Deploy OvisOCR2 OCR demo

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+dist/examples/handwriting-040745.jpg filter=lfs diff=lfs merge=lfs -text
+dist/examples/handwriting-042609.jpg filter=lfs diff=lfs merge=lfs -text
+dist/examples/page-9477c155-19b3-4f2d-ac18-ee3f8997d2f8.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,109 @@
 ---
 title: OvisOCR2
-emoji: 🐢
-colorFrom: yellow
+emoji: 🔎
+colorFrom: blue
 colorTo: gray
 sdk: gradio
 sdk_version: 6.20.0
-python_version: '3.12'
+python_version: "3.12.12"
 app_file: app.py
 pinned: false
+license: apache-2.0
+hardware: zero-gpu
+startup_duration_timeout: 30m
+models:
+  - ATH-MaaS/OvisOCR2
+preload_from_hub:
+  - ATH-MaaS/OvisOCR2
+short_description: Stream structured Markdown from document images and PDFs.
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# OvisOCR2 — document OCR demo
+
+A focused image and PDF document parsing demo for
+[ATH-MaaS/OvisOCR2](https://huggingface.co/ATH-MaaS/OvisOCR2). It preserves
+reading order and emits Markdown with LaTeX formulas, HTML tables, and rendered
+visual-region crops. Model tokens stream into the interface as they are generated.
+PDFs are rasterized locally and processed one page at a time in Base mode. Each
+page uses an independent streamed GPU request and the browser merges the results,
+so long documents do not have to fit inside one ZeroGPU lease. This demo
+intentionally does not expose a separate Long mode.
+
+The five packaged examples cover two financial tables, two Chinese handwriting
+pages, and one formula-rich OmniDocBench research-paper page. They are selectable
+directly in the interface and are also covered by the local smoke-test workflow.
+The workspace opens empty so examples remain an explicit user choice rather than
+an automatic input.
+
+## Local run
+
+The verified local model directory is `/root/models/ATH-MaaS/OvisOCR2`.
+
+```bash
+conda env create -f environment.yml
+conda activate ovisocr2-demo
+npm ci
+npm run build
+CUDA_VISIBLE_DEVICES=0 \
+OVISOCR_MODEL_PATH=/root/models/ATH-MaaS/OvisOCR2 \
+python app.py
+```
+
+Open `http://127.0.0.1:7860`.
+
+### DSW proxy
+
+The server already listens on `0.0.0.0`. Set a DSW id and the public proxy URL
+is generated from the same `PORT` automatically:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 \
+PORT=7869 \
+OVISOCR_DSW_ID=1976932 \
+OVISOCR_MODEL_PATH=/root/models/ATH-MaaS/OvisOCR2 \
+python app.py
+```
+
+The corresponding URL is
+`https://1976932-proxy-7869.dsw-gateway-cn-hangzhou.data.aliyun.com/`.
+For a different gateway or an explicit public URL, set `OVISOCR_ROOT_PATH`
+instead; it takes precedence over `OVISOCR_DSW_ID`.
+
+DSW's per-port proxy uses its own hostname and serves at `/`, so the ASGI root
+path stays empty. If an explicit URL contains a real subpath, such as
+`https://example.test/ocr/`, the app passes only `/ocr` to Gradio's `root_path`;
+passing the whole URL into this custom FastAPI server would break static mounts.
+
+For a UI-only smoke that does not load the model:
+
+```bash
+OVISOCR_TEST_MODE=1 python app.py
+```
+
+## Hugging Face Space
+
+The checked-in `dist/` directory is the production frontend, so the folder can
+be uploaded directly to a Gradio Space without a Node build step. Select
+ZeroGPU hardware in the Space settings. The README metadata pins Python 3.12.12
+and Gradio 6.20.0
+and declares `hardware: zero-gpu`; `app.py` decorates only the inference endpoint
+with `@spaces.GPU` and loads the model on CUDA at module scope, following the
+ZeroGPU loading contract.
+
+On Hugging Face, the model source automatically falls back from the local path
+to `ATH-MaaS/OvisOCR2`. PDFs are limited to 50 pages by default. Every page gets
+its own ZeroGPU reservation while the frontend preserves already completed pages.
+
+## Runtime knobs
+
+- `OVISOCR_MODEL_PATH`: local directory or Hub model id.
+- `OVISOCR_MAX_NEW_TOKENS`: defaults to `16384`, matching the model README inference configuration.
+- `OVISOCR_MAX_PDF_PAGES`: defaults to `50`.
+- `OVISOCR_PDF_RENDER_SCALE`: PDF rasterization scale, defaults to `2.0`.
+- `OVISOCR_STREAM_MIN_CHARS`: minimum new characters between stream updates, defaults to `24`.
+- `OVISOCR_STREAM_MAX_INTERVAL`: maximum seconds between available stream updates, defaults to `0.12`.
+- `OVISOCR_GPU_DURATION`: per-page ZeroGPU reservation in seconds, defaults to `300`.
+- `OVISOCR_ATTN_IMPLEMENTATION`: defaults to `sdpa`.
+- `OVISOCR_DSW_ID`: generates the Hangzhou DSW proxy root using the active `PORT`.
+- `OVISOCR_ROOT_PATH`: explicit proxy URL or ASGI path prefix; overrides `OVISOCR_DSW_ID`.
+- `OVISOCR_TEST_MODE=1`: deterministic mock inference for UI/integration tests.

app.py

@@ -0,0 +1,574 @@
+from __future__ import annotations
+
+import base64
+import html
+import io
+import os
+import re
+import threading
+import time
+from collections.abc import Iterator
+from pathlib import Path
+from typing import Any, Callable
+from urllib.parse import urlsplit
+
+os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
+os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
+
+try:
+    import spaces  # Must be imported before torch on Hugging Face ZeroGPU.
+except ImportError:  # Local development uses a no-op decorator.
