Name 3 vector DB technologies
What is Milvus, PineCone, Chroma, LanceDB, Reddis, Weaviate
Large language models are based on what ML Architecure
What is the transformer architecture
Phoenix can be views in either ___ or ____.
What is browser or notebook
The first issue we see, and often the easiest to uncover is ___.
What is bad responses
Name two LLM Orchestration tools.
What is llamaidex and langchain
Name 2 Types of unstructured data that can be stored in a Vector DB
What are tokens, documents, text, audio, images, etc...
LLMs tend to run on what type of chip for inferences and training?
What is a GPU
What model type is not supported by Phoenix?
What is Ranking
What are three issues that might come up in a search and retreival system?
What is poor retrieval, bad responses, hallucinations, missing context, bad prompts
What's the difference between a chain and an agent?
What is chains have a sequence of actions that is hardcoded and agents use a language model as a reasoning engine to determine which actions to take and in which order.
Splitting a document into smaller ones in a hierarchical and iterative manner until certain criteria (e.g. number of tokens) are reached, is what type of chunking strategy.
What is recursive chunking
Name at least 2 stages in the creation or deveopment of LLMs model. To be more specifc, how do we create base LLM models and how do we improve upon them
what is Pretraining, Fine Tuning, RL, HF, RLHF
collection of documents used in RAG usecases
what is a corpus dataset
What are four steps that could be taken to improve a search & retrieval system?
What is (1)add docs to knowledge base, (2)change chunking strategy, (3)retreive additional context, and (4)rank context prior to repsponding?
This capability enables chatbots to maintain coherence across multiple queries, allowing a chat-like manner of interaction with users.
What is conversational memory
What retreival strategy is used in which you retrieve the top most relevant chunks, query each chunk with the prompt, and then ask the LLM to combine all the responses.
What is map-reduce
Name 3 LLM model names (hint: animal plant names are common)
What is Vicuna, Alpaca, Bison (TextBison), Falcon, Firefly, Palm, GPT etc...
What app allows users to run Phoenix in a databricks or locally from a remote Jupyter instance?
What is using ngrok
True or False: Query density allows you to identify small concept gaps in your knowledge base.
What is false.
This is the default state of a Large Language Model (LLM), where each query is processed without considering past interactions.
What is stateless
What are 3 ways to determine similiarty when searching for emebddings?
What is euclidean dist, cosine similarity, dot product
In the context of neural sequence-to-sequence models, decoding strategies help determine the next token (e.g., word or subword) to be generated, based on the model's predictions. Name 1 decoding strategy
what is greedy decoding, topK, nucleus sampling, beam search
What is required that you don't need a new ngrok tunnel every time?
What is Phoenix default port.
Ranking metrics allow you to directly measuring effectiveness of retrieval, however they require ____ & ____.
What is ground truth and additional LLM calls.
What are the four document chains offered by langchain?
what is stuff, map-reduce, refine, map re-rank