Global PyTorch Config

docs/config.md

@@ -0,0 +1,13 @@
+If you're using the ml dependencies, the config class manages the computational device settings for your project,
+
+### Config
+
+This is a simple wrapper around pytorch. Setting the device will fail if pytorch is not installed.
+
+```python
+import openparse
+
+openparse.config.set_device("cpu")
+```
+
+Note if you're on apple silicon, setting this to `mps` runs significantly slower than on `cpu`.

mkdocs.yml

@@ -79,6 +79,7 @@ nav:
           - Customization: processing/customization.md
   - Serializing Results: serialization.md
   - Visualization: visualization.md
+  - Config: config.md
 
 plugins:
   - search:

src/openparse/__init__.py

@@ -2,7 +2,8 @@
 from openparse.doc_parser import (
     DocumentParser,
 )
-from openparse import processing
+from openparse import processing, version
+from openparse.config import config
 from openparse.schemas import (
     Bbox,
     LineElement,
@@ -13,13 +14,18 @@
 )
 
 __all__ = [
+    # core
     "DocumentParser",
     "Pdf",
+    # Schemas
     "Bbox",
     "LineElement",
     "Node",
     "TableElement",
     "TextElement",
     "TextSpan",
+    # Modules
     "processing",
+    "version",
+    "config",
 ]

src/openparse/config.py

@@ -0,0 +1,42 @@
+from typing import Literal
+
+
+TorchDevice = Literal["cuda", "cpu", "mps"]
+
+
+class Config:
+    def __init__(self):
+        self._device = "cpu"  # Default to CPU
+        self._torch_available = False
+        self._cuda_available = False
+        try:
+            import torch
+
+            self._torch_available = True
+            if torch.cuda.is_available():
+                self._device = "cuda"
+                self._cuda_available = True
+        except ImportError:
+            pass
+
+    def set_device(self, device: TorchDevice) -> None:
+        if not self._torch_available and device == "cuda":
+            raise RuntimeError(
+                "CUDA device requested but torch is not available. Have you installed ml dependencies?"
+            )
+        if not self._cuda_available and device == "cuda":
+            raise RuntimeError("CUDA device requested but CUDA is not available")
+        if device not in ["cuda", "cpu", "mps"]:
+            raise ValueError("Device must be 'cuda', 'cpu' or 'mps'")
+        self._device = device
+
+    def get_device(self):
+        if self._torch_available:
+            import torch
+
+            return torch.device(self._device)
+        else:
+            return self._device
+
+
+config = Config()

src/openparse/tables/table_transformers/ml.py

@@ -10,6 +10,7 @@
     TableTransformerForObjectDetection,  # type: ignore
 )  # type: ignore
 
+from openparse.config import config
 from ..schemas import (
     BBox,
     Size,
@@ -34,12 +35,7 @@
 )
 
 t0 = time.time()
-
-cuda_available = torch.cuda.is_available()
-user_preferred_device = "cuda"
-device = torch.device(
-    "cuda" if cuda_available and user_preferred_device != "cpu" else "cpu"
-)
+device = config.get_device()
 
 
 class MaxResize:

src/openparse/tables/unitable/config.py

@@ -1,10 +1,8 @@
-import torch
 from pydantic import BaseModel
 from pathlib import Path
 import sys
 from openparse import consts
 
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 root = Path(consts.__file__).parent
 
 

src/openparse/tables/unitable/core.py

@@ -6,7 +6,6 @@
 from torchvision import transforms  # type: ignore
 from torch import nn, Tensor  # type: ignore
 
-from .config import device
 from .tokens import VALID_HTML_TOKEN, VALID_BBOX_TOKEN, INVALID_CELL_TOKEN
 from .utils import (
     subsequent_mask,
@@ -27,10 +26,13 @@
     cell_model,
     EncoderDecoder,
 )
+from openparse.config import config
 
 Size = Tuple[int, int]
 BBox = Tuple[int, int, int, int]
 
+device = config.get_device()
+
 
 def _image_to_tensor(image: Image, size: Size) -> Tensor:
     T = transforms.Compose(

src/openparse/tables/unitable/unitable_model.py

@@ -7,14 +7,16 @@
 from torch import nn
 from functools import partial
 
-from .config import device, config
+from .config import config
 from .tabular_transformer import (
     EncoderDecoder,
     ImgLinearBackbone,
     Encoder,
     Decoder,
 )
+from openparse.config import config as global_config
 
+device = global_config.get_device()
 warnings.filterwarnings("ignore")
 
 

src/openparse/version.py

@@ -0,0 +1,43 @@
+OPEN_PARSE_VERSION = "0.5.1"
+
+
+def version_info() -> str:
+    """Return complete version information for OpenParse and its dependencies."""
+    import importlib.metadata as importlib_metadata
+    import platform
+    import sys
+    from pathlib import Path
+
+    python_version = sys.version.split()[0]
+    operating_system = platform.system()
+    os_version = platform.release()
+
+    package_names = {
+        "email-validator",
+        "torch",
+        "torchvision",
+        "transformers",
+        "tokenizers",
+        "PyMuPDF",
+        "pydantic",
+    }
+    related_packages = []
+
+    for dist in importlib_metadata.distributions():
+        name = dist.metadata["Name"]
+        if name in package_names:
+            related_packages.append(f"{name}-{dist.version}")
+
+    info = {
+        "python_version": python_version,
+        "operating_system": operating_system,
+        "os_version": os_version,
+        "open-parse version": OPEN_PARSE_VERSION,
+        "install path": Path(__file__).resolve().parent,
+        "python version": sys.version,
+        "platform": platform.platform(),
+        "related packages": " ".join(related_packages),
+    }
+    return "\n".join(
+        "{:>30} {}".format(k + ":", str(v).replace("\n", " ")) for k, v in info.items()
+    )