How do I chain multiple prompts?

[BillRaymond]

I am the host of a podcast. After a recording, I generate a transcript in markdown. I then use that markdown to develop the following:

1. A summary of the podcast for the website
2. A social media post for my guest
3. A social media post for me

I can do all this, but my code is inefficient because I make multiple calls, uploading the document each time with a new prompt template. Langchain should be able to do that for me.

I stripped out the detailed question for brevity, but as you can see, I created a prompt_template for the summary like this:

prompt_template = """Write a summary of the following podcast text: 
{text}
SUMMARY :"""
PROMPT = PromptTemplate(template=prompt_template, input_variables=["text"])
chain = load_summarize_chain(llm, chain_type="stuff", prompt=PROMPT)
summary=chain.run(docs)

print(summary)

Then I create yet another one for the guest, like this:

podcast_template = """Write a summary of the following podcast text as if you are the guest(s) posting on social media.
{text}

SUMMARY :"""
PROMPT = PromptTemplate(template=podcast_template, input_variables=["text"])
chain = load_summarize_chain(llm, chain_type="stuff", prompt=PROMPT)
podcast_summary=chain.run(docs)

# Label the podcast summary and add some space at the end
podcast_summary = "\n\nPODCAST SUMMARY:\n\n" + podcast_summary
podcast_summary += "\n\n"

print(podcast_summary)

With my limited experience in Langchain, I think my code is doing the following:

1. Upload the document
2. Ask a prompt
3. Upload the document again
4. Ask a prompt again
5. And keep doing that as many times as needed

What I want is:

1. Upload the document
2. Ask as many prompts as needed to get the summaries I am looking for

Is it possible to do that, or is what I am using the most efficient?

FYI: I searched the questions here and tried to find similar answers, but I feel like searching for "chain," "link," or "multiple" provides too many results that are not necessarily the answer I am looking for. I'm sorry if this has been asked and answered.

Thanks!

[dosubot[bot]]

Answer generated by a 🤖

Answer

I understand that you're looking to optimize your code by chaining multiple prompts in LangChain. You want to upload your podcast transcript once and then generate a summary, a social media post for your guest, and a social media post for yourself, without having to upload the document multiple times.

You're on the right track. LangChain does indeed allow you to chain multiple prompts using the SequentialDocumentsChain class. This class lets you execute multiple prompts in a sequence, each with a different prompt template.

Here's how you can modify your code to achieve this:

from langchain.chains import SequentialDocumentsChain, LLMChain
from langchain.llms import OpenAI
from langchain.prompts import PromptTemplate

# Define your prompt templates
summary_template = """Write a summary of the following podcast text: 
{text}
SUMMARY :"""
guest_template = """Write a summary of the following podcast text as if you are the guest(s) posting on social media.
{text}
SUMMARY :"""
host_template = """Write a summary of the following podcast text as if you are the host posting on social media.
{text}
SUMMARY :"""

# Create PromptTemplates
SUMMARY_PROMPT = PromptTemplate(template=summary_template, input_variables=["text"])
GUEST_PROMPT = PromptTemplate(template=guest_template, input_variables=["text"])
HOST_PROMPT = PromptTemplate(template=host_template, input_variables=["text"])

# LLM to use in each stage 
llm = OpenAI()

# Create LLMChains for each stage
summary_llm_chain = LLMChain(llm=llm, prompt=SUMMARY_PROMPT)
guest_llm_chain = LLMChain(llm=llm, prompt=GUEST_PROMPT)
host_llm_chain = LLMChain(llm=llm, prompt=HOST_PROMPT)

# Create SequentialDocumentsChain
chain = SequentialDocumentsChain(
    chains=[summary_llm_chain, guest_llm_chain, host_llm_chain]
)

# Run the chain
results = chain.run(docs)

# Print the results
print(results)

In this code, we're using SequentialDocumentsChain to chain the three prompts. The chains parameter is a list of the chains to be executed in sequence. We then call the run method on the chain with the document as the argument, and print the results.

This is just one potential solution to your problem. There might be other ways to achieve the same result depending on your specific needs and the complexity of your project.

If you find this solution helpful and think it might be useful to others, I encourage you to make a pull request to update the LangChain documentation. This way, other users facing the same issue can benefit from your experience.

Sources

• docs/snippets/modules/chains/popular/summarize.mdx
• docs/extras/modules/model_io/prompts/prompt_templates/prompts_pipelining.ipynb
• docs/snippets/modules/chains/foundational/sequential_chains.mdx

---

This response is meant to be useful, save you time, and share context. 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.

[BillRaymond]

Trying the example, I get the following error message:

ImportError: cannot import name 'SequentialDocumentsChain' from 'langchain.chains' (/usr/local/lib/python3.10/dist-packages/langchain/chains/__init__.py)

I believe I have everything installed, but to check I ran the following lines and reset my virtual environment, but still receive the error.

pip install langchain[all]
pip install --upgrade langchain

I went to the docs and I see no mention of SequentialDocumentsChain. I do, however, see SimpleSequentialChain and SequentialChain.

