技法
训练、检索与适配方法，从微调、LoRA 到 RAG 与量化。每一项都指出它针对的问题，以及在算力或质量上的代价。
35 technologies
- GraphRAG technique Retrieval over a knowledge graph built from the corpus.
- Beam Search technique Keeping several candidate sequences alive and extending the best.
- DPO technique Aligning a model directly on preference pairs, without training a reward model.
- Query Rewriting technique Reformulating a question before retrieval to improve what comes back.
- HyDE technique Embedding a generated answer instead of the question.
- Quantization technique Storing model weights at lower precision to cut memory and speed up inference.
- Self-RAG technique Model decides for itself when to retrieve and critiques what it retrieved.
- Diffusion technique Generates data by learning to reverse a gradual noising process.
- Distillation technique Training a small model to reproduce the behaviour of a larger one.
- LoRA technique Fine-tuning by training small low-rank adapter matrices instead of all weights.
- Tokenization technique Splitting text into the subword units a language model actually reads.