xfmr-zem 0.2.4__tar.gz → 0.2.5__tar.gz

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/.gitignore

@@ -55,3 +55,8 @@ Thumbs.db
 # Project specific
 outputs/
 cache/
+
+# Binary Models (OCR) - Should be downloaded separately
+src/xfmr_zem/servers/ocr/deepdoc_vietocr/onnx/*.onnx
+src/xfmr_zem/servers/ocr/deepdoc_vietocr/vietocr/weight/*.pth
+src/xfmr_zem/servers/ocr/deepdoc_vietocr/vietocr/weight/*.onnx

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/AGENTS.md

@@ -42,6 +42,18 @@
     - Providers: **Ollama** (local default), **OpenAI**.
 - **Sinks Server** (`src/xfmr_zem/servers/sinks/server.py`):
     - Tools: `to_huggingface`, `to_vector_db`.
+### 5. Unstructured Server (`servers/unstructured`)
+Advanced document parsing for multimodal data ingestion.
+- `parse_document`: Convert PDF, DOCX, HTML to structured text.
+- `extract_tables`: Specifically isolate and extract table data from documents.
+
+### 6. OCR Server (`servers/ocr`)
+Unified OCR processing with multiple engine support (SOLID Strategy Pattern).
+- `extract_text`: Extract text from images using different engines:
+    - `tesseract`: Lightweight and fast.
+    - `paddle`: Medium weight, high accuracy.
+    - `qwen`: Heavy Vision-Language Model (Qwen3-VL-8B) for state-of-the-art OCR.
+    - `viet`: Specialized Vietnamese OCR using built-in `deepdoc_vietocr` pipeline. Optimized for Vietnamese diacritics and document layout reconstruction.
 
 ## Orchestration & Concurrency
 

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to this project will be documented in this file.
 
+## [0.2.5] - 2026-02-03
+
+### Added
+- **Vietnamese OCR**: Integrated specialized Deep-ocr DocumentPipeline for high-accuracy Vietnamese text extraction with layout analysis.
+- **Improved OCR Dependencies**: Added `pdfplumber`, `ruamel.yaml`, and `cachetools` to resolve OCR server tool errors.
+
 ## [0.2.4] - 2026-02-02
 
 ### Fixed

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xfmr-zem
-Version: 0.2.4
+Version: 0.2.5
 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
@@ -22,8 +22,10 @@ Requires-Dist: numpy>=1.24.0
 Requires-Dist: pandas>=2.0.0
 Requires-Dist: pyarrow>=15.0.0
 Requires-Dist: pydantic>=2.0.0
+Requires-Dist: python-magic>=0.4.27
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: rich>=13.0.0
+Requires-Dist: unstructured[all-docs]>=0.16.0
 Requires-Dist: zenml[local,server]>=0.75.0
 Provides-Extra: all
 Requires-Dist: nemo-curator>=0.6.0; extra == 'all'
@@ -39,6 +41,22 @@ Requires-Dist: pytest>=7.0.0; extra == 'dev'
 Requires-Dist: ruff>=0.1.0; extra == 'dev'
 Provides-Extra: nemo
 Requires-Dist: nemo-curator>=0.6.0; extra == 'nemo'
+Provides-Extra: ocr
+Requires-Dist: cachetools>=5.0.0; extra == 'ocr'
+Requires-Dist: einops; extra == 'ocr'
+Requires-Dist: onnxruntime>=1.16.0; extra == 'ocr'
+Requires-Dist: opencv-python>=4.8.0; extra == 'ocr'
+Requires-Dist: paddleocr>=2.7.0; extra == 'ocr'
+Requires-Dist: paddlepaddle>=2.6.0; extra == 'ocr'
+Requires-Dist: pdfplumber>=0.11.0; extra == 'ocr'
+Requires-Dist: pillow>=10.0.0; extra == 'ocr'
+Requires-Dist: pyclipper; extra == 'ocr'
+Requires-Dist: pytesseract>=0.3.10; extra == 'ocr'
+Requires-Dist: ruamel-yaml>=0.17.0; extra == 'ocr'
+Requires-Dist: shapely; extra == 'ocr'
+Requires-Dist: torch==2.5.1; extra == 'ocr'
+Requires-Dist: torchvision==0.20.1; extra == 'ocr'
+Requires-Dist: transformers>=4.40.0; extra == 'ocr'
 Provides-Extra: zenml
 Requires-Dist: zenml>=0.75.0; extra == 'zenml'
 Description-Content-Type: text/markdown

