the first to respond!
This is a sample discussion. Start a real discussion to engage with the community!
Start Discussion