PromptTemplate.input_types is ignored on validation

Checked other resources

[X] I added a very descriptive title to this issue.
[X] I searched the LangChain documentation with the integrated search.
[X] I used the GitHub search to find a similar question and didn't find it.
[X] I am sure that this is a bug in LangChain rather than my code.
[X] The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

        document_prompt = PromptTemplate(
            input_variables=["page_content", "metadata"],
            input_types={
                "page_content": str,
                "metadata": dict[str, Any],
            },
            output_parser=None,
            partial_variables={},
            template="{metadata['source']}: {page_content}",
            template_format="f-string",
            validate_template=True
        )

Error Message and Stack Trace (if applicable)

  File "/home/fules/src/ChatPDF/streamlitui.py", line 90, in <module>
    main()
  File "/home/fules/src/ChatPDF/streamlitui.py", line 51, in main
    st.session_state["pdfquery"] = PDFQuery(st.session_state["OPENAI_API_KEY"])
  File "/home/fules/src/ChatPDF/pdfquery.py", line 32, in __init__
    document_prompt = PromptTemplate(
  File "/home/fules/src/ChatPDF/_venv/lib/python3.10/site-packages/pydantic/v1/main.py", line 341, in __init__
    raise validation_error
pydantic.v1.error_wrappers.ValidationError: 1 validation error for PromptTemplate
__root__
  string indices must be integers (type=type_error)

Description

I'm trying to create a document formatting template that uses not only the content of the documents but their metadata as well
As I explicitly specify that the metadata member is a dict, I expect that the validation logic honors that information
I've experienced that all input variables are treated as strs, regardless of input_types

At this point input_types is not passed on to check_valid_template, so that type information is lost beyond this point, and therefore the validator couldn't consider the type even if it tried to.

At this point the validator validate_input_variables tries to resolve the template by assigning the string "foo" to all input variables, and this is where the exception is raised.

The documentation of PromptTemplate.input_types states that

A dictionary of the types of the variables the prompt template expects. If not provided, all variables are assumed to be strings.

If this behaviour (input_types is ignored and all variables are always assumed to be strings) is the intended one, then it might be good to reflect this in the documentation too.

System Info

$ pip freeze | grep langchain
langchain==0.2.2
langchain-community==0.2.3
langchain-core==0.2.4
langchain-openai==0.1.8
langchain-text-splitters==0.2.1

$ uname -a
Linux Lya 5.15.146.1-microsoft-standard-WSL2 #1 SMP Thu Jan 11 04:09:03 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

$ python --version
Python 3.10.6

langchain-ai / langchain