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