feat add glm-ocr core

app/core/config.py

@@ -3,9 +3,8 @@
 from functools import lru_cache
 from pathlib import Path
 
-from pydantic_settings import BaseSettings, SettingsConfigDict
 import torch
-from typing import Optional
+from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
 class Settings(BaseSettings):
@@ -48,21 +47,25 @@ class Settings(BaseSettings):
     is_padding: bool = True
     padding_ratio: float = 0.1
 
+    max_tokens: int = 4096
+
     # Model Paths
-    pp_doclayout_model_dir: Optional[str] = "/home/yoge/.cache/modelscope/hub/models/PaddlePaddle/PP-DocLayoutV3"
+    pp_doclayout_model_dir: str | None = (
+        "/home/yoge/.cache/modelscope/hub/models/PaddlePaddle/PP-DocLayoutV3"
+    )
 
     # Image Processing
     max_image_size_mb: int = 10
     image_padding_ratio: float = 0.1  # 10% on each side = 20% total expansion
 
-    device: torch.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")  # cuda:0 or cpu
+    device: torch.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
     # Server Settings
     host: str = "0.0.0.0"
     port: int = 8053
 
     # Logging Settings
-    log_dir: Optional[str] = None  # Defaults to /app/logs in container or ./logs locally
+    log_dir: str | None = None  # Defaults to /app/logs in container or ./logs locally
     log_level: str = "INFO"  # DEBUG, INFO, WARNING, ERROR, CRITICAL
 
     @property

app/core/dependencies.py

@@ -2,7 +2,7 @@
 
 from app.services.image_processor import ImageProcessor
 from app.services.layout_detector import LayoutDetector
-from app.services.ocr_service import OCRService, MineruOCRService, GLMOCRService
+from app.services.ocr_service import GLMOCREndToEndService
 from app.services.converter import Converter
 from app.core.config import get_settings
 
@@ -31,40 +31,17 @@ def get_image_processor() -> ImageProcessor:
     return ImageProcessor()
 
 
-def get_ocr_service() -> OCRService:
-    """Get an OCR service instance."""
-    return OCRService(
-        vl_server_url=get_settings().paddleocr_vl_url,
-        layout_detector=get_layout_detector(),
-        image_processor=get_image_processor(),
-        converter=get_converter(),
-    )
-
-
 def get_converter() -> Converter:
     """Get a DOCX converter instance."""
     return Converter()
 
 
-def get_mineru_ocr_service() -> MineruOCRService:
-    """Get a MinerOCR service instance."""
+def get_glmocr_endtoend_service() -> GLMOCREndToEndService:
+    """Get end-to-end GLM-OCR service (layout detection + per-region OCR)."""
     settings = get_settings()
-    api_url = getattr(settings, "miner_ocr_api_url", "http://127.0.0.1:8000/file_parse")
-    glm_ocr_url = getattr(settings, "glm_ocr_url", "http://localhost:8002/v1")
-    return MineruOCRService(
-        api_url=api_url,
-        converter=get_converter(),
-        image_processor=get_image_processor(),
-        glm_ocr_url=glm_ocr_url,
-    )
-
-
-def get_glmocr_service() -> GLMOCRService:
-    """Get a GLM OCR service instance."""
-    settings = get_settings()
-    glm_ocr_url = getattr(settings, "glm_ocr_url", "http://127.0.0.1:8002/v1")
-    return GLMOCRService(
-        vl_server_url=glm_ocr_url,
+    return GLMOCREndToEndService(
+        vl_server_url=settings.glm_ocr_url,
         image_processor=get_image_processor(),
         converter=get_converter(),
+        layout_detector=get_layout_detector(),
     )