+    class _LocalSpaces:
+        @staticmethod
+        def GPU(*decorator_args: Any, **decorator_kwargs: Any) -> Callable:
+            def decorate(function: Callable) -> Callable:
+                return function
+
+            if decorator_args and callable(decorator_args[0]) and len(decorator_args) == 1:
+                return decorator_args[0]
+            return decorate
+
+    spaces = _LocalSpaces()
+
+import gradio as gr
+import fitz
+import torch
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import FileResponse, JSONResponse
+from gradio.data_classes import FileData
+from PIL import Image, ImageOps
+from starlette.staticfiles import StaticFiles
+
+
+BASE_DIR = Path(__file__).resolve().parent
+DIST_DIR = BASE_DIR / "dist"
+MODEL_ID = "ATH-MaaS/OvisOCR2"
+LOCAL_MODEL_DEFAULT = Path("/root/models/ATH-MaaS/OvisOCR2")
+TEST_MODE = os.getenv("OVISOCR_TEST_MODE", "0").lower() in {"1", "true", "yes"}
+MODEL_SOURCE = os.getenv(
+    "OVISOCR_MODEL_PATH",
+    str(LOCAL_MODEL_DEFAULT if LOCAL_MODEL_DEFAULT.is_dir() else MODEL_ID),
+)
+MAX_NEW_TOKENS = int(os.getenv("OVISOCR_MAX_NEW_TOKENS", "16384"))
+MAX_PDF_PAGES = int(os.getenv("OVISOCR_MAX_PDF_PAGES", "50"))
+PDF_RENDER_SCALE = float(os.getenv("OVISOCR_PDF_RENDER_SCALE", "2.0"))
+STREAM_MIN_CHARS = int(os.getenv("OVISOCR_STREAM_MIN_CHARS", "24"))
+STREAM_MAX_INTERVAL = float(os.getenv("OVISOCR_STREAM_MAX_INTERVAL", "0.12"))
+MIN_PIXELS = 448 * 448
+MAX_PIXELS = 2880 * 2880
+
+
+def server_config() -> tuple[int, str | None, str | None]:
+    """Resolve the port, ASGI path prefix, and optional public proxy URL."""
+    port = int(os.getenv("PORT", os.getenv("GRADIO_SERVER_PORT", "7860")))
+    configured_root = (
+        os.getenv("OVISOCR_ROOT_PATH", "").strip()
+        or os.getenv("GRADIO_ROOT_PATH", "").strip()
+    )
+    dsw_id = os.getenv("OVISOCR_DSW_ID", "").strip()
+    public_url = None
+    root_path = None
+    if configured_root.startswith(("http://", "https://")):
+        public_url = configured_root
+        path = urlsplit(configured_root).path.rstrip("/")
+        root_path = path or None
+    elif configured_root:
+        root_path = configured_root.rstrip("/") or None
+    elif dsw_id:
+        public_url = (
+            f"https://{dsw_id}-proxy-{port}."
+            "dsw-gateway-cn-hangzhou.data.aliyun.com/"
+        )
+    return port, root_path, public_url
+
+
+SERVER_PORT, ROOT_PATH, PUBLIC_URL = server_config()
+
+OCR_PROMPT = (
+    "\nExtract all readable content from the image in natural human reading order "
+    "and output the result as a single Markdown document. For charts or images, "
+    'represent them using an HTML image tag: <img src="images/bbox_{left}_{top}_{right}_{bottom}.jpg" />, '
+    "where left, top, right, bottom are bounding box coordinates scaled to [0, 1000). "
+    "Format formulas as LaTeX. Format tables as HTML: <table>...</table>. "
+    "Transcribe all other text as standard Markdown. Preserve the original text "
+    "without translation or paraphrasing."
+)
+
+BBOX_IMAGE_PATTERN = re.compile(
+    r'<img\s+src=["\']images/bbox_(\d+)_(\d+)_(\d+)_(\d+)\.jpg["\']\s*/?>',
+    flags=re.IGNORECASE,
+)
+UNMATERIALIZED_BBOX_IMAGE_PATTERN = re.compile(
+    r'<img\b[^>]*\bsrc=["\']images/bbox_[^"\']+["\'][^>]*>',
+    flags=re.IGNORECASE,
+)
+
+MOCK_MARKDOWN = r"""# 盈利预测、估值与评级
+
+我们预测公司 2024—2026 年营业收入与归母净利润将保持稳健增长，当前股价对应估值如下。
+
+<table>
+  <thead><tr><th>项目</th><th>2023A</th><th>2024E</th><th>2025E</th><th>2026E</th></tr></thead>
+  <tbody>
+    <tr><td>营业收入（百万元）</td><td>9,423</td><td>10,516</td><td>11,873</td><td>13,441</td></tr>
+    <tr><td>归母净利润（百万元）</td><td>1,267</td><td>1,452</td><td>1,681</td><td>1,946</td></tr>
+    <tr><td>每股收益（元）</td><td>1.02</td><td>1.17</td><td>1.36</td><td>1.57</td></tr>
+    <tr><td>市盈率</td><td>18.4</td><td>16.1</td><td>13.8</td><td>12.0</td></tr>
+  </tbody>
+</table>
+
+## 财务摘要
+
+净资产收益率采用 $ROE = \frac{NP}{E}$ 计算；预计 2025 年利润同比增速为：
+
+\[
+g = \frac{1{,}681 - 1{,}452}{1{,}452} \times 100\% = 15.8\%.
+\]
+
+<img src="images/bbox_120_130_880_420.jpg" />
+
+资料来源：公司公告，研究团队整理。"""
+
+
+processor = None
+model = None
+
+
+def _load_model() -> None:
+    global processor, model
+    if TEST_MODE:
+        return
+
+    from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
+
+    local_only = Path(MODEL_SOURCE).is_dir()
+    processor = AutoProcessor.from_pretrained(
+        MODEL_SOURCE,
+        min_pixels=MIN_PIXELS,
+        max_pixels=MAX_PIXELS,
+        local_files_only=local_only,
+    )
+    model = Qwen3_5ForConditionalGeneration.from_pretrained(
+        MODEL_SOURCE,
+        dtype=torch.bfloat16,
+        attn_implementation=os.getenv("OVISOCR_ATTN_IMPLEMENTATION", "sdpa"),
+        local_files_only=local_only,
+    ).to("cuda")
+    model.eval()
+
+
+def clean_truncated_repeats(
+    text: str,
+    min_text_len: int = 8000,
+    max_period: int = 200,
+    min_period: int = 1,
+    min_repeat_chars: int = 100,
+    min_repeat_times: int = 5,
+) -> str:
+    """Remove a repeated suffix created when generation reaches its token ceiling."""
