add supported_devices

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

docling/cli/main.py

@@ -22,6 +22,7 @@ from docling.backend.docling_parse_v2_backend import DoclingParseV2DocumentBacke
 from docling.backend.docling_parse_v4_backend import DoclingParseV4DocumentBackend
 from docling.backend.pdf_backend import PdfDocumentBackend
 from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import (
     ConversionStatus,
     FormatToExtensions,
@@ -30,8 +31,6 @@ from docling.datamodel.base_models import (
 )
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     EasyOcrOptions,
     OcrOptions,
     PaginatedPipelineOptions,

docling/datamodel/pipeline_options.py

@@ -1,6 +1,4 @@
 import logging
-import os
-import re
 from enum import Enum
 from pathlib import Path
 from typing import Any, ClassVar, Dict, List, Literal, Optional, Union
@@ -10,13 +8,11 @@ from pydantic import (
     BaseModel,
     ConfigDict,
     Field,
-    field_validator,
-    model_validator,
 )
-from pydantic_settings import BaseSettings, SettingsConfigDict
 from typing_extensions import deprecated
 
 # Import the following for backwards compatibility
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.pipeline_options_vlm_model import (
     ApiVlmOptions,
     InferenceFramework,
@@ -34,64 +30,6 @@ from docling.datamodel.vlm_model_spec import (
 _log = logging.getLogger(__name__)
 
 
-class AcceleratorDevice(str, Enum):
-    """Devices to run model inference"""
-
-    AUTO = "auto"
-    CPU = "cpu"
-    CUDA = "cuda"
-    MPS = "mps"
-
-
-class AcceleratorOptions(BaseSettings):
-    model_config = SettingsConfigDict(
-        env_prefix="DOCLING_", env_nested_delimiter="_", populate_by_name=True
-    )
-
-    num_threads: int = 4
-    device: Union[str, AcceleratorDevice] = "auto"
-    cuda_use_flash_attention2: bool = False
-
-    @field_validator("device")
-    def validate_device(cls, value):
-        # "auto", "cpu", "cuda", "mps", or "cuda:N"
-        if value in {d.value for d in AcceleratorDevice} or re.match(
-            r"^cuda(:\d+)?$", value
-        ):
-            return value
-        raise ValueError(
-            "Invalid device option. Use 'auto', 'cpu', 'mps', 'cuda', or 'cuda:N'."
-        )
-
-    @model_validator(mode="before")
-    @classmethod
-    def check_alternative_envvars(cls, data: Any) -> Any:
-        r"""
-        Set num_threads from the "alternative" envvar OMP_NUM_THREADS.
-        The alternative envvar is used only if it is valid and the regular envvar is not set.
-
-        Notice: The standard pydantic settings mechanism with parameter "aliases" does not provide
-        the same functionality. In case the alias envvar is set and the user tries to override the
-        parameter in settings initialization, Pydantic treats the parameter provided in __init__()
-        as an extra input instead of simply overwriting the evvar value for that parameter.
-        """
-        if isinstance(data, dict):
-            input_num_threads = data.get("num_threads")
-            # Check if to set the num_threads from the alternative envvar
-            if input_num_threads is None:
-                docling_num_threads = os.getenv("DOCLING_NUM_THREADS")
-                omp_num_threads = os.getenv("OMP_NUM_THREADS")
-                if docling_num_threads is None and omp_num_threads is not None:
-                    try:
-                        data["num_threads"] = int(omp_num_threads)
-                    except ValueError:
-                        _log.error(
-                            "Ignoring misformatted envvar OMP_NUM_THREADS '%s'",
-                            omp_num_threads,
-                        )
-        return data
-
-
 class BaseOptions(BaseModel):
     """Base class for options."""
 

docling/datamodel/pipeline_options_vlm_model.py

@@ -1,9 +1,11 @@
 from enum import Enum
-from typing import Any, Dict, Literal
+from typing import Any, Dict, List, Literal
 
 from pydantic import AnyUrl, BaseModel
 from typing_extensions import deprecated
 
+from docling.datamodel.accelerator_options import AcceleratorDevice
+
 
 class BaseVlmOptions(BaseModel):
     kind: str
@@ -35,6 +37,12 @@ class InlineVlmOptions(BaseVlmOptions):
     inference_framework: InferenceFramework
     response_format: ResponseFormat
 
+    supported_devices: List[AcceleratorDevice] = [
+        AcceleratorDevice.CPU,
+        AcceleratorDevice.CUDA,
+        AcceleratorDevice.MPS,
+    ]
+
     scale: float = 2.0
 
     temperature: float = 0.0

docling/datamodel/vlm_model_spec.py

@@ -5,6 +5,7 @@ from pydantic import (
     AnyUrl,
 )
 
