Following along with documentation and existing code to try out BAML generation based on custom documents/strings. I have a working FastAPI instance now. BAML generation and execution works on the Playground as expected. Went ahead and tested the extraction and generation POST requests on the Docs page - Generation works, but for some reason extraction does not. Any pointers?

py
...
async def exec_baml(
    stream: bool,
    file: UploadFile = File(None),
    content: str = Form(None),
    url: str = Form(None),
    baml_code: str = Form(...),
    return_type: str = Form(...)
):
    final_content = await read_content(file, content, url)
    tb = TypeBuilder()
    try:
        tb.add_baml(f"""
        {baml_code}

        dynamic class Response {{
            data {return_type}
        }}
        """)
    except BamlError as e:
        raise HTTPException(status_code=400, detail={
            "error": "BamlError",
            "message": str(e),
        })
    if stream:
        stream = b.stream.ExecuteBAML(final_content, { "tb": tb })
        return handle_stream(stream, lambda x: x.data)
    else:
        response = await b.ExecuteBAML(final_content, { "tb": tb })
        return response.data
...

https://cdn.discordapp.com/attachments/1449701403693023293/1449701404406059018/image.png?ex=693fdb24&is=693e89a4&hm=c3ae1e6c8404533a6bf38197615153f2a09c79cee24a29aa5b615062ccd9efa4&

Hey @GR™, the problem with your screenshot (Syntax Error in type builder) happens because you set

baml_code=string

and then you call tb.add_baml with

string

. You end up with something like this:

string

dynamic class Response {
    data string
}

And that's invalid syntax. You have to either skip the {baml_code} part if it''s just a type, or make it a class. One of these:

dynamic class Response {
    data string
}

Or

class WantThisData {
    name string
    skills string[]
}

dynamic class Response {
    data WantThisData
}

In the second case

baml_code

has to be the entire class definition and

return_type

has to be the class name.

Tried poppler, didn't work

python-poppler

The example I'll take a look tomorrow

No dice https://cdn.discordapp.com/attachments/1449701403693023293/1449983890373607556/image.png?ex=6940e23a&is=693f90ba&hm=a8a7133b2fb5676499643f1014bc9693d798dd5b053dcb71b8516ca65e27399a&

It's missing pkg-config and some other stuff

Build worked, but the same PDF2Image error pops up when I run a test gen_call

Try removing the .venv and doing the setup again installing all deps

Same error

Wait what are you trying to do with pdfs

The example processes a PDF/Image/Text String to generate BAML schema and then extract data in said BAML schema

Yeah the example was written before we added the

pdf

type I think

Any luck with this?

Sorry, I haven't tried yet, I'll do tomorrow (on Fridays we fix bugs/docs etc for next release)

Hey there - Sorry about the giant delay. Job change and move had me pre-occupied. I've restarted learning BAML.

There's no restriction on where

baml-src

can be, could be outside backend if you're sharing Baml types between frontend and backend, or you wanna generate those types for the frontend as well. In this particular example of

extract-anything

I don't think there's a reason it needs to be outside of

backend

. Steps to update, yes you have to manually update pyproject and generator, the extension only downloads the correct LSP version for you but it doesn't auto update generators and pyproject versions

Just spotted this for some reason - No idea why Discord didn't notify as unread by default. Noted on all counts. I think all that needs answering at this point is the PII question from earlier.

What's the PII question?

^ this one

Yes I think that's OpenAI filter. I am not sure if there are options to bypass that one