+    n = len(text)
+    if n < min_text_len:
+        return text
+
+    max_period = min(max_period, n - 1)
+    for unit_len in range(min_period, max_period + 1):
+        if text[n - 1] != text[n - 1 - unit_len]:
+            continue
+        match_len = 1
+        idx = n - 2
+        while idx >= unit_len and text[idx] == text[idx - unit_len]:
+            match_len += 1
+            idx -= 1
+        total_len = match_len + unit_len
+        repeat_times = total_len // unit_len
+        tail_len = total_len % unit_len
+        if repeat_times >= min_repeat_times and total_len >= min_repeat_chars:
+            return text[: n - total_len + unit_len] + text[n - tail_len :]
+    return text
+
+
+def materialize_bbox_images(markdown: str, page_image: Image.Image) -> str:
+    """Replace bbox image placeholders in rendered output with safe data-URI crops.
+
+    Raw model Markdown is returned separately and remains unchanged.
+    """
+    width, height = page_image.size
+
+    def replace(match: re.Match[str]) -> str:
+        left, top, right, bottom = (int(value) for value in match.groups())
+        x1 = max(0, min(width, round(left * width / 1000)))
+        y1 = max(0, min(height, round(top * height / 1000)))
+        x2 = max(0, min(width, round(right * width / 1000)))
+        y2 = max(0, min(height, round(bottom * height / 1000)))
+        if x2 <= x1 or y2 <= y1:
+            return match.group(0)
+
+        crop = page_image.crop((x1, y1, x2, y2)).convert("RGB")
+        crop.thumbnail((1600, 1600), Image.Resampling.LANCZOS)
+        buffer = io.BytesIO()
+        crop.save(buffer, format="JPEG", quality=88, optimize=True)
+        payload = base64.b64encode(buffer.getvalue()).decode("ascii")
+        return f'<img src="data:image/jpeg;base64,{payload}" alt="Visual region" />'
+
+    return neutralize_unmaterialized_bbox_images(BBOX_IMAGE_PATTERN.sub(replace, markdown))
+
+
+def neutralize_unmaterialized_bbox_images(markdown: str) -> str:
+    """Render placeholder examples as code instead of issuing broken requests."""
+
+    def replace(match: re.Match[str]) -> str:
+        escaped = html.escape(match.group(0), quote=False)
+        return f'<code class="unresolved-image-reference">{escaped}</code>'
+
+    return UNMATERIALIZED_BBOX_IMAGE_PATTERN.sub(replace, markdown)
+
+
+def stream_safe_markdown(markdown: str) -> str:
+    """Avoid broken image requests until a page's bbox crops are materialized."""
+    return neutralize_unmaterialized_bbox_images(
+        BBOX_IMAGE_PATTERN.sub(
+            '<div class="visual-placeholder">Preparing visual region…</div>',
+            markdown,
+        )
+    )
+
+
+def _file_path(file_data: FileData | dict[str, Any]) -> str:
+    if isinstance(file_data, dict):
+        path = file_data.get("path")
+    else:
+        path = getattr(file_data, "path", None)
+    if not path:
+        raise ValueError("No uploaded document was provided.")
+    return str(path)
+
+
+def generation_token_ids(active_processor: Any) -> dict[str, int]:
+    """Use tokenizer stop IDs; this checkpoint's config and tokenizer differ."""
+    tokenizer = active_processor.tokenizer
+    return {
+        "eos_token_id": int(tokenizer.eos_token_id),
+        "pad_token_id": int(tokenizer.pad_token_id),
+    }
+
+
+def document_info(path: str) -> tuple[str, int]:
+    suffix = Path(path).suffix.lower()
+    try:
+        with Path(path).open("rb") as file:
+            header = file.read(5)
+    except OSError as error:
+        raise ValueError("The uploaded document could not be read.") from error
+
+    # Browser uploads can arrive at Gradio as an extensionless temporary `blob`.
+    if suffix == ".pdf" or header == b"%PDF-":
+        with fitz.open(path) as document:
+            total_pages = document.page_count
+        if total_pages < 1:
+            raise ValueError("The uploaded PDF has no pages.")
+        if total_pages > MAX_PDF_PAGES:
+            raise ValueError(
+                f"This demo accepts up to {MAX_PDF_PAGES} PDF pages; received {total_pages}."
+            )
+        return "pdf", total_pages
+
+    try:
+        with Image.open(path) as source:
+            source.verify()
+    except Exception as error:
+        raise ValueError("Please upload a valid PNG, JPEG, WebP, or PDF file.") from error
+    return "image", 1
+
+
+def load_document_page(path: str, document_type: str, page_index: int) -> Image.Image:
+    if document_type == "pdf":
+        with fitz.open(path) as document:
+            page = document.load_page(page_index)
+            pixmap = page.get_pixmap(
+                matrix=fitz.Matrix(PDF_RENDER_SCALE, PDF_RENDER_SCALE),
+                colorspace=fitz.csRGB,
+                alpha=False,
+            )
+        return Image.frombytes("RGB", (pixmap.width, pixmap.height), pixmap.samples)
+
+    with Image.open(path) as source:
+        return ImageOps.exif_transpose(source).convert("RGB")
+
+
+def page_preview_data_uri(page_image: Image.Image) -> str:
+    preview = page_image.copy().convert("RGB")
+    preview.thumbnail((1400, 1800), Image.Resampling.LANCZOS)
+    buffer = io.BytesIO()
+    preview.save(buffer, format="JPEG", quality=82, optimize=True)
+    payload = base64.b64encode(buffer.getvalue()).decode("ascii")
+    return f"data:image/jpeg;base64,{payload}"
+
+
+def _model_inputs(page_image: Image.Image) -> Any:
+    if processor is None or model is None:
+        raise RuntimeError("OvisOCR2 is not loaded.")
+
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "image", "image": page_image},
+                {"type": "text", "text": OCR_PROMPT},
+            ],
+        }
+    ]
+    return processor.apply_chat_template(
+        messages,
+        tokenize=True,
+        add_generation_prompt=True,
+        return_dict=True,
+        return_tensors="pt",
+        enable_thinking=False,
+    ).to(model.device)
+
+
+def infer_stream(page_image: Image.Image) -> Iterator[str]:
+    if TEST_MODE:
+        for end in range(64, len(MOCK_MARKDOWN) + 64, 64):
+            yield MOCK_MARKDOWN[:end]
+        return
+    if processor is None or model is None:
+        raise RuntimeError("OvisOCR2 is not loaded.")
