app.py

import gradio as gr
from huggingface_hub import InferenceClient
import pytesseract
from PIL import Image
from pypdf import PdfReader
import ocrmypdf
import os

# Image to Text

def fn_image_to_text(input_image):
    return pytesseract.image_to_string(Image.open(input_image))

# PDF to Text

def fn_pdf_to_text(input_pdf):
    reader = PdfReader(input_pdf)
    
    output_pdf = ""
    for page in reader.pages:
        output_pdf+=page.extract_text()
  
    image_count = 0
    for page in reader.pages:
        image_count += len(page.images)

    if image_count > 0 and len(output_pdf) < 1000:
        input_pdf_ocr = input_pdf.replace(".pdf", " - OCR.pdf")
        ocrmypdf.ocr(input_pdf, input_pdf_ocr, force_ocr=True)
    
        reader = PdfReader(input_pdf_ocr)
        output_pdf = ""
        for page in reader.pages:
            output_pdf+=page.extract_text()

        os.remove(input_pdf_ocr)
  
    return output_pdf

# Inference

model_text = "google/gemma-2-27b-it"
model_vision = "google/paligemma2-3b-pt-224"

client = InferenceClient()

def fn_text(
    prompt,
    history,
    input,
    #system_prompt,
    max_tokens,
    temperature,
    top_p,
):
    if input:
        if os.path.splitext(input)[1].lower() in [".png", ".jpg", ".jpeg"]:
            output = fn_image_to_text(input)
        if os.path.splitext(input)[1].lower() == ".pdf":
            output = fn_pdf_to_text(input)
    else:
        output = ""
    
    #messages = [{"role": "system", "content": system_prompt}]
    #history.append(messages[0])
    #messages.append({"role": "user", "content": prompt})
    #history.append(messages[1])

    messages = [{"role": "user", "content": prompt + " " + output}]
    history.append(messages[0])
    
    stream = client.chat.completions.create(
        model = model_text,
        messages = history,
        max_tokens = max_tokens,
        temperature = temperature,
        top_p = top_p,
        stream = True,
    )
    
    chunks = []
    for chunk in stream:
        chunks.append(chunk.choices[0].delta.content or "")
        yield "".join(chunks)

app_text = gr.ChatInterface(
    fn = fn_text,
    type = "messages",
    additional_inputs = [
        gr.File(type="filepath", label="Input"),
        #gr.Textbox(value="You are a helpful assistant.", label="System Prompt"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P"),
    ],
    title = "Google Gemma",
    description = model_text,
)

def fn_vision(
    prompt,
    image_url,
    #system_prompt,
    max_tokens,
    temperature,
    top_p,
):
    messages = [{"role": "user", "content": [{"type": "text", "text": prompt}]}]
    
    if image_url:
        messages[0]["content"].append({"type": "image_url", "image_url": {"url": image_url}})
    
    stream = client.chat.completions.create(
        model = model_vision,
        messages = messages,
        max_tokens = max_tokens,
        temperature = temperature,
        top_p = top_p,
        stream = True,
    )
    
    chunks = []
    for chunk in stream:
        chunks.append(chunk.choices[0].delta.content or "")
        yield "".join(chunks)

app_vision = gr.Interface(
    fn = fn_vision,
    inputs = [
        gr.Textbox(label="Prompt"),
        gr.Textbox(label="Image URL")
    ],
    outputs = [
        gr.Textbox(label="Output")
    ],
    additional_inputs = [
        #gr.Textbox(value="You are a helpful assistant.", label="System Prompt"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-P"),
    ],
    title = "Google Gemma",
    description = model_vision,
)

app = gr.TabbedInterface(
    [app_text, app_vision],
    ["Text", "Vision"]
).launch()

#if __name__ == "__main__":
#    app.launch()