abdulazeem / edx_databricks_llm

LLM

Vector Stores

1. Specialized full-fledged databases for unstructured data.

2. Popular vector database comparisons

3. How to imporove retrieval performance?

   1. Embedding Model Selection
      1. Do i have the right embedding model for my data?
      2. Do my embeddings capture BOTH my documents and queries?
   2. Document Storage Strategy
      1. Should i store the whole document as one? Or split up into chunks?

   

4. Other Resources:

   1. dfd
   2. Do you actually need a vector database? by Ethan Rosenthal
   3. https://learning.edx.org/course/course-v1:Databricks+LLM101x+2T2023/block-v1:Databricks+LLM101x+2T2023+type@sequential+block@a7512e678c8a4fd98ae3635977fb55bb/block-v1:Databricks+LLM101x+2T2023+type@vertical+block@b475e8c705f448818785a4d5dc2952e4

Fine-Tuning LLM's

1. Typical Flavors or fine tuned versions:

   1. Base - I know what word comes next
   2. Chat - I know how to engage in conversation
   3. Instruct - I know how to respond to instructions

2. Multiple Sizes:

   1. Small
   2. Base
   3. Large

3. Which one to use? Need to balance below three:

   1. Accuracy
   2. Speed
   3. Task Specific Performance in a3.5

Fine Tuning - Few Shot Learning

Instruction Following LLM

Fine tune existing model

1. Fine tuning using Dollyv2

Evaluation of LLM Models

1. Need to look at task specific evaluation.

Harrison Chase Overview of LangChain

Natural language answer vs Natural language groud truch and compare them for semantic equivalence.