LangChain summarizing/truncating instead of returning verbatim

[uhwan888]

Example Code

from langchain.prompts import PromptTemplate

verbatim_prompt = PromptTemplate(
    input_variables=["context", "question"],
    template="""
Below is the raw text:
----------------
{context}
----------------

Question: {question}

Please return the exact matching text from the section above. 
Do not summarize, paraphrase, or alter the text in any way. 
Return the full excerpt verbatim.
"""
)

    def get_conversational_chain(self):
        model = ChatGoogleGenerativeAI(model="gemini-1.5-pro", temperature=0.0)
        chain = load_qa_chain(
            llm=model,
            chain_type="stuff",
            prompt=verbatim_prompt,
            document_variable_name="context",
            verbose=True,
        )
        return chain

Description

What I’m doing:

1. I upload one or more PDFs, split them into 10000-token chunks, and build a FAISS index of those chunks.
2. I retrieve the top-k chunks with vector_store.similarity_search(…).
3. I feed them into LangChain’s “stuff” QA chain with a verbatim prompt template.

The problem:
Instead of spitting back the full chunk I asked for, Gemini still summarizes or cuts off the text midway. I need the entire verbatim excerpt, but every response is truncated (regardless of how large I set my chunks).

Question:
What am I missing? Is there a chain configuration, prompt format, or Gemini parameter that forces a full-text return instead of a summary/truncation? Or do I need to use a different chain type (e.g. map-reduce or refine) or a different model setting to get unabridged verbatim output?

Any pointers or sample code would be hugely appreciated...thanks!

System Info

Environment:

• Python: 3.11
• Flask: 3.0.3
• LangChain: 0.2.6
• langchain-community: 0.2.6
• langchain-core: 0.2.10
• langchain-text-splitters: 0.2.2
• FAISS (faiss-cpu): 1.8.0.post1
• PyPDF2: 3.0.1
• google-generativeai: 0.8.4
• google-ai-generativelanguage: 0.6.15
• langchain-google-genai: (whatever version you have)