xfmr_zem-0.2.5/parameters.yml

@@ -0,0 +1,7 @@
+# Parameters for Zem Pipeline
+# This file allows overriding tool arguments at runtime
+
+ocr:
+  extract_text:
+    engine: "huggingface"
+    model_id: "Qwen/Qwen2-VL-2B-Instruct"

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "xfmr-zem"
-version = "0.2.4"
+version = "0.2.5"
 description = "Zem: Unified Data Pipeline Framework (ZenML + NeMo Curator + DataJuicer) for multi-domain processing"
 readme = "README.md"
 requires-python = ">=3.10,<3.13"
@@ -25,6 +25,8 @@ dependencies = [
     "dask-cuda>=24.0.0",
     "ftfy>=6.3.1",
     "pydantic>=2.0.0",
+    "unstructured[all-docs]>=0.16.0",
+    "python-magic>=0.4.27",
 ]
 
 [project.urls]
@@ -48,6 +50,23 @@ all = [
     "nemo-curator>=0.6.0",
     "py-data-juicer>=1.0.0",
 ]
+ocr = [
+    "pytesseract>=0.3.10",
+    "paddleocr>=2.7.0",
+    "paddlepaddle>=2.6.0",
+    "transformers>=4.40.0",
+    "torch==2.5.1",
+    "torchvision==0.20.1",
+    "pillow>=10.0.0",
+    "onnxruntime>=1.16.0",
+    "opencv-python>=4.8.0",
+    "shapely",
+    "pyclipper",
+    "einops",
+    "pdfplumber>=0.11.0",
+    "ruamel.yaml>=0.17.0",
+    "cachetools>=5.0.0",
+]
 dev = [
     "pytest>=7.0.0",
     "pytest-cov>=4.0.0",

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/cli.py

@@ -173,8 +173,19 @@ def _print_static_operators():
 @main.command()
 @click.argument("config_file", type=click.Path(exists=True))
 @click.option("--params", "-p", type=click.Path(exists=True), help="Path to custom parameters.yml")
-def run(config_file, params):
+@click.option("--verbose", "-v", is_flag=True, help="Enable verbose/debug logging")
+def run(config_file, params, verbose):
     """Run a pipeline from a YAML configuration file"""
+    # Configure logging based on verbosity
+    if verbose:
+        os.environ["ZEM_VERBOSE"] = "1"
+        logger.remove()
+        logger.add(sys.stderr, level="DEBUG", format="<green>{time:HH:mm:ss}</green> | <level>{level: <8}</level> | <cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - <level>{message}</level>")
+        console.print("[bold yellow]Verbose mode enabled - DEBUG logging active[/bold yellow]")
+    else:
+        logger.remove()
+        logger.add(sys.stderr, level="INFO", format="<level>{message}</level>")
+
     abs_config = os.path.abspath(config_file)
     console.print(f"[bold green]Starting Pipeline:[/bold green] {abs_config}")
     if params:
@@ -239,12 +250,18 @@ def preview(artifact_id, id2, limit, sample):
                 return pd.DataFrame(lines)
         elif isinstance(d, list): return pd.DataFrame(d)
         elif isinstance(d, pd.DataFrame): return d
-        return None
+        elif isinstance(d, dict): return pd.DataFrame([d])
+        return d
 
     try:
         df1 = load_art_df(artifact_id)
         if df1 is None:
-            console.print("[bold red]Error:[/bold red] Could not load artifact as tabular data.")
+            console.print("[bold red]Error:[/bold red] Artifact is empty or could not be loaded.")
+            return
+
+        if not isinstance(df1, pd.DataFrame):
+            console.print(f"[bold blue]Artifact Preview (Type: {type(df1).__name__}):[/bold blue]")
+            console.print(str(df1))
             return
 
         if id2:
@@ -290,6 +307,18 @@ def preview(artifact_id, id2, limit, sample):
         console.print(f"[bold red]Error previewing artifact:[/bold red] {e}")
 
