Add GoT OCR 2.0

Signed-off-by: Christoph Auer <[email protected]>

Author: cau-git

docling/cli/main.py

@@ -59,6 +59,7 @@
 )
 from docling.datamodel.settings import settings
 from docling.datamodel.vlm_model_specs import (
+    GOT2_TRANSFORMERS,
     GRANITE_VISION_OLLAMA,
     GRANITE_VISION_TRANSFORMERS,
     SMOLDOCLING_MLX,
@@ -621,6 +622,8 @@ def convert(  # noqa: C901
                 pipeline_options.vlm_options = GRANITE_VISION_TRANSFORMERS
             elif vlm_model == VlmModelType.GRANITE_VISION_OLLAMA:
                 pipeline_options.vlm_options = GRANITE_VISION_OLLAMA
+            elif vlm_model == VlmModelType.GOT_OCR_2:
+                pipeline_options.vlm_options = GOT2_TRANSFORMERS
             elif vlm_model == VlmModelType.SMOLDOCLING:
                 pipeline_options.vlm_options = SMOLDOCLING_TRANSFORMERS
                 if sys.platform == "darwin":

docling/datamodel/vlm_model_specs.py

@@ -194,6 +194,26 @@
     temperature=0.0,
 )
 
+# GoT 2.0
+GOT2_TRANSFORMERS = InlineVlmOptions(
+    repo_id="stepfun-ai/GOT-OCR-2.0-hf",
+    prompt="",
+    response_format=ResponseFormat.MARKDOWN,
+    inference_framework=InferenceFramework.TRANSFORMERS,
+    transformers_prompt_style=TransformersPromptStyle.RAW,
+    transformers_model_type=TransformersModelType.AUTOMODEL_IMAGETEXTTOTEXT,
+    supported_devices=[
+        AcceleratorDevice.CPU,
+        AcceleratorDevice.CUDA,
+        #    AcceleratorDevice.MPS,
+    ],
+    scale=2.0,
+    temperature=0.0,
+    stop_strings=["<|im_end|>"],
+    #extra_generation_config={"format": True},
+)
+
+
 # Gemma-3
 GEMMA3_12B_MLX = InlineVlmOptions(
     repo_id="mlx-community/gemma-3-12b-it-bf16",
@@ -215,6 +235,8 @@
     temperature=0.0,
 )
 
+# Dolphin
+
 DOLPHIN_TRANSFORMERS = InlineVlmOptions(
     repo_id="ByteDance/Dolphin",
     prompt="<s>Read text in the image. <Answer/>",
@@ -238,3 +260,4 @@ class VlmModelType(str, Enum):
     GRANITE_VISION = "granite_vision"
     GRANITE_VISION_VLLM = "granite_vision_vllm"
     GRANITE_VISION_OLLAMA = "granite_vision_ollama"
+    GOT_OCR_2 = "got_ocr_2"