Parser SPAMING the API with garbage

[compatible9]

After upgrading to 20.1 just found out that for any reasong the parsing process is sending a lot more than the document to be parsed to the OpenAI API.

I am parsing a knowledgebase with over 110,000 documents and there is a lot of spam going to the api with the work "Are you strong enough!?" consuming lots of tokens.

Please see pictures from the OpenAI Logs:

[compatible9]

Just found this code inside the file: "https://github.com/infiniflow/ragflow/blob/main/api/utils/api_utils.py"

Why do the taskexecutor needs to spam the model apis?

async def is_strong_enough(chat_model, embedding_model):
count = settings.STRONG_TEST_COUNT
if not chat_model or not embedding_model:
return
if isinstance(count, int) and count <= 0:
return

@timeout(60, 2)
async def _is_strong_enough():
    nonlocal chat_model, embedding_model
    if embedding_model:
        with trio.fail_after(10):
            _ = await trio.to_thread.run_sync(lambda: embedding_model.encode(["Are you strong enough!?"]))
    if chat_model:
        with trio.fail_after(30):
            res = await trio.to_thread.run_sync(lambda: chat_model.chat("Nothing special.", [{"role": "user", "content": "Are you strong enough!?"}], {}))
        if res.find("**ERROR**") >= 0:
            raise Exception(res)

[Llala0928]

@compatible9 Understood, we have noted it down.

[ZhenhangTung]

Hi! Sorry about it. It's our ping message. Will talk to our eng team about it to see if we can make it better. cc @KevinHuSh