MasterAI-EAM/Darwin

Pretrains and fine-tunes LLaMA on 100,000+ instruction-following examples synthesized from 6M scientific papers and 16 FAIR datasets via a custom Scientific Instruction Generator (SIG). Employs QA + multi-task fine-tuning strategies that outperform specialized ML methods and GPT-3.5 in chemistry tasks, with demonstrated state-of-the-art results on MatBench bandgap and metallic classification benchmarks. Integrates with LangChain for complex scientific workflows and supports inference on 10GB+ GPU memory while enabling full fine-tuning via distributed training.

247 stars. No commits in the last 6 months.