 
+@main.group()
+def ocr():
+    """OCR related commands (Installation, etc.)"""
+    pass
+
+@ocr.command(name="install")
+def ocr_install():
+    """Install OCR model weights (ONNX/PTH)"""
+    from xfmr_zem.servers.ocr.install_models import main as install_main
+    install_main()
+
+
 if __name__ == "__main__":
     main()
 

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/client.py

@@ -4,6 +4,7 @@ import yaml
 from pathlib import Path
 from zenml import pipeline
 from .zenml_wrapper import mcp_generic_step
+from loguru import logger
 import os
 import sys
 
@@ -39,6 +40,17 @@ class PipelineClient:
                     params.update(yaml.safe_load(f) or {})
         return params
 
+    def _flatten_params(self, d: Dict[str, Any], prefix: str = "") -> Dict[str, Any]:
+        """Flatten nested dictionary into dot-notation keys."""
+        items = []
+        for k, v in d.items():
+            new_key = f"{prefix}.{k}" if prefix else k
+            if isinstance(v, dict):
+                items.extend(self._flatten_params(v, new_key).items())
+            else:
+                items.append((new_key, v))
+        return dict(items)
+
     def _load_config_dict(self, path: Path) -> Dict[str, Any]:
         """Load YAML config and perform substitution."""
         with open(path, "r") as f:
@@ -54,8 +66,13 @@ class PipelineClient:
             custom_params = self._load_params(self.params_path)
             self.params.update(custom_params)
             
+        # Flatten params for template substitution
+        flat_params = self._flatten_params(self.params)
+        
         content = raw_content
-        for key, value in self.params.items():
+        # Use reversed sorted keys to avoid partial replacements (e.g. ocr before ocr.engine)
+        for key in sorted(flat_params.keys(), key=len, reverse=True):
+            value = flat_params[key]
             content = content.replace(f"{{{{ {key} }}}}", str(value))
             content = content.replace(f"{{{{{key}}}}}", str(value))
             
@@ -96,6 +113,11 @@ class PipelineClient:
                     server_specific_params[key] = value
             
             env["ZEM_PARAMETERS"] = yaml.dump(server_specific_params)
+            
+            # Pass verbose flag to subprocess
+            if os.environ.get("ZEM_VERBOSE"):
+                env["ZEM_VERBOSE"] = "1"
+            
             configs[name] = {
                 "command": sys.executable,
                 "args": [str(abs_path)],
@@ -146,11 +168,35 @@ class PipelineClient:
                         
                     srv, tool = key.split(".")
                     
-                    if not step_alias:
-                        step_alias = step_def[key].get("name")
-                    
                     step_alias = step_alias or f"{srv}.{tool}.{i}"
-                    tool_args = step_def[key].get("input", {}) or {}
+                    
+                    val = step_def[key]
+                    if isinstance(val, dict):
+                        if "input" in val:
+                            tool_args = val.get("input", {}) or {}
+                        else:
+                            # Use everything except 'name' as tool_args
+                            tool_args = {k: v for k, v in val.items() if k != "name"}
+                    else:
+                        tool_args = {}
+                    
+                # Standardized Parameter Injection: 
+                # Merge parameters from the 'parameters' section.
+                # Priority: Step-specific args > parameters.<srv>.<tool> > parameters.<srv>
+                srv_params = self.params.get(srv, {})
+                if isinstance(srv_params, dict):
+                    # 1. Server-wide defaults
+                    for k, v in srv_params.items():
+                        if k != tool and not isinstance(v, dict) and k not in tool_args:
+                            tool_args[k] = v
+                    
+                    # 2. Tool-specific overrides
+                    tool_params = srv_params.get(tool, {})
+                    if isinstance(tool_params, dict):
+                        for k, v in tool_params.items():
+                            if k not in tool_args:
+                                tool_args[k] = v
+                     
 
