Solutions to the three programming exercises. Each folder has its own README with setup, commands, and results.
- hw3_q1_rag - a simple RAG system from scratch (Ollama embeddings + local LLM)
- hw3_q2_retrieval_embedding - a tiny retrieval embedding model trained with a contrastive hinge loss
- hw3_q3_dpo_objective - the DPO objective implemented on toy preference data