Langchain_keywords

Langchain_keywords

   - LangChain inbuilt cache
    - OpenAI chatgpt 3.5 vs 4
    - Streaming LLM with attention sink
    - Prompt templates, few shot templates
    - Text model, chat model, text embedding
    - Prompts
    - SemanticsSimilarityExampleSelector
    - FewShotPromptTemplate
    - OutputParsers
    - DocumentLoaders
    - Text Splitters
    - Retrievers
    - Vector embeddings
    - ChatMessageHistory
    - Chains 
        - SimpleSequentialChain
        - load_summarize_chain - go over chunks and summarize - map reduce
    - Agents, Tool, Toolkit
    - cosine similarity 
    - Text embeddings generators - OpenAI, Word2Vec, GloVe
    - model.most_similar(postive=, negative=)
    - Vector embeddings of 
        - Word/Text
        - Graph
        - Image
        - Documents
        - Sentences
        - Graph embeddings
    - Some uses
        - Recommendation System
        - Anomaly detection
        - Transfer leaning
        - Visualisation
        - Information retreival
        - Audo + Speech processing
        - Facial Recognition
    - vectorIndex = VectorStoreIndexWrapper(vectorstore=myCassandraVStore)
    - SentenceTransformer SBERT.net
        - Semantic Search
    - HuggingFace
    - Similarity:
        - Dot product
        - cosine-similarity
        - Euclidean
    - Bitext Mining
        - finding translated sentence pairs in two languages
    - Image and Text models
        - Both images and text in the same vector space
        - zero-shot image classification
    - OpenAI text embedding length 1536
        - Sequence Lengtg = Token, might not qual to number of lenghts
        - Byte pair encoding
    - fast approximate nearest neighbor search
    - Embedding benchmark: https://github.com/embeddings-benchmark/mteb
    - Hugging face - python sdk
    - ONNX Runtime - inference engine c++
    - quantized model - reduce the size of the model
    - Portkey for llm observability. Langchain agent
        - Passed as header in ChatOpenAI object