Suya OCR API — vLLM-backed, OpenAI-compatible OCR service

FastAPI service wrapping the Surya-OCR-2 model (datalab-to) served through vLLM:
legacy /v1/api/ai/* endpoints, an OpenAI-compatible /v1/chat/completions endpoint,
a coalescing request batcher, a local OCR CLI, Docker packaging, multilingual
example outputs, and quantization/concurrency benchmarks.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

surya/input/load.py

@@ -0,0 +1,77 @@
+from typing import List
+import PIL
+
+from surya.input.processing import open_pdf, get_page_images
+from surya.logging import get_logger
+from surya.settings import settings
+import os
+import filetype
+from PIL import Image
+
+logger = get_logger()
+
+
+def get_name_from_path(path):
+    return os.path.basename(path).split(".")[0]
+
+
+def load_pdf(pdf_path, page_range: List[int] | None = None, dpi=settings.IMAGE_DPI):
+    doc = open_pdf(pdf_path)
+    last_page = len(doc)
+
+    if page_range:
+        assert all([0 <= page < last_page for page in page_range]), (
+            f"Invalid page range: {page_range}"
+        )
+    else:
+        page_range = list(range(last_page))
+
+    images = get_page_images(doc, page_range, dpi=dpi)
+    doc.close()
+    names = [get_name_from_path(pdf_path) for _ in page_range]
+    return images, names
+
+
+def load_image(image_path):
+    image = Image.open(image_path).convert("RGB")
+    name = get_name_from_path(image_path)
+    return [image], [name]
+
+
+def load_from_file(
+    input_path, page_range: List[int] | None = None, dpi=settings.IMAGE_DPI
+):
+    input_type = filetype.guess(input_path)
+    if input_type and input_type.extension == "pdf":
+        return load_pdf(input_path, page_range, dpi=dpi)
+    else:
+        return load_image(input_path)
+
+
+def load_from_folder(
+    folder_path, page_range: List[int] | None = None, dpi=settings.IMAGE_DPI
+):
+    image_paths = [
+        os.path.join(folder_path, image_name)
+        for image_name in os.listdir(folder_path)
+        if not image_name.startswith(".")
+    ]
+    image_paths = [ip for ip in image_paths if not os.path.isdir(ip)]
+
+    images = []
+    names = []
+    for path in image_paths:
+        extension = filetype.guess(path)
+        if extension and extension.extension == "pdf":
+            image, name = load_pdf(path, page_range, dpi=dpi)
+            images.extend(image)
+            names.extend(name)
+        else:
+            try:
+                image, name = load_image(path)
+                images.extend(image)
+                names.extend(name)
+            except PIL.UnidentifiedImageError:
+                logger.warning(f"Could not load image {path}")
+                continue
+    return images, names

surya/input/processing.py

@@ -0,0 +1,17 @@
+from typing import List
+
+import pypdfium2
+
+from surya.settings import settings
+
+
+def open_pdf(pdf_filepath):
+    return pypdfium2.PdfDocument(pdf_filepath)
+
+
+def get_page_images(doc, indices: List, dpi=settings.IMAGE_DPI):
+    images = [
+        doc[i].render(scale=dpi / 72, draw_annots=False).to_pil() for i in indices
+    ]
+    images = [image.convert("RGB") for image in images]
+    return images