+from docling.datamodel.accelerator_options import AcceleratorDevice
 from docling.datamodel.pipeline_options_vlm_model import (
     ApiVlmOptions,
     InferenceFramework,
@@ -21,6 +22,7 @@ SMOLDOCLING_MLX = InlineVlmOptions(
     prompt="Convert this page to docling.",
     response_format=ResponseFormat.DOCTAGS,
     inference_framework=InferenceFramework.MLX,
+    supported_devices=[AcceleratorDevice.MPS],
     scale=2.0,
     temperature=0.0,
 )
@@ -30,6 +32,11 @@ SMOLDOCLING_TRANSFORMERS = InlineVlmOptions(
     prompt="Convert this page to docling.",
     response_format=ResponseFormat.DOCTAGS,
     inference_framework=InferenceFramework.TRANSFORMERS_VISION2SEQ,
+    supported_devices=[
+        AcceleratorDevice.CPU,
+        AcceleratorDevice.CUDA,
+        AcceleratorDevice.MPS,
+    ],
     scale=2.0,
     temperature=0.0,
 )
@@ -40,6 +47,7 @@ GRANITE_VISION_TRANSFORMERS = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare MarkDown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.TRANSFORMERS_VISION2SEQ,
+    supported_devices=[AcceleratorDevice.CPU, AcceleratorDevice.CUDA],
     scale=2.0,
     temperature=0.0,
 )
@@ -60,6 +68,7 @@ PIXTRAL_12B_TRANSFORMERS = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare markdown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.TRANSFORMERS_VISION2SEQ,
+    supported_devices=[AcceleratorDevice.CPU, AcceleratorDevice.CUDA],
     scale=2.0,
     temperature=0.0,
 )
@@ -69,6 +78,7 @@ PIXTRAL_12B_MLX = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare markdown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.MLX,
+    supported_devices=[AcceleratorDevice.MPS],
     scale=2.0,
     temperature=0.0,
 )
@@ -77,8 +87,10 @@ PIXTRAL_12B_MLX = InlineVlmOptions(
 PHI4_TRANSFORMERS = InlineVlmOptions(
     repo_id="microsoft/Phi-4-multimodal-instruct",
     prompt="Convert this page to MarkDown. Do not miss any text and only output the bare markdown",
+    trust_remote_code=True,
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.TRANSFORMERS_CAUSALLM,
+    supported_devices=[AcceleratorDevice.CPU, AcceleratorDevice.CUDA],
     scale=2.0,
     temperature=0.0,
 )
@@ -89,6 +101,7 @@ QWEN25_VL_3B_MLX = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare markdown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.MLX,
+    supported_devices=[AcceleratorDevice.MPS],
     scale=2.0,
     temperature=0.0,
 )
@@ -99,6 +112,7 @@ GEMMA3_12B_MLX = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare markdown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.MLX,
+    supported_devices=[AcceleratorDevice.MPS],
     scale=2.0,
     temperature=0.0,
 )
@@ -108,6 +122,7 @@ GEMMA3_27B_MLX = InlineVlmOptions(
     prompt="Convert this page to markdown. Do not miss any text and only output the bare markdown!",
     response_format=ResponseFormat.MARKDOWN,
     inference_framework=InferenceFramework.MLX,
+    supported_devices=[AcceleratorDevice.MPS],
     scale=2.0,
     temperature=0.0,
 )

docling/models/base_ocr_model.py

@@ -11,9 +11,10 @@ from PIL import Image, ImageDraw
 from rtree import index
 from scipy.ndimage import binary_dilation, find_objects, label
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import AcceleratorOptions, OcrOptions
+from docling.datamodel.pipeline_options import OcrOptions
 from docling.datamodel.settings import settings
 from docling.models.base_model import BaseModelWithOptions, BasePageModel
 

docling/models/code_formula_model.py

@@ -16,8 +16,8 @@ from docling_core.types.doc.labels import CodeLanguageLabel
 from PIL import Image, ImageOps
 from pydantic import BaseModel
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import ItemAndImageEnrichmentElement
-from docling.datamodel.pipeline_options import AcceleratorOptions
 from docling.models.base_model import BaseItemAndImageEnrichmentModel
 from docling.utils.accelerator_utils import decide_device
 

docling/models/document_picture_classifier.py

@@ -13,7 +13,7 @@ from docling_core.types.doc import (
 from PIL import Image
 from pydantic import BaseModel
 
-from docling.datamodel.pipeline_options import AcceleratorOptions
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.models.base_model import BaseEnrichmentModel
 from docling.utils.accelerator_utils import decide_device
 

docling/models/easyocr_model.py

@@ -9,11 +9,10 @@ import numpy
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import BoundingRectangle, TextCell
 