+
+    from transformers import TextIteratorStreamer
+
+    inputs = _model_inputs(page_image)
+    streamer = TextIteratorStreamer(
+        processor.tokenizer,
+        skip_prompt=True,
+        skip_special_tokens=True,
+        clean_up_tokenization_spaces=False,
+        timeout=30.0,
+    )
+    errors: list[BaseException] = []
+
+    def generate() -> None:
+        try:
+            with torch.inference_mode():
+                model.generate(
+                    **inputs,
+                    streamer=streamer,
+                    max_new_tokens=MAX_NEW_TOKENS,
+                    do_sample=False,
+                    temperature=None,
+                    top_p=None,
+                    top_k=None,
+                    **generation_token_ids(processor),
+                )
+        except BaseException as error:
+            errors.append(error)
+            streamer.on_finalized_text("", stream_end=True)
+
+    worker = threading.Thread(target=generate, name="ovisocr2-generate", daemon=True)
+    worker.start()
+    text = ""
+    last_yielded = ""
+    last_yield_time = time.monotonic()
+    for fragment in streamer:
+        text += fragment
+        now = time.monotonic()
+        if (
+            len(text) - len(last_yielded) >= STREAM_MIN_CHARS
+            or now - last_yield_time >= STREAM_MAX_INTERVAL
+        ):
+            yield text
+            last_yielded = text
+            last_yield_time = now
+
+    worker.join()
+    if errors:
+        raise RuntimeError("Model generation failed.") from errors[0]
+    final_text = clean_truncated_repeats(text.strip())
+    if final_text and final_text != last_yielded:
+        yield final_text
+
+
+def combine_pages(pages: list[dict[str, Any]], field: str) -> str:
+    if len(pages) <= 1:
+        return pages[0].get(field, "") if pages else ""
+    return "\n\n---\n\n".join(
+        f"<!-- Page {page['page_number']} -->\n\n{page.get(field, '')}" for page in pages
+    )
+
+
+def stream_payload(
+    *,
+    event: str,
+    pages: list[dict[str, Any]],
+    current_page: int,
+    total_pages: int,
+    document_type: str,
+    started: float,
+    page_preview: str | None = None,
+) -> dict[str, Any]:
+    return {
+        "event": event,
+        "markdown": combine_pages(pages, "markdown"),
+        "render_markdown": combine_pages(pages, "render_markdown"),
+        "pages": pages,
+        "current_page": current_page,
+        "total_pages": total_pages,
+        "document_type": document_type,
+        "page_preview": page_preview,
+        "char_count": sum(len(page.get("markdown", "")) for page in pages),
+        "elapsed_seconds": round(time.perf_counter() - started, 3),
+        "model": MODEL_ID,
+        "backend": "mock" if TEST_MODE else "transformers",
+        "mode": "base",
+    }
+
+
+def _gpu_duration(
+    image_path: FileData | dict[str, Any], page_index: int = 0
+) -> int:
+    return int(os.getenv("OVISOCR_GPU_DURATION", "300"))
+
+
+_load_model()
+app = gr.Server()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["http://127.0.0.1:4173", "http://localhost:4173"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+@app.api(name="run_ocr", concurrency_limit=1, time_limit=300)
+@spaces.GPU(duration=_gpu_duration)
+def run_ocr(image_path: FileData, page_index: int = 0) -> Iterator[dict[str, Any]]:
+    """Stream one document page per GPU request.
+
+    The browser calls this endpoint sequentially for PDFs. Keeping each page in
+    its own request prevents long documents from exhausting one ZeroGPU lease.
+    """
+    started = time.perf_counter()
+    path = _file_path(image_path)
+    document_type, total_pages = document_info(path)
+    page_index = int(page_index)
+    if page_index < 0 or page_index >= total_pages:
+        raise ValueError(
+            f"Requested PDF page {page_index + 1}, but this document has {total_pages} pages."
+        )
+
+    page_number = page_index + 1
+    page_image = load_document_page(path, document_type, page_index)
+    page_started = time.perf_counter()
+    current = {
+        "page_number": page_number,
+        "markdown": "",
+        "render_markdown": "",
+        "status": "streaming",
+        "elapsed_seconds": 0.0,
+    }
+    yield stream_payload(
+        event="page_start",
+        pages=[current],
+        current_page=page_number,
+        total_pages=total_pages,
+        document_type=document_type,
+        started=started,
+        page_preview=page_preview_data_uri(page_image),
+    )
+
+    markdown = ""
+    for partial in infer_stream(page_image):
+        markdown = partial
+        current = {
+            "page_number": page_number,
+            "markdown": markdown,
+            "render_markdown": stream_safe_markdown(markdown),
+            "status": "streaming",
+            "elapsed_seconds": round(time.perf_counter() - page_started, 3),
+        }
+        yield stream_payload(
+            event="stream",
+            pages=[current],
+            current_page=page_number,
+            total_pages=total_pages,
+            document_type=document_type,
+            started=started,
+        )
+
+    markdown = markdown.strip()
+    if not markdown:
+        raise RuntimeError(f"The model returned an empty result for page {page_number}.")
