sequential-decisions

Direct Diversity Optimization (DDO) is an offline post-training method that improves successful strategy coverage in LLM agents for sequential decision tasks, outperforming other methods in benchmarks like BabyAI, BabaIsAI, and WebShop.