diff --git a/src/instructlab/sdg/generate_data.py b/src/instructlab/sdg/generate_data.py
index cf65ae14..2aac5028 100644
--- a/src/instructlab/sdg/generate_data.py
+++ b/src/instructlab/sdg/generate_data.py
@@ -15,6 +15,7 @@
 # instructlab - All of these need to go away (other than sdg) - issue #6
 from xdg_base_dirs import xdg_data_dirs, xdg_data_home
 import openai
+import yaml
 
 # First Party
 # pylint: disable=ungrouped-imports
@@ -220,6 +221,23 @@ def _sdg_init(ctx, pipeline):
     data_dirs = [os.path.join(xdg_data_home(), "instructlab", "sdg")]
     data_dirs.extend(os.path.join(dir, "instructlab", "sdg") for dir in xdg_data_dirs())
 
+    docling_model_path = None
+    sdg_models_path = docling_model_path
+    for d in data_dirs:
+        if os.path.exists(os.path.join(d, "models")):
+            sdg_models_path = os.path.join(d, "models")
+            break
+
+    if sdg_models_path is not None:
+        try:
+            with open(
+                os.path.join(sdg_models_path, "config.yaml"), "r", encoding="utf-8"
+            ) as file:
+                config = yaml.safe_load(file)
+                docling_model_path = config["models"][0]["path"]
+        except (FileNotFoundError, NotADirectoryError, PermissionError) as e:
+            logger.warning(f"unable to read docling models path from config.yaml {e}")
+
     for d in data_dirs:
         pipeline_path = os.path.join(d, "pipelines", pipeline)
         if os.path.exists(pipeline_path):
@@ -251,6 +269,7 @@ def load_pipeline(yaml_basename):
         load_pipeline("knowledge.yaml"),
         load_pipeline("freeform_skills.yaml"),
         load_pipeline("grounded_skills.yaml"),
+        docling_model_path,
     )
 
 
@@ -363,8 +382,8 @@ def generate_data(
         max_num_tokens=max_num_tokens,
     )
 
-    knowledge_pipe, freeform_skills_pipe, grounded_skills_pipe = _sdg_init(
-        ctx, pipeline
+    knowledge_pipe, freeform_skills_pipe, grounded_skills_pipe, docling_model_path = (
+        _sdg_init(ctx, pipeline)
     )
 
     # Make sure checkpointing is disabled (we don't want this pipeline to load checkpoints from the main pipeline)
@@ -392,6 +411,7 @@ def generate_data(
             chunk_word_count,
             document_output_dir,
             model_name,
+            docling_model_path=docling_model_path,
         )
 
         if not samples:
diff --git a/src/instructlab/sdg/utils/chunkers.py b/src/instructlab/sdg/utils/chunkers.py
index 50fd692c..0a72e3c4 100644
--- a/src/instructlab/sdg/utils/chunkers.py
+++ b/src/instructlab/sdg/utils/chunkers.py
@@ -90,6 +90,7 @@ def __new__(
         server_ctx_size=4096,
         chunk_word_count=1024,
         tokenizer_model_name: str | None = None,
+        docling_model_path: str | None = None,
     ):
         """Insantiate the appropriate chunker for the provided document
 
@@ -145,6 +146,7 @@ def __new__(
                 output_dir,
                 chunk_word_count,
                 tokenizer_model_name,
+                docling_model_path=docling_model_path,
             )
 
     @staticmethod
@@ -219,6 +221,7 @@ def __init__(
         output_dir: Path,
         chunk_word_count: int,
         tokenizer_model_name="mistralai/Mixtral-8x7B-Instruct-v0.1",
+        docling_model_path=None,
     ):
         self.document_paths = document_paths
         self.filepaths = filepaths
@@ -231,6 +234,7 @@ def __init__(
         )
 
         self.tokenizer = self.create_tokenizer(tokenizer_model_name)
+        self.docling_model_path = docling_model_path
 
     def chunk_documents(self) -> List:
         """Semantically chunk PDF documents.
@@ -247,6 +251,8 @@ def chunk_documents(self) -> List:
         if self.document_paths == []:
             return []
 
