Bangla LLM Applications
Bengali NLP and LLM fine-tuning toolkit featuring curated datasets, instruction-tuning notebooks, and tokenization evaluation benchmarks.
Overview
Comprehensive resource repository and code library for fine-tuning open-source LLMs (Llama 3, Mistral, Gemma) on low-resource Bengali linguistic corpora.
Architecture & Design
QLoRA parameter-efficient fine-tuning workflows on Hugging Face PEFT with custom tokenizers and translation alignment pipelines.
Key Capabilities
- Instruction-tuning datasets prepared in Bengali dialect and formal register
- QLoRA fine-tuning recipes for 8B and 7B parameter foundation models
- Bengali evaluation benchmarks for grammar, summarization, and sentiment