Delete app.py

app.py

@@ -1,322 +0,0 @@
-import os
-import random
-import uuid
-import json
-import time
-import asyncio
-from threading import Thread
-
-import gradio as gr
-import spaces
-import torch
-import numpy as np
-from PIL import Image
-import cv2
-
-from transformers import (
-    Qwen2VLForConditionalGeneration,
-    Qwen2_5_VLForConditionalGeneration,
-    Gemma3ForConditionalGeneration,
-    AutoModelForImageTextToText,
-    AutoProcessor,
-    TextIteratorStreamer,
-)
-from transformers.image_utils import load_image
-
-# Optionally enable synchronous CUDA errors for debugging:
-os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
-
-# Constants for text generation
-MAX_MAX_NEW_TOKENS = 2048
-DEFAULT_MAX_NEW_TOKENS = 1024
-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 models and processors
-# -------------------------------------------------------------------
-
-# VIREX (Video Information Retrieval & Extraction)
-MODEL_ID_VIREX = "prithivMLmods/VIREX-062225-exp"
-processor_virex = AutoProcessor.from_pretrained(MODEL_ID_VIREX, trust_remote_code=True)
-model_virex = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_VIREX,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
-# DREX (Document Retrieval & Extraction Expert)
-MODEL_ID_DREX = "prithivMLmods/DREX-062225-exp"
-processor_drex = AutoProcessor.from_pretrained(MODEL_ID_DREX, trust_remote_code=True)
-model_drex = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_DREX,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
-# Typhoon-OCR-3B (Thai/English OCR parser)
-MODEL_ID_TYPHOON = "sarvamai/sarvam-translate"
-processor_typhoon = AutoProcessor.from_pretrained(MODEL_ID_TYPHOON, trust_remote_code=True)
-model_typhoon = Gemma3ForConditionalGeneration.from_pretrained(
-    MODEL_ID_TYPHOON,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
-# olmOCR-7B-0225-preview (document OCR + LaTeX)
-MODEL_ID_OLM = "allenai/olmOCR-7B-0225-preview"
-processor_olm = AutoProcessor.from_pretrained(MODEL_ID_OLM, trust_remote_code=True)
-model_olm = Qwen2VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_OLM,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
-# -------------------------------------------------------------------
-# Video downsampling helper
-# -------------------------------------------------------------------
-def downsample_video(video_path):
-    """
-    Downsamples the video to 10 evenly spaced frames.
-    Returns a list of (PIL.Image, timestamp) tuples.
-    """
-    vidcap = cv2.VideoCapture(video_path)
-    total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
-    fps = vidcap.get(cv2.CAP_PROP_FPS) or 30.0
-    frames = []
-    frame_indices = np.linspace(0, total_frames - 1, 10, dtype=int)
-    for idx in frame_indices:
-        vidcap.set(cv2.CAP_PROP_POS_FRAMES, idx)
-        success, img = vidcap.read()
-        if not success:
-            continue
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
-        frames.append((Image.fromarray(img), round(idx / fps, 2)))
-    vidcap.release()
-    return frames
-
-# -------------------------------------------------------------------
-# Generation loops
-# -------------------------------------------------------------------
-def _make_generation_kwargs(processor, inputs, streamer, max_new_tokens, do_sample=False, temperature=1.0, top_p=1.0, top_k=0, repetition_penalty=1.0):
-    # ensure pad/eos tokens are defined
-    tok = processor.tokenizer
-    return {
-        **inputs,
-        "streamer": streamer,
-        "max_new_tokens": max_new_tokens,
-        "do_sample": do_sample,
-        "temperature": temperature,
-        "top_p": top_p,
-        "top_k": top_k,
-        "repetition_penalty": repetition_penalty,
-        "pad_token_id": tok.eos_token_id,
-        "eos_token_id": tok.eos_token_id,
-    }
-
-@spaces.GPU
-def generate_image(model_name: str, text: str, image: Image.Image,
-                   max_new_tokens: int = DEFAULT_MAX_NEW_TOKENS,
-                   temperature: float = 0.6,
-                   top_p: float = 0.9,
-                   top_k: int = 50,
-                   repetition_penalty: float = 1.2):
-    # select
-    if model_name.startswith("VIREX"):
-        processor, model = processor_virex, model_virex
-    elif model_name.startswith("DREX"):
-        processor, model = processor_drex, model_drex
-    elif model_name.startswith("olmOCR"):
-        processor, model = processor_olm, model_olm
-    elif model_name.startswith("Typhoon"):
-        processor, model = processor_typhoon, model_typhoon
-    else:
-        yield "Invalid model selected.", "Invalid model selected."
-        return
-
-    if image is None:
-        yield "Please upload an image.", ""
-        return
-
-    # build the chat-style prompt
-    messages = [{
-        "role": "user",
-        "content": [
-            {"type": "image", "image": image},
-            {"type": "text",  "text": text},
-        ]
-    }]
-    prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    inputs = processor(
-        text=[prompt],
-        images=[image],
-        return_tensors="pt",
-        padding=True,
-        truncation=False,
-        max_length=MAX_INPUT_TOKEN_LENGTH
-    ).to(device)
-
-    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-    gen_kwargs = _make_generation_kwargs(
-        processor, inputs, streamer, max_new_tokens,
-        do_sample=True,
-        temperature=temperature,
-        top_p=top_p,
-        top_k=top_k,
-        repetition_penalty=repetition_penalty
-    )
-
-    # launch
-    Thread(target=model.generate, kwargs=gen_kwargs).start()
-    buffer = ""
-    for chunk in streamer:
-        buffer += chunk
-        yield buffer, buffer
-
-@spaces.GPU
-def generate_video(model_name: str, text: str, video_path: str,
-                   max_new_tokens: int = DEFAULT_MAX_NEW_TOKENS,
-                   temperature: float = 0.6,
-                   top_p: float = 0.9,
-                   top_k: int = 50,
-                   repetition_penalty: float = 1.2):
-    # select model
-    if model_name.startswith("VIREX"):
-        processor, model = processor_virex, model_virex
-    elif model_name.startswith("DREX"):
-        processor, model = processor_drex, model_drex
-    elif model_name.startswith("olmOCR"):
-        processor, model = processor_olm, model_olm
-    elif model_name.startswith("Typhoon"):
-        processor, model = processor_typhoon, model_typhoon
-    else:
-        yield "Invalid model selected.", "Invalid model selected."
