You can deploy this model and start earning money today!

strangervisionhf/
paddle.ocr_path_expose

Name: paddle.ocr_path_expose
Rating: 2.5 (8 reviews)

text-generation-inference

Views

No views yet

[!important] This is the OCR weight component of the PaddlePaddle/PaddleOCR-VL model. These weights cannot be used for other use cases. If you wish to do so, please visit the original model page!

This repository directly exposes the OCR-only weights for smoother transformers implementation of the PaddleOCR-VL model.

[!note] Last updated: 4:00 AM (IST), October 25, 2025.

[!note] The latest transformers version used as of the above date is transformers==4.57.1 and the torch version 2.8.0+cu126

Quick Start with Transformers

Install the required packages

!pip install transformers torch torchvision gradio hf_xet \
             huggingface_hub pillow accelerate peft \
             matplotlib requests einops av sentencepiece\
             transformers-stream-generator

notebook login

from huggingface_hub import notebook_login, HfApi
notebook_login()

Run [app.py]

import os
import sys
from threading import Thread
from typing import Iterable

import gradio as gr
import torch
from PIL import Image
from transformers import (
    AutoModelForCausalLM,
    AutoProcessor,
    TextIteratorStreamer,
)

from gradio.themes import Soft
from gradio.themes.utils import colors, fonts, sizes

# --- Theme and CSS Setup ---
colors.steel_blue = colors.Color(
    name="steel_blue",
    c50="#EBF3F8",
    c100="#D3E5F0",
    c200="#A8CCE1",
    c300="#7DB3D2",
    c400="#529AC3",
    c500="#4682B4",
    c600="#3E72A0",
    c700="#36638C",
    c800="#2E5378",
    c900="#264364",
    c950="#1E3450",
)

class SteelBlueTheme(Soft):
    def __init__(
        self,
        *,
        primary_hue: colors.Color | str = colors.gray,
        secondary_hue: colors.Color | str = colors.steel_blue,
        neutral_hue: colors.Color | str = colors.slate,
        text_size: sizes.Size | str = sizes.text_lg,
        font: fonts.Font | str | Iterable[fonts.Font | str] = (
            fonts.GoogleFont("Outfit"), "Arial", "sans-serif",
        ),
        font_mono: fonts.Font | str | Iterable[fonts.Font | str] = (
            fonts.GoogleFont("IBM Plex Mono"), "ui-monospace", "monospace",
        ),
    ):
        super().__init__(
            primary_hue=primary_hue,
            secondary_hue=secondary_hue,
            neutral_hue=neutral_hue,
            text_size=text_size,
            font=font,
            font_mono=font_mono,
        )
        super().set(
            background_fill_primary="*primary_50",
            background_fill_primary_dark="*primary_900",
            body_background_fill="linear-gradient(135deg, *primary_200, *primary_100)",
            body_background_fill_dark="linear-gradient(135deg, *primary_900, *primary_800)",
            button_primary_text_color="white",
            button_primary_text_color_hover="white",
            button_primary_background_fill="linear-gradient(90deg, *secondary_500, *secondary_600)",
            button_primary_background_fill_hover="linear-gradient(90deg, *secondary_600, *secondary_700)",
            button_primary_background_fill_dark="linear-gradient(90deg, *secondary_600, *secondary_700)",
            button_primary_background_fill_hover_dark="linear-gradient(90deg, *secondary_500, *secondary_600)",
            slider_color="*secondary_500",
            slider_color_dark="*secondary_600",
            block_title_text_weight="600",
            block_border_width="3px",
            block_shadow="*shadow_drop_lg",
            button_primary_shadow="*shadow_drop_lg",
            button_large_padding="11px",
            color_accent_soft="*primary_100",
            block_label_background_fill="*primary_200",
        )

steel_blue_theme = SteelBlueTheme()

css = """
#main-title h1 {
    font-size: 2.3em !important;
}
#output-title h2 {
    font-size: 2.1em !important;
}
"""

# --- Model Configuration and Loading ---
MAX_MAX_NEW_TOKENS = 4096
DEFAULT_MAX_NEW_TOKENS = 2048
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# Load PaddleOCR
MODEL_ID_P = "strangervisionhf/paddle.ocr_path_expose" # ->  Original model: https://huggingface.co/PaddlePaddle/PaddleOCR-VL
processor = AutoProcessor.from_pretrained(MODEL_ID_P, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID_P,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16
).to(device).eval()

# --- Generation Function ---
def generate_image(text: str, image: Image.Image,
                   max_new_tokens: int = 1024,
                   temperature: float = 0.6,
                   top_p: float = 0.9,
                   top_k: int = 50,
                   repetition_penalty: float = 1.2):
    """Generate responses for image input using the PaddleOCR model."""
    if image is None:
        yield "Please upload an image.", "Please upload an image."
        return

    images = [image.convert("RGB")]

    # PaddleOCR has a specific message format
    messages = [
        {"role": "user", "content": text}
    ]
        
    prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
    inputs = processor(text=prompt, images=images, return_tensors="pt").to(device)

    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
    generation_kwargs = {
        **inputs,
        "streamer": streamer,
        "max_new_tokens": max_new_tokens,
        "temperature": temperature,
        "top_p": top_p,
        "top_k": top_k,
        "repetition_penalty": repetition_penalty,
        "do_sample": True
    }
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    buffer = ""
    for new_text in streamer:
        buffer += new_text.replace("<|im_end|>", "").replace("<end_of_utterance>", "")
        yield buffer, buffer


with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
    gr.Markdown("# **Paddle OCR Only**", elem_id="main-title")
    with gr.Row():
        with gr.Column(scale=2):
            image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
            image_upload = gr.Image(type="pil", label="Upload Image", height=320)
            image_submit = gr.Button("Submit", variant="primary")

            with gr.Accordion("Advanced options", open=False):
                max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
                temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6)
                top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9)
                top_k = gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50)
                repetition_penalty = gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2)

        with gr.Column(scale=3):
            gr.Markdown("## Output", elem_id="output-title")
            raw_output = gr.Textbox(label="Raw Output Stream", interactive=False, lines=11, show_copy_button=True)
            with gr.Accordion("[Result.md]", open=False):
                formatted_output = gr.Markdown(label="Formatted Result")
            
            gr.Markdown("Note: Currently, PaddleOCR VL only supports OCR inference. Structured OCR document parsing transformer inference is coming soon. [Report – Bug/Issue](https://huggingface.co/spaces/prithivMLmods/Multimodal-OCR3/discussions/1)")
            
    image_submit.click(
        fn=generate_image,
        inputs=[image_query, image_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
        outputs=[raw_output, formatted_output]
    )

if __name__ == "__main__":
    demo.queue(max_size=50).launch(mcp_server=True, ssr_mode=False, show_error=True)