Reproduce “AssameseLM-E2B: An Open-Weight Assamese Language Model for Efficient Consumer-Grade Deployment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pragyan Phukan
AssameseLM-E2B adapts a 2B Gemma-based model for Assamese with an expanded subword vocabulary, continued pretraining, and instruction tuning for retrieval-augmented generation (RAG), answering from supplied text and abstaining when it lacks an answer. On 337,084 UTF-8 bytes of FLORES-200 Assamese, it uses 35,107 tokens versus Llama-3.2-3B’s 166,218, with similar but higher BPB; its 96.7% RAG score comes from ten prompts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.