-        return
-
-    if video_path is None:
-        yield "Please upload a video.", ""
-        return
-
-    # downsample frames
-    frames = downsample_video(video_path)
-
-    # system + user
-    messages = [
-        {"role": "system", "content": [{"type":"text", "text":"You are a helpful assistant."}]},
-        {"role": "user",   "content": [{"type":"text", "text": text}]}
-    ]
-    for img, ts in frames:
-        messages[1]["content"].append({"type":"text", "text":f"Frame {ts}s:"})
-        messages[1]["content"].append({"type":"image", "image":img})
-
-    inputs = processor.apply_chat_template(
-        messages,
-        tokenize=True,
-        add_generation_prompt=True,
-        return_dict=True,
-        return_tensors="pt",
-        truncation=False,
-        max_length=MAX_INPUT_TOKEN_LENGTH
-    ).to(device)
-
-    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-    gen_kwargs = _make_generation_kwargs(
-        processor, inputs, streamer, max_new_tokens,
-        do_sample=True,
-        temperature=temperature,
-        top_p=top_p,
-        top_k=top_k,
-        repetition_penalty=repetition_penalty
-    )
-
-    Thread(target=model.generate, kwargs=gen_kwargs).start()
-    buffer = ""
-    for chunk in streamer:
-        buffer += chunk.replace("<|im_end|>", "")
-        yield buffer, buffer
-
-# -------------------------------------------------------------------
-# Examples, CSS, and launch
-# -------------------------------------------------------------------
-image_examples = [
-    ["Convert this page to doc [text] precisely.", "images/3.png"],
-    ["Convert this page to doc [text] precisely.", "images/4.png"],
-    ["Convert this page to doc [text] precisely.", "images/1.png"],
-    ["Convert chart to OTSL.", "images/2.png"]
-]
-
-video_examples = [
-    ["Explain the video in detail.", "videos/2.mp4"],
-    ["Explain the ad in detail.", "videos/1.mp4"]
-]
-
-css = """
-.submit-btn {
-    background-color: #2980b9 !important;
-    color: white !important;
-}
-.submit-btn:hover {
-    background-color: #3498db !important;
-}
-.canvas-output {
-    border: 2px solid #4682B4;
-    border-radius: 10px;
-    padding: 20px;
-}
-"""
-
-with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
-    gr.Markdown("# **[Doc VLMs OCR](https://huggingface.co/collections/prithivMLmods/multimodal-implementations-67c9982ea04b39f0608badb0)**")
-    with gr.Row():
-        with gr.Column():
-            with gr.Tabs():
-                with gr.TabItem("Image Inference"):
-                    image_query   = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
-                    image_upload  = gr.Image(type="pil", label="Image")
-                    image_submit  = gr.Button("Submit", elem_classes="submit-btn")
-                    gr.Examples(examples=image_examples, inputs=[image_query, image_upload])
-                with gr.TabItem("Video Inference"):
-                    video_query  = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
-                    video_upload = gr.Video(label="Video")
-                    video_submit = gr.Button("Submit", elem_classes="submit-btn")
-                    gr.Examples(examples=video_examples, inputs=[video_query, video_upload])
-
-            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(elem_classes="canvas-output"):
-            gr.Markdown("## Result Canvas")
-            output = gr.Textbox(label="Raw Output Stream", interactive=False, lines=2)
-            markdown_output = gr.Markdown(label="Formatted Result (Result.Md)")
-
-        model_choice = gr.Radio(
-            choices=["DREX-062225-7B-exp", "olmOCR-7B-0225-preview", "VIREX-062225-7B-exp", "Typhoon-OCR-3B"],
-            label="Select Model",
-            value="DREX-062225-7B-exp"
-        )
-
-        gr.Markdown("**Model Info 💻** | [Report Bug](https://huggingface.co/spaces/prithivMLmods/Doc-VLMs/discussions)")
-        gr.Markdown("> [DREX-062225-7B-exp](https://huggingface.co/prithivMLmods/DREX-062225-exp): ...")
-        gr.Markdown("> [VIREX-062225-7B-exp](https://huggingface.co/prithivMLmods/VIREX-062225-exp): ...")
-        gr.Markdown("> [Typhoon-OCR-3B](https://huggingface.co/scb10x/typhoon-ocr-3b): ...")
-        gr.Markdown("> [olmOCR-7B-0225](https://huggingface.co/allenai/olmOCR-7B-0225-preview): ...")
-        gr.Markdown("> ⚠️ note: video inference may be less reliable.")
-
-    image_submit.click(
-        fn=generate_image,
-        inputs=[model_choice, image_query, image_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
-        outputs=[output, markdown_output]
-    )
-    video_submit.click(
-        fn=generate_video,
-        inputs=[model_choice, video_query, video_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
-        outputs=[output, markdown_output]
-    )
-
-if __name__ == "__main__":
-    demo.queue(max_size=30).launch(share=True, mcp_server=True, ssr_mode=False, show_error=True)