                 # Smart Parallelization & DAG Logic:
                 # 1. By default, a step is a root (None) unless it has no 'data' input,
@@ -161,8 +207,13 @@ class PipelineClient:
                 has_explicit_data = "data" in tool_args
 
                 if not has_explicit_data:
-                    # No data provided? Inherit from the last executed step to keep simple sequences working
-                    current_prev_output = last_output
+                    # Smart Source Detection: If a step has 'file_path', 'url', etc.,
+                    # it's likely a primary ingestion step and shouldn't inherit 'data' from the previous step.
+                    source_keys = {"file_path", "url", "uri", "path"}
+                    is_source = any(k in tool_args for k in source_keys)
+                    
+                    if not is_source:
+                        current_prev_output = last_output
                 else:
                     # Data provided? Check if it's a reference or raw data
                     for k, v in list(tool_args.items()):
@@ -174,7 +225,7 @@ class PipelineClient:
                                     del tool_args[k]
                                 else:
                                     # Limitation: ZenML doesn't materialize artifacts nested in dicts
-                                    print(f"[Warning] Tool argument '{k}' uses a step reference '{v}'. "
+                                    logger.warning(f" Tool argument '{k}' uses a step reference '{v}'. "
                                           "Currently, only the 'data' field supports cross-step dependencies. "
                                           "This value will be passed as a raw string.")
                             else:

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/server.py

@@ -3,7 +3,9 @@ from typing import Any, Callable, Dict, List, Optional, Union
 import yaml
 from pathlib import Path
 from fastmcp import FastMCP
+from loguru import logger
 import inspect
+import functools
 
 class ZemServer(FastMCP):
     """
@@ -22,9 +24,24 @@ class ZemServer(FastMCP):
         self.parameter_file = parameter_file
         self.parameters = {}
         
-        # 1. Load from file
+        # Configure logging based on ZEM_VERBOSE
+        import os
+        import sys
+        if os.environ.get("ZEM_VERBOSE"):
+            logger.remove()
+            logger.add(sys.stderr, level="DEBUG", format="<green>{time:HH:mm:ss}</green> | <level>{level: <8}</level> | <cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - <level>{message}</level>")
+        
+        # 1. Load from file or auto-detect in server directory
         if parameter_file:
             self.load_parameters(parameter_file)
+        else:
+            # Auto-detect parameters.yml in the same directory as the server script
+            import inspect
+            caller_frame = inspect.stack()[1]
+            caller_file = caller_frame.filename
+            auto_path = Path(caller_file).parent / "parameters.yml"
+            if auto_path.exists():
+                self.load_parameters(str(auto_path))
             
         # 2. Override with env params (from PipelineClient)
         import os
@@ -35,7 +52,7 @@ class ZemServer(FastMCP):
                 if isinstance(env_params, dict):
                     self._merge_parameters(env_params)
             except Exception as e:
-                print(f"Error loading ZEM_PARAMETERS: {e}")
+                logger.error(f"Error loading ZEM_PARAMETERS: {e}")
 
     def load_parameters(self, file_path: str) -> Dict[str, Any]:
         """Load parameters from YAML file and merge them."""
@@ -79,8 +96,8 @@ class ZemServer(FastMCP):
             else:
                 target[k] = v
 
-    # Removed custom tool decorator to fix multiple values for argument 'name' error
-    # Inherit directly from FastMCP.tool
+    # NOTE: Parameter injection is handled by PipelineClient._merge_parameters
+    # and the tool decorator is inherited from FastMCP
 
     def get_data(self, data: Any) -> List[Dict[str, Any]]:
         """

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/servers/data_juicer/server.py

@@ -9,7 +9,7 @@ from loguru import logger
 logger.remove()
 logger.add(sys.stderr, level="INFO")
 
-server = ZemServer("data_juicer", parameter_file=os.path.join(os.path.dirname(__file__), "parameter.yaml"))
+server = ZemServer("data_juicer", parameter_file=os.path.join(os.path.dirname(__file__), "parameters.yml"))
 
 @server.tool()
 def clean_content(

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/servers/instruction_gen/server.py

@@ -10,7 +10,7 @@ from loguru import logger
 logger.remove()
 logger.add(sys.stderr, level="INFO")
 