+    completed_page = {
+        "page_number": page_number,
+        "markdown": markdown,
+        "render_markdown": materialize_bbox_images(markdown, page_image),
+        "status": "complete",
+        "elapsed_seconds": round(time.perf_counter() - page_started, 3),
+    }
+    yield stream_payload(
+        event="page_complete",
+        pages=[completed_page],
+        current_page=page_number,
+        total_pages=total_pages,
+        document_type=document_type,
+        started=started,
+    )
+
+    yield stream_payload(
+        event="complete",
+        pages=[completed_page],
+        current_page=page_number,
+        total_pages=total_pages,
+        document_type=document_type,
+        started=started,
+    )
+
+
+@app.get("/healthz")
+def healthz() -> JSONResponse:
+    return JSONResponse(
+        {
+            "status": "ok",
+            "model": MODEL_ID,
+            "model_source": MODEL_SOURCE,
+            "backend": "mock" if TEST_MODE else "transformers",
+            "loaded": TEST_MODE or (processor is not None and model is not None),
+            "max_pdf_pages": MAX_PDF_PAGES,
+            "root_path": ROOT_PATH,
+            "public_url": PUBLIC_URL,
+        }
+    )
+
+
+if DIST_DIR.is_dir():
+    for route, directory in (
+        ("/assets", DIST_DIR / "assets"),
+        ("/examples", DIST_DIR / "examples"),
+        ("/brand", DIST_DIR / "brand"),
+        ("/vendor", DIST_DIR / "vendor"),
+    ):
+        if directory.is_dir():
+            app.mount(route, StaticFiles(directory=directory), name=route.strip("/").replace("/", "-"))
+
+
+@app.get("/")
+def homepage() -> FileResponse:
+    index_path = DIST_DIR / "index.html"
+    if not index_path.is_file():
+        raise RuntimeError("Frontend build missing. Run `npm run build` before launching app.py.")
+    return FileResponse(index_path)
+
+
+@app.get("/favicon.ico")
+def favicon() -> FileResponse:
+    return FileResponse(DIST_DIR / "favicon.ico")
+
+
+if __name__ == "__main__":
+    app.launch(
+        server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"),
+        server_port=SERVER_PORT,
+        root_path=ROOT_PATH,
+        show_error=True,
+    )

dist/assets/__vite-browser-external-DYxpcVy9-BIHI7g3E.js

@@ -0,0 +1 @@
+const e={};export{e as default};

dist/assets/index-CCa7-Cez.css

@@ -0,0 +1 @@
+@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(data:font/woff2;base64,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) format("woff2-variations");unicode-range:U+0460-052F,U+1C80-1C8A,U+20B4,U+2DE0-2DFF,U+A640-A69F,U+FE2E-FE2F}@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(./manrope-cyrillic-wght-normal-Dvxsihut.woff2) format("woff2-variations");unicode-range:U+0301,U+0400-045F,U+0490-0491,U+04B0-04B1,U+2116}@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(./manrope-greek-wght-normal-DL7QRZyv.woff2) format("woff2-variations");unicode-range:U+0370-0377,U+037A-037F,U+0384-038A,U+038C,U+038E-03A1,U+03A3-03FF}@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(./manrope-vietnamese-wght-normal-usUDDRr7.woff2) format("woff2-variations");unicode-range:U+0102-0103,U+0110-0111,U+0128-0129,U+0168-0169,U+01A0-01A1,U+01AF-01B0,U+0300-0301,U+0303-0304,U+0308-0309,U+0323,U+0329,U+1EA0-1EF9,U+20AB}@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(./manrope-latin-ext-wght-normal-Ch3YOpNY.woff2) format("woff2-variations");unicode-range:U+0100-02BA,U+02BD-02C5,U+02C7-02CC,U+02CE-02D7,U+02DD-02FF,U+0304,U+0308,U+0329,U+1D00-1DBF,U+1E00-1E9F,U+1EF2-1EFF,U+2020,U+20A0-20AB,U+20AD-20C0,U+2113,U+2C60-2C7F,U+A720-A7FF}@font-face{font-family:Manrope Variable;font-style:normal;font-display:swap;font-weight:200 800;src:url(./manrope-latin-wght-normal-DHIcAJRg.woff2) format("woff2-variations");unicode-range:U+0000-00FF,U+0131,U+0152-0153,U+02BB-02BC,U+02C6,U+02DA,U+02DC,U+0304,U+0308,U+0329,U+2000-206F,U+20AC,U+2122,U+2191,U+2193,U+2212,U+2215,U+FEFF,U+FFFD}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(data:font/woff2;base64,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) format("woff2-variations");unicode-range:U+0460-052F,U+1C80-1C8A,U+20B4,U+2DE0-2DFF,U+A640-A69F,U+FE2E-FE2F}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(./jetbrains-mono-cyrillic-wght-normal-D73BlboJ.woff2) format("woff2-variations");unicode-range:U+0301,U+0400-045F,U+0490-0491,U+04B0-04B1,U+2116}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(./jetbrains-mono-greek-wght-normal-Bw9x6K1M.woff2) format("woff2-variations");unicode-range:U+0370-0377,U+037A-037F,U+0384-038A,U+038C,U+038E-03A1,U+03A3-03FF}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(./jetbrains-mono-vietnamese-wght-normal-Bt-aOZkq.woff2) format("woff2-variations");unicode-range:U+0102-0103,U+0110-0111,U+0128-0129,U+0168-0169,U+01A0-01A1,U+01AF-01B0,U+0300-0301,U+0303-0304,U+0308-0309,U+0323,U+0329,U+1EA0-1EF9,U+20AB}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(./jetbrains-mono-latin-ext-wght-normal-DBQx-q_a.woff2) format("woff2-variations");unicode-range:U+0100-02BA,U+02BD-02C5,U+02C7-02CC,U+02CE-02D7,U+02DD-02FF,U+0304,U+0308,U+0329,U+1D00-1DBF,U+1E00-1E9F,U+1EF2-1EFF,U+2020,U+20A0-20AB,U+20AD-20C0,U+2113,U+2C60-2C7F,U+A720-A7FF}@font-face{font-family:JetBrains Mono Variable;font-style:normal;font-display:swap;font-weight:100 800;src:url(./jetbrains-mono-latin-wght-normal-B9CIFXIH.woff2) format("woff2-variations");unicode-range:U+0000-00FF,U+0131,U+0152-0153,U+02BB-02BC,U+02C6,U+02DA,U+02DC,U+0304,U+0308,U+0329,U+2000-206F,U+20AC,U+2122,U+2191,U+2193,U+2212,U+2215,U+FEFF,U+FFFD}:root{font-family:Manrope