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     EasyOcrOptions,
     OcrOptions,
 )

docling/models/layout_model.py

@@ -10,9 +10,9 @@ from docling_core.types.doc import DocItemLabel
 from docling_ibm_models.layoutmodel.layout_predictor import LayoutPredictor
 from PIL import Image
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import BoundingBox, Cluster, LayoutPrediction, Page
 from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import AcceleratorOptions
 from docling.datamodel.settings import settings
 from docling.models.base_model import BasePageModel
 from docling.utils.accelerator_utils import decide_device

docling/models/ocr_mac_model.py

@@ -8,10 +8,10 @@ from typing import Optional, Type
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import BoundingRectangle, TextCell
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     OcrMacOptions,
     OcrOptions,
 )

docling/models/picture_description_api_model.py

@@ -5,8 +5,8 @@ from typing import Optional, Type, Union
 
 from PIL import Image
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     PictureDescriptionApiOptions,
     PictureDescriptionBaseOptions,
 )

docling/models/picture_description_base_model.py

@@ -13,8 +13,8 @@ from docling_core.types.doc.document import (  # TODO: move import to docling_co
 )
 from PIL import Image
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     PictureDescriptionBaseOptions,
 )
 from docling.models.base_model import (

docling/models/picture_description_vlm_model.py

@@ -4,8 +4,8 @@ from typing import Optional, Type, Union
 
 from PIL import Image
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     PictureDescriptionBaseOptions,
     PictureDescriptionVlmOptions,
 )

docling/models/rapid_ocr_model.py

@@ -7,11 +7,10 @@ import numpy
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import BoundingRectangle, TextCell
 
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     OcrOptions,
     RapidOcrOptions,
 )

docling/models/table_structure_model.py

@@ -13,11 +13,10 @@ from docling_core.types.doc.page import (
 from docling_ibm_models.tableformer.data_management.tf_predictor import TFPredictor
 from PIL import ImageDraw
 
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import Page, Table, TableStructurePrediction
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     TableFormerMode,
     TableStructureOptions,
 )

docling/models/tesseract_ocr_cli_model.py

@@ -13,10 +13,10 @@ import pandas as pd
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import TextCell
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     OcrOptions,
     TesseractCliOcrOptions,
 )

docling/models/tesseract_ocr_model.py

@@ -7,10 +7,10 @@ from typing import Iterable, Optional, Type
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import TextCell
 
+from docling.datamodel.accelerator_options import AcceleratorOptions
 from docling.datamodel.base_models import Page
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorOptions,
     OcrOptions,
     TesseractOcrOptions,
 )

docling/models/vlm_models_inline/hf_transformers_causallm_model.py

@@ -4,11 +4,11 @@ from collections.abc import Iterable
 from pathlib import Path
 from typing import Optional
 
-from docling.datamodel.base_models import Page, VlmPrediction
-from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import (
+from docling.datamodel.accelerator_options import (
     AcceleratorOptions,
 )
+from docling.datamodel.base_models import Page, VlmPrediction
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options_vlm_model import InlineVlmOptions
 from docling.models.base_model import BasePageModel
 from docling.models.hf_vlm_model import HuggingFaceVlmModel
@@ -39,8 +39,10 @@ class HuggingFaceVlmModel_AutoModelForCausalLM(BasePageModel):
                 GenerationConfig,
             )
 
-            self.device = decide_device(accelerator_options.device)
-            self.device = HuggingFaceVlmModel.map_device_to_cpu_if_mlx(self.device)
+            self.device = decide_device(
+                accelerator_options.device,
+                supported_devices=vlm_options.supported_devices,
+            )
             _log.debug(f"Available device for VLM: {self.device}")
 
             self.use_cache = vlm_options.use_kv_cache

docling/models/vlm_models_inline/hf_transformers_vision2seq_model.py

@@ -4,11 +4,11 @@ from collections.abc import Iterable
 from pathlib import Path
 from typing import Optional
 
-from docling.datamodel.base_models import Page, VlmPrediction
-from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import (
+from docling.datamodel.accelerator_options import (
     AcceleratorOptions,
 )
+from docling.datamodel.base_models import Page, VlmPrediction
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options_vlm_model import InlineVlmOptions
 from docling.models.base_model import BasePageModel
 from docling.models.hf_vlm_model import HuggingFaceVlmModel
@@ -38,9 +38,11 @@ class HuggingFaceVlmModel_AutoModelForVision2Seq(BasePageModel):
                 BitsAndBytesConfig,
             )
 
