added initial version for summarize chains
parent
1eacc74221
commit
1e0a70b03e
@ -0,0 +1,9 @@
|
||||
# Summarize
|
||||
This is a type of chain to distill large amounts of information into a small amount.
|
||||
|
||||
There are three types of summarize chains:
|
||||
1. Stuff: This is a simple chain to stuff all the text from each document into one propmt. This is limited by token window so this approach only works for smaller amounts of data.
|
||||
2. Map Reduce: This maps a summarize prompt onto each data chunk and then combines all the outputs to finally reduce using a summarization prompt.
|
||||
3. Refine: This iteratively passes in each chunk of data and update a continously update an evolving summary to be more accurate based on the new chunk of data given.
|
||||
|
||||
## Usage
|
@ -0,0 +1,2 @@
|
||||
# Map Reduce
|
||||
This maps a summarize prompt onto each data chunk and then combines all the outputs to finally reduce using a summarization prompt.
|
@ -0,0 +1,10 @@
|
||||
input_variables: [text]
|
||||
output_parser: null
|
||||
template: 'Write a concise summary of the following:
|
||||
|
||||
|
||||
{text}
|
||||
|
||||
|
||||
CONCISE SUMMARY:'
|
||||
template_format: f-string
|
@ -0,0 +1,2 @@
|
||||
# Refine
|
||||
This iteratively passes in each chunk of data and update a continously update an evolving summary to be more accurate based on the new chunk of data given.
|
@ -0,0 +1,13 @@
|
||||
# Prompt config for a chain?
|
||||
|
||||
input_variables: ['text']
|
||||
output_parser: null
|
||||
# what primitives do we allow?
|
||||
template: [
|
||||
map:
|
||||
input: 'text'
|
||||
prompt: 'prompt.yaml',
|
||||
reduce:
|
||||
prompt: 'promptSummarize.yaml'
|
||||
]
|
||||
template_format: f-string
|
@ -0,0 +1,13 @@
|
||||
input_variables: [existing_answer, text]
|
||||
output_parser: null
|
||||
template: '
|
||||
Your job is to produce a final summary\n
|
||||
We have provided an existing summary up to a certain point: {existing_answer}\n
|
||||
We have the opportunity to refine the existing summary
|
||||
(only if needed) with some more context below.\n
|
||||
------------\n
|
||||
{text}\n
|
||||
------------\n
|
||||
Given the new context, refine the original summary
|
||||
If the context isn't useful, return the original summary.'
|
||||
template_format: f-string
|
@ -0,0 +1,10 @@
|
||||
input_variables: [text]
|
||||
output_parser: null
|
||||
template: 'Write a concise summary of the following:
|
||||
|
||||
|
||||
{text}
|
||||
|
||||
|
||||
CONCISE SUMMARY:'
|
||||
template_format: f-string
|
@ -0,0 +1,2 @@
|
||||
# Stuff
|
||||
This is a simple chain to stuff all the text from each document into one propmt. This is limited by token window so this approach only works for smaller amounts of data.
|
@ -0,0 +1,10 @@
|
||||
input_variables: [text]
|
||||
output_parser: null
|
||||
template: 'Write a concise summary of the following:
|
||||
|
||||
|
||||
{text}
|
||||
|
||||
|
||||
CONCISE SUMMARY:'
|
||||
template_format: f-string
|
Loading…
Reference in New Issue