xfmr-zem 0.2.6__tar.gz → 0.2.7__tar.gz

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to this project will be documented in this file.
 
+## [0.2.7] - 2026-02-03
+
+### Added
+- **Configurable OCR Parameters**: Added `scanned_threshold`, `zoom`, and `temp_dir` parameters to the OCR server for finer control over PDF processing.
+
 ## [0.2.6] - 2026-02-03
 
 ### Added

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xfmr-zem
-Version: 0.2.6
+Version: 0.2.7
 Summary: Zem: Unified Data Pipeline Framework (ZenML + NeMo Curator + DataJuicer) for multi-domain processing
 Project-URL: Homepage, https://github.com/OAI-Labs/xfmr-zem
 Project-URL: Repository, https://github.com/OAI-Labs/xfmr-zem

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "xfmr-zem"
-version = "0.2.6"
+version = "0.2.7"
 description = "Zem: Unified Data Pipeline Framework (ZenML + NeMo Curator + DataJuicer) for multi-domain processing"
 readme = "README.md"
 requires-python = ">=3.10,<3.13"

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/src/xfmr_zem/servers/ocr/parameters.yml

@@ -2,3 +2,6 @@
 extract_text:
   engine: "tesseract"
   model_id: null
+  scanned_threshold: 50
+  zoom: 2.0
+  temp_dir: "/tmp"

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/src/xfmr_zem/servers/ocr/server.py

@@ -9,29 +9,39 @@ import io
 # Initialize ZemServer for OCR
 mcp = ZemServer("ocr")
 
-def extract_pdf_pages(file_path: str, engine: str, ocr_engine, model_id: str = None):
+def extract_pdf_pages(
+    file_path: str, 
+    engine: str, 
+    ocr_engine, 
+    scanned_threshold: int = 50, 
+    zoom: float = 2.0, 
+    temp_dir: str = "/tmp"
+):
     """Helper to process PDF pages with optional OCR for scanned content."""
     import fitz  # PyMuPDF
     
     results = []
     doc = fitz.open(file_path)
     
+    # Ensure temp_dir exists
+    os.makedirs(temp_dir, exist_ok=True)
+    
     for page_num in range(len(doc)):
         page = doc[page_num]
         text = page.get_text().strip()
         
         # Determine if we need to OCR (Strategy: text is too short or empty)
-        is_scanned = len(text) < 50 
+        is_scanned = len(text) < scanned_threshold
         
         if is_scanned:
-            logger.info(f"Page {page_num + 1} appears scanned. Running OCR with {engine}...")
+            logger.info(f"Page {page_num + 1} appears scanned (text length: {len(text)}). Running OCR with {engine}...")
             # Render page to image for OCR
-            pix = page.get_pixmap(matrix=fitz.Matrix(2, 2)) # 2x zoom for better OCR
+            pix = page.get_pixmap(matrix=fitz.Matrix(zoom, zoom))
             img_data = pix.tobytes("png")
             img = Image.open(io.BytesIO(img_data))
             
             # Temporary save for engine compatibility (engines expect path)
-            temp_path = f"/tmp/ocr_page_{page_num}.png"
+            temp_path = os.path.join(temp_dir, f"ocr_page_{os.getpid()}_{page_num}.png")
             img.save(temp_path)
             
             try:
@@ -56,7 +66,14 @@ def extract_pdf_pages(file_path: str, engine: str, ocr_engine, model_id: str = N
     return results
 
 @mcp.tool()
-async def extract_text(file_path: str, engine: str = "tesseract", model_id: str = None) -> pd.DataFrame:
+async def extract_text(
+    file_path: str, 
+    engine: str = "tesseract", 
+    model_id: str = None,
+    scanned_threshold: int = 50,
+    zoom: float = 2.0,
+    temp_dir: str = "/tmp"
+) -> pd.DataFrame:
     """
     Extracts text from an image or PDF using the specified OCR engine.
     For PDFs, it will automatically handle scanned pages using the OCR engine.
@@ -65,8 +82,11 @@ async def extract_text(file_path: str, engine: str = "tesseract", model_id: str
         file_path: Path to the image or PDF file.
         engine: The OCR engine to use ("tesseract", "paddle", "huggingface", "viet"). Defaults to "tesseract".
         model_id: Optional model ID for the 'huggingface' engine.
+        scanned_threshold: Min characters required to skip OCR on PDF page. Defaults to 50.
+        zoom: Rendering zoom factor for scanned PDF pages. Defaults to 2.0.
+        temp_dir: Directory for temporary page images. Defaults to "/tmp".
     """
-    logger.info(f"OCR Extraction: {file_path} using {engine}")
+    logger.info(f"OCR Extraction: {file_path} using {engine} (scanned_threshold={scanned_threshold}, zoom={zoom})")
     
     if not os.path.exists(file_path):
         raise FileNotFoundError(f"File not found: {file_path}")
@@ -78,7 +98,14 @@ async def extract_text(file_path: str, engine: str = "tesseract", model_id: str
         # Handle PDF vs Image
         if file_path.lower().endswith(".pdf"):
             logger.info(f"Processing PDF file: {file_path}")
-            data = extract_pdf_pages(file_path, engine, ocr_engine, model_id)
+            data = extract_pdf_pages(
+                file_path, 
+                engine, 
+                ocr_engine, 
+                scanned_threshold=scanned_threshold, 
+                zoom=zoom, 
+                temp_dir=temp_dir
+            )
             df = pd.DataFrame(data)
         else:
             # Process image

{xfmr_zem-0.2.6 → xfmr_zem-0.2.7}/uv.lock

@@ -5534,7 +5534,7 @@ wheels = [
 
 [[package]]
 name = "xfmr-zem"
-version = "0.2.6"
+version = "0.2.7"
 source = { editable = "." }
 dependencies = [
     { name = "click" },