+<<<<<<< HEAD
+<<<<<<< HEAD
         model_artifacts_path = StandardPdfPipeline.download_models_hf()
         pipeline_options = PdfPipelineOptions(
             artifacts_path=model_artifacts_path,
@@ -256,6 +262,27 @@ def chunk_documents(self) -> List:
         if ocr_options is not None:
             pipeline_options.do_ocr = True
             pipeline_options.ocr_options = ocr_options
+=======
+        if not self.docling_model_path.exists():
+=======
+        if self.docling_model_path is None:
+<<<<<<< HEAD
+>>>>>>> f8f6959 (Update src/instructlab/sdg/utils/chunkers.py)
+            logger.info(
+                f"Docling models not found on disk, downloading models..."
+            )
+=======
+            logger.info("Docling models not found on disk, downloading models...")
+>>>>>>> 16c6f45 (Address mypy issues and small typos)
+            self.docling_model_path = StandardPdfPipeline.download_models_hf()
+        else:
+            logger.info("Found the docling models")
+
+        pipeline_options = PdfPipelineOptions(artifacts_path=self.docling_model_path)
+
+        # Keep OCR models on the CPU instead of GPU
+        pipeline_options.ocr_options.use_gpu = False
+>>>>>>> 1b984e0 (Rebase)
         converter = DocumentConverter(
             format_options={
                 InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
diff --git a/src/instructlab/sdg/utils/taxonomy.py b/src/instructlab/sdg/utils/taxonomy.py
index a6f9b381..00743d93 100644
--- a/src/instructlab/sdg/utils/taxonomy.py
+++ b/src/instructlab/sdg/utils/taxonomy.py
@@ -416,6 +416,7 @@ def _knowledge_leaf_node_to_samples(
     chunk_word_count,
     document_output_dir,
     model_name,
+    docling_model_path=None,
 ):
     chunker = DocumentChunker(
         leaf_node=leaf_node,
@@ -424,6 +425,7 @@ def _knowledge_leaf_node_to_samples(
         server_ctx_size=server_ctx_size,
         chunk_word_count=chunk_word_count,
         tokenizer_model_name=model_name,
+        docling_model_path=docling_model_path,
     )
     chunks = chunker.chunk_documents()
 
@@ -453,6 +455,7 @@ def leaf_node_to_samples(
     chunk_word_count,
     document_output_dir,
     model_name,
+    docling_model_path=None,
 ):
     if not leaf_node:
         return []
@@ -464,5 +467,6 @@ def leaf_node_to_samples(
             chunk_word_count,
             document_output_dir,
             model_name,
+            docling_model_path,
         )
     return _skill_leaf_node_to_samples(leaf_node)
diff --git a/tests/conftest.py b/tests/conftest.py
index 80d61903..ed3fd8c4 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -6,6 +6,8 @@
 
 # Standard
 from unittest import mock
+import pathlib
+import typing
 
 # Third Party
 from datasets import Dataset
@@ -17,6 +19,14 @@
 # Local
 from .taxonomy import MockTaxonomy
 
+TESTS_PATH = pathlib.Path(__file__).parent.absolute()
+
+
+@pytest.fixture
+def testdata_path() -> typing.Generator[pathlib.Path, None, None]:
+    """Path to local test data directory"""
+    yield TESTS_PATH / "testdata"
+
 
 def get_ctx(**kwargs) -> PipelineContext:
     kwargs.setdefault("client", mock.MagicMock())
diff --git a/tests/test_generate_data.py b/tests/test_generate_data.py
index f382a351..0d04a80f 100644
--- a/tests/test_generate_data.py
+++ b/tests/test_generate_data.py
@@ -20,7 +20,7 @@
 import yaml
 
 # First Party
-from instructlab.sdg.generate_data import _context_init, generate_data
+from instructlab.sdg.generate_data import _context_init, _sdg_init, generate_data
 from instructlab.sdg.llmblock import LLMBlock
 from instructlab.sdg.pipeline import PipelineContext
 
@@ -548,3 +548,37 @@ def test_context_init_batch_size_optional():
         batch_num_workers=32,
     )
     assert ctx.batch_size == 20
+
+
+def test_sdg_init_docling_path_config_found(testdata_path):
+    with patch.dict(os.environ):
+        os.environ["XDG_DATA_HOME"] = str(testdata_path.joinpath("mock_xdg_data_dir"))
+        ctx = _context_init(
+            None,
+            "mixtral",
+            "foo.bar",
+            1,
+            "/checkpoint/dir",
+            1,
+            batch_size=20,
+            batch_num_workers=32,
+        )
+        _, _, _, docling_model_path = _sdg_init(ctx, "full")
+        assert docling_model_path == "/mock/docling-models"
+
+
+def test_sdg_init_docling_path_config_not_found(testdata_path):
+    with patch.dict(os.environ):
+        os.environ["XDG_DATA_HOME"] = str(testdata_path.joinpath("nonexistent_dir"))
+        ctx = _context_init(
+            None,
+            "mixtral",
+            "foo.bar",
+            1,
+            "/checkpoint/dir",
+            1,
+            batch_size=20,
+            batch_num_workers=32,
+        )
+        _, _, _, docling_model_path = _sdg_init(ctx, "full")
+        assert docling_model_path is None
diff --git a/tests/testdata/mock_xdg_data_dir/instructlab/sdg/models/config.yaml b/tests/testdata/mock_xdg_data_dir/instructlab/sdg/models/config.yaml
new file mode 100644
index 00000000..657cfdf3
--- /dev/null
+++ b/tests/testdata/mock_xdg_data_dir/instructlab/sdg/models/config.yaml
@@ -0,0 +1,4 @@
+models:
+- path: /mock/docling-models
+  source: https://huggingface.co/ds4sd/docling-models
+  revision: main