feat: integrate got-ocr2.0 as image reader

.github/workflows/stale.yaml

@@ -0,0 +1,25 @@
+name: Close stale issues
+on:
+  schedule:
+    - cron: "0 0 * * *" # Runs at 00:00 UTC every day
+
+jobs:
+  stale:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/stale@v9
+        with:
+          repo-token: ${{ secrets.GITHUB_TOKEN }}
+
+          stale-issue-message: |
+            Hi folk 😊, it looks like this issue has been inactive for a while. Is there any update or further action
+            needed? If not, we might consider closing it soon to keep our board clean and focused.
+
+            But don't worry, you can reopen it anytime when needed.Thank you for your contributions to Kotaemon🪴
+
+          days-before-issue-stale: 30
+          days-before-issue-close: 3
+          days-before-pr-stale: 90
+          days-before-pr-close: -1
+          exempt-issue-labels: "documentation,tutorial,TODO"
+          operations-per-run: 300 # The maximum number of operations per run, used to control rate limiting.

docker-compose.dev.yml

@@ -0,0 +1,9 @@
+services:
+  kotaemon:
+    volumes:
+      - "./ktem_app_data:/app/ktem_app_data"
+      - "./libs/kotaemon:/app/kotaemon"
+      - "./libs/ktem:/app/ktem"
+      - "./flowsettings.py:/app/flowsettings.py"
+    ports:
+      - "7860:7860"

docker-compose.yml

@@ -0,0 +1,30 @@
+services:
+  kotaemon:
+    build:
+      context: .
+      target: lite
+      dockerfile: Dockerfile
+    env_file: .env
+    environment:
+      - GRADIO_SERVER_NAME=0.0.0.0
+      - GRADIO_SERVER_PORT=7860
+    ports:
+      - "7860:7860"
+    networks:
+      - backend
+  # gocr:
+  #   image: ghcr.io/phv2312/got-ocr2.0:main
+  #   ports:
+  #     - "8881:8881"
+  #   deploy:
+  #     resources:
+  #       reservations:
+  #         devices:
+  #           - driver: nvidia
+  #             count: 1
+  #             capabilities: [gpu]
+  #   networks:
+  #     - backend
+networks:
+  backend:
+    driver: bridge

integration/got-ocr2.md

@@ -0,0 +1,30 @@
+## Extension Manager and GOT-OCR2.0 Loader
+
+## Key Features
+
+### 1. **GOCR2 as Image Reader**
+
+- **GOCR2ImageReader** is a new class designed to read images using the [**GOCR-2.0** OCR engine](https://github.com/Ucas-HaoranWei/GOT-OCR2.0).
+- This reader is initialized with an endpoint that defaults to `http://localhost:8881/ai/infer/` for the OCR service, but can be configured through an environment variable `GOCR2_ENDPOINT` or passed explicitly.
+- It uses exponential backoff retry mechanisms to ensure robustness during API calls.
+- Supports loading image files and extracting their text content, returning structured document data.
+
+#### Setup
+
+- We provide the docker image, with fastapi for serving the GOT-OCR2.0. Pull the image from:
+
+```bash
+docker run -d --gpus all -p 8881:8881 ghcr.io/phv2312/got-ocr2.0:main
+```
+
+- Detail implementation is placed at [ocr_loader.py](/libs/kotaemon/kotaemon/loaders/ocr_loader.py)
+
+### 2. **Extension Manager**
+
+- ExtensionManager allows users to dynamically manage multiple loaders for different file types.
+
+- Users can switch between multiple loaders for the same file extension, such as using the GOCR2ImageReader or a
+  different unstructured data parser for .png files. This provides the flexibility to choose the best-suited loader for the task at hand.
+
+- To change the default loader, go to **Settings**, then **Loader settings**. It displays a grid of extensions and
+  its supported loaders. Any modification will be saved to DB as other settings do.