-server = ZemServer("instruction", parameter_file=os.path.join(os.path.dirname(__file__), "parameter.yaml"))
+server = ZemServer("instruction", parameter_file=os.path.join(os.path.dirname(__file__), "parameters.yml"))
 
 @server.tool()
 def generate_qa_pairs(

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/servers/io/server.py

@@ -7,7 +7,7 @@ from loguru import logger
 import sys
 
 # Initialize server
-server = ZemServer("io", parameter_file=os.path.join(os.path.dirname(__file__), "parameter.yaml"))
+server = ZemServer("io", parameter_file=os.path.join(os.path.dirname(__file__), "parameters.yml"))
 
 @server.tool()
 def load_jsonl(path: str, return_reference: bool = False) -> Any:

xfmr_zem-0.2.5/src/xfmr_zem/servers/llm/parameters.yml

@@ -0,0 +1,10 @@
+# Default parameters for LLM Server
+mask_pii:
+  provider: "ollama"
+
+classify_domain:
+  provider: "ollama"
+  categories:
+    - Tech
+    - Finance
+    - Legal

{xfmr_zem-0.2.4 → xfmr_zem-0.2.5}/src/xfmr_zem/servers/nemo_curator/server.py

@@ -10,7 +10,7 @@ from loguru import logger
 logger.remove()
 logger.add(sys.stderr, level="INFO")
 
-server = ZemServer("nemo", parameter_file=os.path.join(os.path.dirname(__file__), "parameter.yaml"))
+server = ZemServer("nemo", parameter_file=os.path.join(os.path.dirname(__file__), "parameters.yml"))
 
 @server.tool()
 def normalize(

xfmr_zem-0.2.5/src/xfmr_zem/servers/ocr/deepdoc_vietocr/__init__.py

@@ -0,0 +1,90 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import io
+import sys
+import threading
+import pdfplumber
+
+from .ocr import OCR
+from .recognizer import Recognizer
+from .layout_recognizer import LayoutRecognizerDocLayoutYOLO as LayoutRecognizer
+from .table_structure_recognizer import TableStructureRecognizer
+# from .engine import VietDocEngine
+
+# New Phase-Based Architecture
+from .pipeline import DocumentPipeline
+from .phases import (
+    LayoutAnalysisPhase,
+    TextDetectionPhase,
+    TextRecognitionPhase,
+    PostProcessingPhase,
+    DocumentReconstructionPhase,
+)
+from .implementations import (
+    DocLayoutYOLOAnalyzer,
+    PaddleOCRTextDetector,
+    VietOCRRecognizer,
+    SVTRv2Recognizer,
+    LandingAIRecognizer,
+    VietnameseTextPostProcessor,
+    SmartMarkdownReconstruction,
+    create_default_pipeline,
+    create_svtrv2_pipeline,
+    create_experimental_pipeline,
+)
+
+
+LOCK_KEY_pdfplumber = "global_shared_lock_pdfplumber"
+if LOCK_KEY_pdfplumber not in sys.modules:
+    sys.modules[LOCK_KEY_pdfplumber] = threading.Lock()
+
+
+# Removed init_in_out
+
+
+__all__ = [
+    # Legacy API (backward compatibility)
+    "OCR",
+    "Recognizer",
+    "LayoutRecognizer",
+    "TableStructureRecognizer",
+    # "VietDocEngine",
+    # "init_in_out",
+
+    # New Phase-Based Architecture
+    "DocumentPipeline",
+
+    # Abstract Phase Interfaces
+    "LayoutAnalysisPhase",
+    "TextDetectionPhase",
+    "TextRecognitionPhase",
+    "PostProcessingPhase",
+    "DocumentReconstructionPhase",
+
+    # Concrete Implementations
+    "DocLayoutYOLOAnalyzer",
+    "PaddleOCRTextDetector",
+    "VietOCRRecognizer",
+    "SVTRv2Recognizer",
+    "LandingAIRecognizer",
+    "VietnameseTextPostProcessor",
+    "SmartMarkdownReconstruction",
+
+    # Factory Functions
+    "create_default_pipeline",
+    "create_svtrv2_pipeline",
+    "create_experimental_pipeline",
+]