Implements and benchmarks optimal demonstration selection strategies for In-Context Learning (ICL) using LLMs. Covers IDS, RDES, Influence-based Selection, Se², and TopK+ConE across reasoning and classification tasks, analyzing the impact of example relevance, diversity, and ordering on model performance across multiple architectures.
python nlp benchmarking information-retrieval reinforcement-learning reproducible-research evaluation transformers beam-search semantic-search selection-algorithms meta-learning few-shot-learning sequence-modeling mahine-learning prompt-tuning in-context-learning llm prompt-engineering demonstration-selection
-
Updated
Jun 28, 2025 - Jupyter Notebook