github/langchain
1
0
Fork 0
mirror of https://github.com/hwchase17/langchain.git synced 2025-09-02 11:39:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

add concept of prompt collection (#1507)

Browse Source
This commit is contained in:
Harrison Chase
2023-03-08 08:31:29 -08:00
committed by GitHub
parent 97e3666e0d
commit c4a557bdd4
8 changed files with 204 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
docs/modules/chains/key_concepts.md
View File

@@ -9,3 +9,12 @@ This is a specific type of chain where multiple other chains are run in sequence
to the next. A subtype of this type of chain is the [`SimpleSequentialChain`](./generic/sequential_chains.html#simplesequentialchain), where all subchains have only one input and one output,
and the output of one is therefore used as sole input to the next chain.
## Prompt Selectors
One thing that we've noticed is that the best prompt to use is really dependent on the model you use.
Some prompts work really good with some models, but not great with others.
One of our goals is provide good chains that "just work" out of the box.
A big part of chains like that is having prompts that "just work".
So rather than having a default prompt for chains, we are moving towards a paradigm where if a prompt is not explicitly
provided we select one with a PromptSelector. This class takes in the model passed in, and returns a default prompt.
The inner workings of the PromptSelector can look at any aspect of the model - LLM vs ChatModel, OpenAI vs Cohere, GPT3 vs GPT4, etc.
Due to this being a newer feature, this may not be implemented for all chains, but this is the direction we are moving.