libs/kotaemon/kotaemon/loaders/__init__.py

@@ -7,7 +7,7 @@
 from .excel_loader import ExcelReader, PandasExcelReader
 from .html_loader import HtmlReader, MhtmlReader
 from .mathpix_loader import MathpixPDFReader
-from .ocr_loader import ImageReader, OCRReader
+from .ocr_loader import GOCR2ImageReader, ImageReader, OCRReader
 from .pdf_loader import PDFThumbnailReader
 from .txt_loader import TxtReader
 from .unstructured_loader import UnstructuredReader
@@ -32,4 +32,5 @@
     "PDFThumbnailReader",
     "WebReader",
     "DoclingReader",
+    "GOCR2ImageReader",
 ]

libs/kotaemon/kotaemon/loaders/ocr_loader.py

@@ -191,3 +191,74 @@ def load_data(
             )
 
         return result
+
+
+class GOCR2ImageReader(BaseReader):
+    default_endpoint = "http://localhost:8881/ai/infer/"
+    """Read Image using GOCR-2.0
+
+    Args:
+        endpoint: URL to GOCR endpoint. If not provided, will look for
+            environment variable `GOCR2_ENDPOINT` or use the default
+            (http://localhost:8881/ai/infer/)
+    """
+
+    def __init__(self, endpoint: Optional[str] = None):
+        """Init the OCR reader with OCR endpoint (FullOCR pipeline)"""
+        super().__init__()
+        self.endpoint = endpoint or os.getenv("GOCR2_ENDPOINT", self.default_endpoint)
+
+    def load_data(
+        self, file_path: Path, extra_info: dict | None = None, **kwargs
+    ) -> List[Document]:
+        """Load data using OCR reader
+
+        Args:
+            file_path (Path): Path to PDF file
+            extra_info (Path): Extra information while inference
+
+        Returns:
+            List[Document]: list of documents extracted from the PDF file
+        """
+
+        @retry(
+            stop=stop_after_attempt(6),
+            wait=wait_exponential(multiplier=20, exp_base=2, min=1, max=1000),
+            after=after_log(logger, logging.WARNING),
+        )
+        def _tenacious_api_post(
+            url: str, file_path: Path, ocr_type: str = "ocr", **kwargs
+        ):
+            with file_path.open("rb") as content:
+                files = {"file": content}
+                data = {"ocr_type": ocr_type}
+                resp = requests.post(url=url, files=files, data=data, **kwargs)
+                resp.raise_for_status()
+            return resp
+
+        file_path = Path(file_path).resolve()
+
+        # call the API from GOCR endpoint
+        if "response_content" in kwargs:
+            # overriding response content if specified
+            ocr_results = kwargs["response_content"]
+        else:
+            # call original API
+            resp = _tenacious_api_post(url=self.endpoint, file_path=file_path)
+            ocr_results = [resp.json()["result"]]
+
+        extra_info = extra_info or {}
+        result = []
+        for ocr_result in ocr_results:
+            metadata = {"file_name": file_path.name, "page_label": 1}
+            if extra_info is not None:
+                metadata.update(extra_info)
+
+            result.append(
+                Document(
+                    content=ocr_result,
+                    metadata=metadata,
+                )
+            )
+
+        return result

libs/ktem/ktem/app.py

@@ -8,6 +8,7 @@
 from ktem.components import reasonings
 from ktem.exceptions import HookAlreadyDeclared, HookNotDeclared
 from ktem.index import IndexManager
+from ktem.loaders.extensions import extension_manager
 from ktem.settings import BaseSettingGroup, SettingGroup, SettingReasoningGroup
 from theflow.settings import settings
 from theflow.utils.modules import import_dotted_string
@@ -63,6 +64,9 @@ def __init__(self):
         self.default_settings = SettingGroup(
             application=BaseSettingGroup(settings=settings.SETTINGS_APP),
             reasoning=SettingReasoningGroup(settings=settings.SETTINGS_REASONING),
+            extension=BaseSettingGroup(
+                settings=extension_manager.generate_gradio_settings()
+            ),
         )
 
         self._callbacks: dict[str, list] = {}