Ask-my-pdf -> Q&A for PDF files using OpenAI API

[Klaudioz]

I just did that:

export S3_PREFIX=""
export S3_REGION="us-east-1"
export S3_BUCKET="my-bucket"
export S3_SECRET="XXXXXXXXXXXXXXXXXXX"
export S3_KEY="YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY"
export S3_URL="https://s3.amazonaws.com"

And I’m not seeing any error, but nothing is uploaded to my bucket. Using debug doesn’t show anything new. I’ve used multiple times this bucket so I know it’s OK.

[kerbal]

In Streamlit Cloud you can set environment variables during the deployment in the “Advanced settings”.

[kerbal]

I’ve tested it only with Digital Ocean. I cannot promise when I’ll be able to test it with AWS.

BUT when running ask-my-pdf localy you can use LocalStorage. Just change line 58 in gui.py:

[Goyo]

Oh, I forgot about that.

[Klaudioz]

@Klaudioz

S3_PREFIX

I was able to make it work with S3, but I don’t understand how to “load” the stored documents when I closed it and opened it again.
I recorded a video, I got the following error:

Traceback (most recent call last):
  File "/usr/local/lib/python3.10/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 565, in _run_script
    exec(code, module.__dict__)
  File "/Users/canalescl/personal/replit/ask-my-pdf/src/gui.py", line 251, in <module>
    b_ask()
  File "/Users/canalescl/personal/replit/ask-my-pdf/src/gui.py", line 168, in b_ask
    summary = ss['index']['summary']
  File "/usr/local/lib/python3.10/site-packages/streamlit/runtime/state/session_state_proxy.py", line 89, in __getitem__
    return get_session_state()[key]
  File "/usr/local/lib/python3.10/site-packages/streamlit/runtime/state/safe_session_state.py", line 110, in __getitem__
    return self._state[key]
  File "/usr/local/lib/python3.10/site-packages/streamlit/runtime/state/session_state.py", line 438, in __getitem__
    raise KeyError(_missing_key_error_message(key))
KeyError: 'st.session_state has no key "index". Did you forget to initialize it? More info: https://docs.streamlit.io/library/advanced-features/session-state#initialization'

[kerbal]

Ok, so around 1:15 you only inspected the available files but you actually haven’t selected your file - just click on the name.

I will disable the “get answer” button when no file is loaded.

[Klaudioz]

Hi, thanks for your help. I was able to make it work.

You are moving very fast, so I’m trying to catch you. I saw your updated README.md but can’t see any reference to REDIS_URL.

Also, don’t you need a password for it?

[kerbal]

Redis url can contain all the important parameters about the db connection (host,port,password,tls/ssl,db-id). But Redis is not required to run the project. When STATS_MODE is not set to REDIS usage stats will be aggregated in a Dict object (not persistent). I’ll add the info about this to the readme.

[Klaudioz]

Cool, it’s working !!

I have another question. For example, if I have a 1000 pages pdf. I think if I’m using an S3 bucket, when I open the app again, I need to re-load in memory the entire document, and I guess it’s downloading everything from S3 every time.
So, how big should the performance improvement be using something like embedding databases like Pinecone or Chroma?

[kerbal]

For a single document - not that big really. You can expect a saving of <3s on startup (S3 fetch + deserialization) and <1s on each query (assuming 4 fragments per page). The reason why I’m using S3 here is that it’s fast enough and dirt-cheap (250GB → $5/month).

[kerbal]

Ask-My-PDF now uses gpt-3.5-turbo and allows the user to switch between models
I’ve also added the option to provide feedback about the answer.

[kerbal]

You can now try ask-my-pdf without an API key
Every day, a pool of 1,000,000 community tokens is available to everyone.

[ferdy]

What is feedback score and community token?

[kerbal]

Feedback score - number of feedback actions (thumb up / down) performed by the user (if API key is provided) or by the community (the default “user”).

Community tokens - measure of a budget that is provided by me (on my OpenAI API key) for people without OpenAI API key, so they can test/use the app. Currently it’s the equivalent of 1,000,000 gtp-3.5-turbo tokens per day.

[osis]

I love this. I am wondering if it is possible to run this script and get responses from terminal without web browser? Could someone help me out?

[dacostadaniel]

looking to develop an app; a startup situtaion; seeking your participation; would you be interested?

Daniel

[shawngiese]

This is really nice. Thank you for sharing the code. What do you think of PDF with mixed data… for example text and tables? For example, if I enter in the CSV directly into my prompt I can query it, but tables / CSV in a PDF are not being detected. I think there needs to be a way to identify sparse and dense embeddings maybe.

[ndrgnz]

Nice project!
A few questions:

• is it possible to hide entirely the API key section and just use the secret one provided as a secret streamlit?
• is it possible to hide the feedback part?
• is it possible to hide the left panel?
• how about implementi some OCR librarie for PDF?
  Thanks!

[Ahmed]

When running the app within streamlit, the community version tab does not show up, only the “Enter your OpenAI API key” does. Do you have to add the OPENAI_KEY set to your openai key within advanced settings? Is there something to edit with the COMMUNITY_USER?