[dosubot[bot]]

Answer generated by a 🤖

Answer

It seems like there might be a misunderstanding in the class name. The SequentialDocumentsChain class does not exist in the LangChain framework. As you've found in the documentation, the available classes for chaining are SimpleSequentialChain and SequentialChain.

Based on your requirements, you want to upload a podcast transcript once and then generate a summary, a social media post for your guest, and a social media post for yourself. This seems like a complex chain that involves multiple inputs and multiple final outputs. Therefore, the SequentialChain class would be more suitable for your needs.

Here's a basic example of how you might set up your chain:

from langchain.chains import SequentialChain
from langchain.memory import SimpleMemory

# Define your chains here
summary_chain = ...
guest_post_chain = ...
self_post_chain = ...

overall_chain = SequentialChain(
    memory=SimpleMemory(memories={"transcript": "Your podcast transcript here"}),
    chains=[summary_chain, guest_post_chain, self_post_chain],
    input_variables=["transcript"],
    output_variables=["summary", "guest_post", "self_post"],
    verbose=True)

results = overall_chain({"transcript": "Your podcast transcript here"})

In this example, summary_chain, guest_post_chain, and self_post_chain are placeholders for the chains you would define to generate a summary, a guest post, and a self post, respectively. You would replace "Your podcast transcript here" with your actual podcast transcript.

Please replace the chains and the transcript with your actual data and see if this resolves your issue. If you encounter any further problems, feel free to ask.

---

This response is meant to be useful, save you time, and share context. 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.

[BillRaymond]

Could you provide an example where I add the transcript as a file, like transcript.md, and then show an example of what a summary_chain, guest_post_chain, and self_post_chain looks like?

[dosubot[bot]]

Answer generated by a 🤖

Answer

Hello,

To define the summary_chain, guest_post_chain, and self_post_chain using the SequentialChain class in the LangChain Python framework, you would first need to define the individual chains that make up each of these chains.

The SequentialChain class is used to create a chain of chains, where the output of one chain is used as the input for the next chain. The SequentialChain class takes a list of chains and a list of input variables as arguments.

Here is an example of how you might define these chains:

# Define the individual chains
summary_chain_1 = FakeChain(input_variables=["transcript"], output_variables=["summary"])
summary_chain_2 = FakeChain(input_variables=["summary"], output_variables=["final_summary"])

guest_post_chain_1 = FakeChain(input_variables=["transcript"], output_variables=["guest_post"])
guest_post_chain_2 = FakeChain(input_variables=["guest_post"], output_variables=["final_guest_post"])

self_post_chain_1 = FakeChain(input_variables=["transcript"], output_variables=["self_post"])
self_post_chain_2 = FakeChain(input_variables=["self_post"], output_variables=["final_self_post"])

# Define the SequentialChains
summary_chain = SequentialChain(chains=[summary_chain_1, summary_chain_2], input_variables=["transcript"])
guest_post_chain = SequentialChain(chains=[guest_post_chain_1, guest_post_chain_2], input_variables=["transcript"])
self_post_chain = SequentialChain(chains=[self_post_chain_1, self_post_chain_2], input_variables=["transcript"])

In this example, each chain is made up of two individual chains. The first chain takes the transcript as input and generates a summary, guest post, or self post. The second chain takes this output and further processes it to generate the final summary, guest post, or self post.

To add a markdown file named transcript.md as an input for an application that generates a summary, a guest post, and a self post from a podcast transcript, you would need to read the contents of the markdown file and pass it as an argument to the SequentialChain:

# Read the contents of the markdown file
with open('transcript.md', 'r') as file:
    transcript = file.read()

# Pass the transcript as an argument to the SequentialChain
summary_output = summary_chain({"transcript": transcript})
guest_post_output = guest_post_chain({"transcript": transcript})
self_post_output = self_post_chain({"transcript": transcript})

Please note that this is a simplified example and the actual implementation may vary depending on the specifics of your application and the LangChain framework.

For more information, you can refer to the LangChain documentation and the unit tests for the SequentialChain class:

• libs/langchain/tests/unit_tests/chains/test_sequential.py
• libs/langchain/langchain/chains/sequential.py
• docs/snippets/modules/chains/foundational/sequential_chains.mdx

I hope this helps! If you have any further questions, feel free to ask.

Best,
Dosu

Sources

• libs/langchain/tests/unit_tests/chains/test_sequential.py
• libs/langchain/langchain/chains/sequential.py
• docs/snippets/modules/chains/foundational/sequential_chains.mdx

---

This response is meant to be useful, save you time, and share context. 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.