-            self.device = decide_device(accelerator_options.device)
-            self.device = HuggingFaceVlmModel.map_device_to_cpu_if_mlx(self.device)
-            _log.debug(f"Available device for HuggingFace VLM: {self.device}")
+            self.device = decide_device(
+                accelerator_options.device,
+                supported_devices=vlm_options.supported_devices,
+            )
+            _log.debug(f"Available device for VLM: {self.device}")
 
             self.use_cache = vlm_options.use_kv_cache
             self.max_new_tokens = vlm_options.max_new_tokens

docling/models/vlm_models_inline/mlx_model.py

@@ -4,11 +4,11 @@ from collections.abc import Iterable
 from pathlib import Path
 from typing import Optional
 
-from docling.datamodel.base_models import Page, VlmPrediction, VlmPredictionToken
-from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import (
+from docling.datamodel.accelerator_options import (
     AcceleratorOptions,
 )
+from docling.datamodel.base_models import Page, VlmPrediction, VlmPredictionToken
+from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options_vlm_model import InlineVlmOptions
 from docling.models.base_model import BasePageModel
 from docling.models.hf_vlm_model import HuggingFaceVlmModel

docling/utils/accelerator_utils.py

@@ -1,13 +1,16 @@
 import logging
+from typing import List, Optional
 
 import torch
 
-from docling.datamodel.pipeline_options import AcceleratorDevice
+from docling.datamodel.accelerator_options import AcceleratorDevice
 
 _log = logging.getLogger(__name__)
 
 
-def decide_device(accelerator_device: str) -> str:
+def decide_device(
+    accelerator_device: str, supported_devices: Optional[List[AcceleratorDevice]] = None
+) -> str:
     r"""
     Resolve the device based on the acceleration options and the available devices in the system.
 
@@ -20,6 +23,18 @@ def decide_device(accelerator_device: str) -> str:
     has_cuda = torch.backends.cuda.is_built() and torch.cuda.is_available()
     has_mps = torch.backends.mps.is_built() and torch.backends.mps.is_available()
 
+    if supported_devices is not None:
+        if has_cuda and AcceleratorDevice.CUDA not in supported_devices:
+            _log.info(
+                f"Removing CUDA from available devices because it is not in {supported_devices=}"
+            )
+            has_cuda = False
+        if has_mps and AcceleratorDevice.MPS not in supported_devices:
+            _log.info(
+                f"Removing MPS from available devices because it is not in {supported_devices=}"
+            )
+            has_mps = False
+
     if accelerator_device == AcceleratorDevice.AUTO.value:  # Handle 'auto'
         if has_cuda:
             device = "cuda:0"

docs/examples/custom_convert.py

@@ -3,10 +3,9 @@ import logging
 import time
 from pathlib import Path
 
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import InputFormat
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     PdfPipelineOptions,
 )
 from docling.document_converter import DocumentConverter, PdfFormatOption

docs/examples/run_with_accelerator.py

@@ -1,9 +1,8 @@
 from pathlib import Path
 
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import InputFormat
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     PdfPipelineOptions,
 )
 from docling.datamodel.settings import settings

tests/test_e2e_conversion.py

@@ -1,9 +1,10 @@
 from pathlib import Path
 
 from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
+from docling.datamodel.accelerator_options import AcceleratorDevice
 from docling.datamodel.base_models import InputFormat
 from docling.datamodel.document import ConversionResult
-from docling.datamodel.pipeline_options import AcceleratorDevice, PdfPipelineOptions
+from docling.datamodel.pipeline_options import PdfPipelineOptions
 from docling.document_converter import DocumentConverter, PdfFormatOption
 
 from .test_data_gen_flag import GEN_TEST_DATA

tests/test_e2e_ocr_conversion.py

@@ -3,10 +3,10 @@ from pathlib import Path
 from typing import List, Tuple
 
 from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
+from docling.datamodel.accelerator_options import AcceleratorDevice
 from docling.datamodel.base_models import InputFormat
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
     EasyOcrOptions,
     OcrMacOptions,
     OcrOptions,

tests/test_options.py

@@ -7,11 +7,10 @@ from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
 from docling.backend.docling_parse_v2_backend import DoclingParseV2DocumentBackend
 from docling.backend.docling_parse_v4_backend import DoclingParseV4DocumentBackend
 from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
+from docling.datamodel.accelerator_options import AcceleratorDevice, AcceleratorOptions
 from docling.datamodel.base_models import ConversionStatus, InputFormat, QualityGrade
 from docling.datamodel.document import ConversionResult
 from docling.datamodel.pipeline_options import (
-    AcceleratorDevice,
-    AcceleratorOptions,
     PdfPipelineOptions,
     TableFormerMode,
 )