Customize the prompts with refine model

[ilyasbendev]

Hello everyone,
I need to customize my summarization using refine model.
I'm working with load_summarize_chain and also i'm working with large documents, so i had to split my documents.

This is a part of code that i wrote :

llm = ChatOpenAI(model_name="gpt-3.5-turbo" , openai_api_key="example_api_key")
#texts is the entire text and t is the text splitted
docs = [Document(page_content=t) for t in texts]
refine_prompt_template = "Write a summary of the following text : {t} please summarize in appromximately 20 words"
refine_prompt = PromptTemplate(template=refine_prompt_template, input_variables=[text])
refine_chain = load_summarize_chain(vertex_llm_text,chain_type="refine", refine_prompt=refine_prompt)

I want the entire text to be summarized in 20 words (how can i combine this prompt as many as number of texts ?(number of t))

[dosubot[bot]]

🤖

Hello,

To summarize the entire text in 20 words by combining the prompt as many times as the number of split texts, you can modify your code as follows:

from langchain import Document, PromptTemplate
from langchain.openai import ChatOpenAI
from langchain.chains.summarize import load_summarize_chain

llm = ChatOpenAI(model_name="gpt-3.5-turbo" , openai_api_key="example_api_key")

# Assuming 'texts' is your list of split texts
docs = [Document(page_content=t) for t in texts]

# Initialize an empty string to store the final summary
final_summary = ""

# Loop through each document
for doc in docs:
    refine_prompt_template = f"Write a summary of the following text : {doc.page_content} please summarize in approximately 20 words"
    refine_prompt = PromptTemplate(template=refine_prompt_template, input_variables=[doc.page_content])
    refine_chain = load_summarize_chain(llm, chain_type="refine", refine_prompt=refine_prompt)
    
    # Generate the summary for the document
    summary = refine_chain.run(doc)
    
    # Add the summary to the final summary
    final_summary += summary

# Now 'final_summary' should contain the combined summary of all documents

This code will loop through each document in your list of split texts, generate a summary for it using the refine_chain, and add the summary to the final_summary. At the end of the loop, final_summary should contain the combined summary of all documents.

Please note that the load_summarize_chain function requires a BaseLanguageModel instance as the first argument, a chain_type as the second argument, and a refine_prompt as the third argument. The refine_prompt should be an instance of PromptTemplate, which requires a template string and a list of input variables. In your case, the template string is the prompt you want to use for summarization, and the input variable is the text you want to summarize.

I hope this helps! Let me know if you have any other questions.

Sources

• libs/langchain/langchain/chains/summarize/refine_prompts.py
• libs/langchain/langchain/chains/summarize/init.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[ilyasbendev]

Hi,
This code will not work, I can see two problems here.
1st : If the text is splitted to 5, the final summary will contain 100 word (5*20), my goal is to have 20 words in the summary
2nd : The

 final_summary 

will not contain comprehensive summary because its just assembling the small summaries together

[lee1221ee]

First summarize a large amount of text into few paragraph.

Use the following code：

chain = load_summarize_chain(llm, chain_type="refine")
chain.run(split_docs)

Then, use a simple custom summary prompt or chain_type="stuff" (load_summarize_chain). To limit 20 words in this step.