Variable,Manrope,Arial,sans-serif;color:#080a0d;background:#fff;font-synthesis:none;--ink: #080a0d;--muted: #626872;--subtle: #8a9099;--soft: #f7f8fa;--canvas: #f3f5f8;--line: #e3e6ea;--line-strong: #cfd4da;--blue: #075ce8;--blue-deep: #0349cf;--blue-soft: #edf4ff;--green: #08a972;--red: #d42d3f}*{box-sizing:border-box}html{min-width:320px;background:#fff}body{margin:0;min-width:320px;min-height:100vh;-webkit-font-smoothing:antialiased;text-rendering:optimizeLegibility}button,input{font:inherit}button,a{-webkit-tap-highlight-color:transparent}button{color:inherit}.app-shell{min-height:100vh;background:#fff}.topbar{height:64px;padding:0 28px;display:flex;align-items:center;justify-content:space-between;border-bottom:1px solid var(--line);background:#fffffff5}.brand{display:inline-flex;align-items:center;gap:14px;color:var(--ink);text-decoration:none}.brand img{width:78px;height:auto;display:block}.brand-divider{width:1px;height:28px;background:var(--line)}.product-name{font-size:24px;line-height:1;font-weight:730;letter-spacing:-.045em}.product-name span{color:var(--blue)}.model-chip{margin-left:2px;padding:4px 9px 3px;border:1px solid #b9d3ff;border-radius:999px;color:var(--blue);background:var(--blue-soft);font-family:JetBrains Mono Variable,monospace;font-size:11px;font-weight:650;letter-spacing:-.02em}.topbar nav{display:flex;align-items:center;gap:32px}.topbar nav a{color:#333840;text-decoration:none;font-size:13px;font-weight:560}.topbar nav a:hover{color:var(--blue)}.main-content{width:min(100%,1600px);margin:0 auto;padding:22px 24px 32px}.intro-row{min-height:72px;display:flex;align-items:center;justify-content:space-between;gap:28px}.eyebrow{margin:0 0 5px;color:var(--blue);font-family:JetBrains Mono Variable,monospace;font-size:10px;font-weight:650;line-height:1.2;letter-spacing:.09em;text-transform:uppercase}h1,h2,p{text-wrap:pretty}h1{margin:0;font-size:clamp(21px,2.1vw,30px);line-height:1.16;letter-spacing:-.035em;font-weight:700}.intro-actions{display:flex;align-items:center;gap:22px;flex-shrink:0}.status-line{display:flex;align-items:center;gap:8px;color:var(--muted);font-family:JetBrains Mono Variable,monospace;font-size:10px;white-space:nowrap}.status-mode:before{content:"·";margin-right:8px;color:var(--line-strong)}.status-dot{width:7px;height:7px;border-radius:50%;background:var(--green);box-shadow:0 0 0 3px #08a9721a}.status-line.running .status-dot{background:var(--blue);box-shadow:0 0 0 3px #075ce81a;animation:pulse 1.25s ease-in-out infinite}.status-line.error .status-dot{background:var(--red);box-shadow:0 0 0 3px #d42d3f1a}.primary-button{height:42px;min-width:132px;padding:0 22px;border:0;border-radius:6px;display:inline-flex;align-items:center;justify-content:center;gap:10px;color:#fff;background:var(--blue);box-shadow:0 7px 18px #075ce82e;cursor:pointer;font-size:13px;font-weight:680;transition:background .14s ease,transform .14s ease,box-shadow .14s ease}.primary-button:hover:not(:disabled){background:var(--blue-deep);box-shadow:0 9px 22px #075ce83d;transform:translateY(-1px)}.primary-button:disabled{opacity:.58;cursor:not-allowed;box-shadow:none}.spinner{width:15px;height:15px;border:2px solid rgba(255,255,255,.38);border-top-color:#fff;border-radius:50%;animation:spin .8s linear infinite}.workspace{height:clamp(510px,61vh,650px);min-height:510px;display:grid;grid-template-columns:minmax(360px,.92fr) minmax(520px,1.22fr);border:1px solid var(--line);border-radius:8px;overflow:hidden;background:#fff}.document-panel,.result-panel{min-width:0;min-height:0;display:flex;flex-direction:column}.document-panel{border-right:1px solid var(--line)}.panel-toolbar,.result-toolbar{height:48px;min-height:48px;padding:0 16px;display:flex;align-items:center;justify-content:space-between;border-bottom:1px solid var(--line);background:#fffffffa}.panel-title{min-width:0;display:flex;align-items:center;gap:8px;color:#4a5059;font-size:11px}.panel-title>span{max-width:210px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}.panel-title strong{margin-left:3px;padding:2px 6px;border-radius:4px;color:var(--green);background:#eaf9f3;font-family:JetBrains Mono Variable,monospace;font-size:8px;letter-spacing:.06em}.text-button,.copy-button,.zoom-toolbar button,.toast button{border:0;background:transparent;cursor:pointer}.text-button{display:inline-flex;align-items:center;gap:6px;color:var(--blue);font-size:11px;font-weight:620}.text-button:hover{color:var(--blue-deep)}.document-canvas{position:relative;min-height:0;flex:1;padding:40px 28px 24px;display:flex;align-items:center;justify-content:center;overflow:auto;background:var(--canvas)}.document-canvas:fullscreen{padding:72px;background:var(--canvas)}.document-canvas.dragging{background:var(--blue-soft)}.document-image{display:block;max-width:100%;height:auto;max-height:calc(100% - 8px);object-fit:contain;background:#fff;box-shadow:0 8px 22px #0f172a26,0 0 0 1px #0f172a0f;transition:width .15s ease}.document-pdf{width:min(88%,720px);height:calc(100% - 8px);min-height:360px;border:0;background:#fff;box-shadow:0 8px 22px #0f172a26,0 0 0 1px #0f172a0f}.pdf-fallback{height:100%;min-height:360px;display:flex;flex-direction:column;align-items:center;justify-content:center;gap:8px;color:var(--muted);background:#fff;text-align:center}.pdf-fallback strong{color:var(--ink)}.pdf-fallback span{max-width:260px;font-size:11px}.zoom-toolbar{position:absolute;z-index:2;top:14px;left:14px;height:32px;padding:0 6px;display:flex;align-items:center;gap:2px;border:1px solid