Intro to LLMS
- Large Language Model: At its core it predicts the next token(word) and it does till it reaches
- limits
<EOS>: End of sequence
- Large Language models predicts the next token from a distribution.
- distribution is all about token and its probability.
- Choosing a token can be influenced by temperature.
-
The quality of output depends on the inputs passed (Prompt)
-
Refer Here for visualization of how an llm generates tokens
-
LLMs are Machine Learning Models (Deep learning Neural Network -> Transformer)
-
To interact with Models we have
- SDKs offered by model provider
- Open AI SDK
- Gemini
-
We need an approach to interact with llms in a similar way irrespective of model provider and if this approach also includes a way to
- deal with tools
- storing and retrieving information
- One answer to the above point is
- langchain
