DEV Community

Michael Anderson
Michael Anderson

Posted on

GENERATIVE AI SUMMARY

  1. Loss metric: Measures how wrong a model's predictions are. Lower loss is better.
  2. Cosine distance of 0: Indicates two embeddings are similar in direction.
  3. RAG (Retrieval Augmented Generation): Uses external information to improve text generation.
  4. String prompt templates: Can use any number of variables.
  5. Retrievers in LangChain: Retrieve relevant information from knowledge bases.
  6. Indexing in vector data: Maps vectors for faster searching.
  7. Accuracy: Measures correct predictions out of total predictions.
  8. Keyword-based search: Evaluates documents based on keyword presence and frequency.
  9. Soft prompting: When there is a need to add learnable parameters to a Large Language Model (LLM) without task-specific training.
  10. Greedy decoding: Selects the most probable word at each step in text generation.
  11. T-Few fine-tuning: Updates only a fraction of model weights.
  12. LangChain: Python library for building LLM applications.
  13. Prompt templates: Use Python's str.format syntax for templating.
  14. RAG Sequence model: Retrieves multiple relevant documents for each query.
  15. Temperature in decoding: Influences probability distribution over vocabulary.
  16. LLM in chatbot: Generates linguistic output.
  17. Chain interaction with memory: Before and after chain execution. 18. Challenge with diffusion models for text: Text is not categorical.
  18. Vector databases vs. relational databases: Based on distances and similarities.
  19. StreamlitChatMessageHistory: Stores messages in Streamlit session state, not persisted.
  20. Semantic relationships in vector databases: Crucial for LLM understanding and generation.
  21. Groundedness vs. Answer Relevance: Groundedness focuses on factual correctness, Answer Relevance on query relevance.
  22. Fine-tuning vs. PEFT: Fine-tuning trains the entire model, PEFT updates a small subset of parameters.
  23. Fine-tuning appropriateness: When LLM doesn't perform well and prompt engineering is insufficient.

Top comments (0)