var(--line);border-radius:6px;background:#fffffff0;box-shadow:0 2px 7px #0f172a12;-webkit-backdrop-filter:blur(8px);backdrop-filter:blur(8px)}.zoom-toolbar button{width:28px;height:28px;display:grid;place-items:center;border-radius:4px;color:#5e646d}.zoom-toolbar button:hover{color:var(--blue);background:var(--blue-soft)}.zoom-toolbar>span:not(.toolbar-divider){min-width:38px;text-align:center;color:#444a53;font-family:JetBrains Mono Variable,monospace;font-size:10px}.zoom-toolbar .toolbar-divider{width:1px;height:16px;margin:0 4px;background:var(--line)}.drop-overlay{position:absolute;top:12px;right:12px;bottom:12px;left:12px;z-index:4;display:grid;place-items:center;border:2px dashed var(--blue);border-radius:8px;color:var(--blue);background:#edf4fff0;font-weight:700}.upload-empty{width:min(360px,80%);min-height:210px;border:1px dashed #aab6c7;border-radius:8px;display:flex;flex-direction:column;align-items:center;justify-content:center;gap:8px;color:var(--muted);background:#ffffff9e;cursor:pointer}.upload-empty strong{color:var(--ink)}.upload-empty span{font-size:11px}.result-toolbar{padding-left:18px}.tabs{height:100%;display:flex;align-items:stretch;gap:24px}.tabs button{position:relative;padding:0 4px;border:0;color:#666c75;background:transparent;cursor:pointer;font-size:11px;font-weight:600}.tabs button:after{content:"";position:absolute;left:0;right:0;bottom:-1px;height:2px;background:transparent}.tabs button.active{color:var(--blue)}.tabs button.active:after{background:var(--blue)}.copy-button{height:30px;padding:0 11px;border:1px solid #a9c9ff;border-radius:5px;display:inline-flex;align-items:center;gap:7px;color:var(--blue);font-size:10px;font-weight:650}.copy-button:hover:not(:disabled){background:var(--blue-soft)}.copy-button:disabled{opacity:.42;cursor:not-allowed}.page-strip{min-height:40px;padding:5px 12px;display:flex;align-items:center;gap:7px;border-bottom:1px solid var(--line);background:#fbfcfe}.page-pills{min-width:0;flex:1;display:flex;align-items:center;gap:6px;overflow-x:auto;scrollbar-width:thin}.page-pills button,.page-arrow{height:28px;border:1px solid var(--line);border-radius:5px;display:inline-flex;align-items:center;justify-content:center;gap:4px;color:#606771;background:#fff;cursor:pointer;font-family:JetBrains Mono Variable,monospace;font-size:9px;white-space:nowrap}.page-pills button{padding:0 9px}.page-pills button.active{border-color:#9fc2fb;color:var(--blue);background:var(--blue-soft)}.page-pills button:disabled{color:#adb3bc;background:#f6f7f9;cursor:default}.page-pills button svg{color:var(--green)}.page-arrow{width:28px;flex:0 0 auto}.page-arrow:hover:not(:disabled){color:var(--blue);border-color:#b8cff5}.page-arrow:disabled{opacity:.36;cursor:default}.result-content{min-height:0;flex:1;overflow:auto;background:#fff}.result-state{min-height:100%;padding:48px;display:flex;flex-direction:column;align-items:center;justify-content:center;text-align:center;color:var(--muted)}.result-state.empty svg{margin-bottom:12px;color:#b4bbc4}.result-state strong{color:#31363d;font-size:14px}.result-state p{max-width:400px;margin:7px 0 0;font-size:11px;line-height:1.55}.processing-mark{width:46px;height:46px;margin-bottom:15px;display:grid;place-items:center;border:1px solid #cfe0ff;border-radius:50%;background:var(--blue-soft)}.processing-mark span{width:16px;height:16px;border:2px solid #bdd4ff;border-top-color:var(--blue);border-radius:50%;animation:spin .8s linear infinite}.stream-banner{position:sticky;z-index:3;top:0;min-height:34px;padding:7px 16px;display:flex;align-items:center;gap:7px;border-bottom:1px solid #d9e7ff;color:#50627b;background:#f1f6fff5;-webkit-backdrop-filter:blur(8px);backdrop-filter:blur(8px);font-size:10px}.stream-banner>span:first-child{width:7px;height:7px;border-radius:50%;background:var(--blue);animation:pulse 1s ease-in-out infinite}.stream-banner strong{color:var(--blue);font-size:10px}.stream-banner>span:last-child{margin-left:auto;font-family:JetBrains Mono Variable,monospace;font-size:8px}.rendered-markdown{padding:24px 30px 46px;color:#20242a;font-size:14px;line-height:1.68}.rendered-markdown>:first-child{margin-top:0}.rendered-markdown h1,.rendered-markdown h2,.rendered-markdown h3,.rendered-markdown h4{margin:1.5em 0 .56em;color:#111318;line-height:1.25;letter-spacing:-.02em}.rendered-markdown h1{font-size:22px}.rendered-markdown h2{font-size:18px}.rendered-markdown h3{font-size:15px}.rendered-markdown h4{font-size:14px}.rendered-markdown p,.rendered-markdown ul,.rendered-markdown ol{margin:.7em 0}.rendered-markdown a{color:var(--blue)}.rendered-markdown blockquote{margin:1em 0;padding:1px 0 1px 16px;border-left:3px solid #bad3ff;color:#535a64}.rendered-markdown table{width:100%;margin:1.2em 0;border-collapse:collapse;display:table;overflow-x:auto;font-size:12px}.rendered-markdown th,.rendered-markdown td{padding:8px 10px;border:1px solid #dfe3e8;text-align:left;vertical-align:top}.rendered-markdown th{background:#f5f7fa;font-weight:700}.rendered-markdown pre{max-width:100%;padding:14px;overflow-x:auto;border:1px solid #dce3ec;border-radius:6px;white-space:pre;background:#f7f9fc}.rendered-markdown code{font-family:JetBrains Mono Variable,monospace;font-size:.9em}.rendered-markdown .unresolved-image-reference{padding:2px 5px;border:1px solid #dce3ec;border-radius:4px;color:#4b5f7a;background:#f5f7fa;overflow-wrap:anywhere}.rendered-markdown img{display:block;max-width:min(100%,620px);height:auto;margin:18px auto;border-radius:3px}.rendered-markdown mjx-container{max-width:100%;overflow-x:auto;overflow-y:hidden}.rendered-markdown .visual-placeholder{min-height:82px;margin:16px 0;border:1px dashed #b7ccec;border-radius:6px;display:grid;place-items:center;color:#6680a5;background:#f5f8fd;font-family:JetBrains Mono Variable,monospace;font-size:9px}.source-code{min-height:100%;margin:0;padding:24px 28px 48px;overflow:visible;color:#183254;background:#fbfcfe;font-family:JetBrains Mono Variable,monospace;font-size:11px;line-height:1.66;white-space:pre-wrap;overflow-wrap:anywhere;-moz-tab-size:2;tab-size:2}.examples-section{padding-top:18px}.examples-heading{margin-bottom:10px;display:flex;align-items:flex-end;justify-content:space-between}.examples-heading h2{margin:0;font-size:14px;line-height:1.2;letter-spacing:-.02em}.examples-heading>p{margin:0;color:var(--subtle);font-family:JetBrains Mono Variable,monospace;font-size:9px}.example-grid{display:grid;grid-template-columns:repeat(5,minmax(0,1fr));gap:12px}.example-card{min-width:0;padding:0;border:1px solid var(--line);border-radius:7px;overflow:hidden;color:#4a5059;background:#fff;cursor:pointer;text-align:left;transition:border-color .14s ease,transform .14s ease,box-shadow .14s ease}.example-card:hover{border-color:#aac8f9;transform:translateY(-1px);box-shadow:0 7px 17px #0f172a0f}.example-card.active{border-color:var(--blue);box-shadow:0 0 0 1px var(--blue)}.example-image-wrap{position:relative;height:112px;display:block;overflow:hidden;background:var(--soft)}.example-card img{width:100%;height:100%;display:block;object-fit:cover;object-position:top center;transition:transform .18s ease}.example-card:hover img{transform:scale(1.018)}.selected-badge{position:absolute;left:8px;bottom:8px;padding:4px 7px;display:inline-flex;align-items:center;gap:4px;border-radius:4px;color:#fff;background:var(--blue);font-size:9px;font-weight:700}.example-label{min-height:36px;padding:9px 11px;display:flex;align-items:center;font-size:10px;font-weight:610}.file-input{position:fixed;width:1px;height:1px;opacity:0;pointer-events:none}.toast{position:fixed;z-index:30;right:22px;bottom:22px;min-width:220px;min-height:42px;padding:0 12px;display:flex;align-items:center;gap:8px;border:1px solid #243142;border-radius:6px;color:#fff;background:#111926;box-shadow:0 14px 35px #0f172a33;opacity:0;transform:translateY(8px);pointer-events:none;transition:opacity .14s ease,transform .14s ease;font-size:11px}.toast.show{opacity:1;transform:translateY(0);pointer-events:auto}.toast button{width:24px;height:24px;margin-left:auto;display:grid;place-items:center;color:#cbd5e1}button:focus-visible,a:focus-visible{outline:2px solid var(--blue);outline-offset:2px}@keyframes spin{to{transform:rotate(360deg)}}@keyframes pulse{50%{opacity:.42}}@media(max-width:1050px){.main-content{padding-inline:18px}.workspace{grid-template-columns:minmax(330px,.85fr) minmax(470px,1.15fr)}.status-mode{display:none}.example-grid{grid-template-columns:repeat(3,minmax(0,1fr))}.example-image-wrap{height:112px}}@media(max-width:820px){.topbar{height:58px;padding:0 16px}.brand{gap:9px}.brand img{width:62px}.brand-divider{height:22px}.product-name{font-size:19px}.model-chip{display:none}.topbar nav{gap:16px}.topbar nav a{font-size:11px}.topbar nav a:last-child{display:none}.main-content{padding:15px 14px 24px}.intro-row{align-items:flex-start;flex-direction:column;gap:14px;padding-bottom:14px}.intro-actions{width:100%;justify-content:space-between;gap:12px}.status-line{min-width:0;white-space:normal}.primary-button{min-width:116px}.workspace{height:auto;min-height:0;grid-template-columns:1fr}.document-panel{border-right:0;border-bottom:1px solid var(--line)}.document-canvas,.result-content{min-height:430px;max-height:62vh}.example-grid{grid-template-columns:repeat(2,minmax(0,1fr))}.examples-heading>p{display:none}}@media(max-width:480px){.topbar nav{display:none}.intro-actions{align-items:stretch;flex-direction:column}.primary-button{width:100%}.workspace{border-radius:6px}.panel-title>span{max-width:150px}.document-canvas{min-height:370px;padding-inline:16px}.result-content{min-height:400px}.rendered-markdown{padding:22px 18px 40px}.example-grid{gap:10px}.example-image-wrap{height:98px}.toast{left:14px;right:14px;bottom:14px}}@media(prefers-reduced-motion:reduce){*,*:before,*:after{scroll-behavior:auto!important;transition-duration:.01ms!important;animation-duration:.01ms!important;animation-iteration-count:1!important}}

dist/index.html

@@ -0,0 +1,31 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <meta name="theme-color" content="#ffffff" />
+    <meta name="description" content="Streaming document parsing for images and PDFs with OvisOCR2." />
+    <link rel="icon" href="./favicon.ico" />
+    <title>OvisOCR2 — Document to Markdown</title>
+    <script>
+      window.MathJax = {
+        options: {
+          skipHtmlTags: ['script', 'noscript', 'style', 'textarea', 'code', 'annotation', 'annotation-xml']
+        },
+        tex: {
+          inlineMath: [['\\(', '\\)'], ['$', '$']],
+          displayMath: [['\\[', '\\]']],
+          processEscapes: true
+        },
+        chtml: { matchFontHeight: false },
+        startup: { typeset: false }
+      };
+    </script>
+    <script defer src="./vendor/mathjax/tex-chtml-full.js"></script>
+    <script type="module" crossorigin src="./assets/index-5Ju-rOmV.js"></script>
+    <link rel="stylesheet" crossorigin href="./assets/index-CCa7-Cez.css">
+  </head>
+  <body>
+    <div id="root"></div>
+  </body>
+</html>

requirements.txt

@@ -0,0 +1,5 @@
+transformers>=5.13.0,<6
+accelerate>=1.12.0
+torchvision>=0.25.0,<0.27
+Pillow>=10.0.0
+PyMuPDF>=1.24.0,<2