[Feat] Add /concatenate-pages for PP-StructureV3 and PaddleOCR-VL services

deploy/hps/sdk/common/server.sh

@@ -14,10 +14,12 @@ rm -rf "${MODEL_REPO_DIR}"
 cp -r model_repo "${MODEL_REPO_DIR}"
 
 find "${MODEL_REPO_DIR}" -mindepth 1 -maxdepth 1 -type d -print0 | while IFS= read -r -d '' dir_; do
-    if [ "${PADDLEX_HPS_DEVICE_TYPE}" = 'gpu' ]; then
-        cp -f "${dir_}/config_gpu.pbtxt" "${dir_}/config.pbtxt"
-    else
-        cp -f "${dir_}/config_cpu.pbtxt" "${dir_}/config.pbtxt"
+    if [ ! -f "${dir_}/config.pbtxt" ]; then
+        if [ "${PADDLEX_HPS_DEVICE_TYPE:-gpu}" = 'gpu' ]; then
+            cp -f "${dir_}/config_gpu.pbtxt" "${dir_}/config.pbtxt"
+        else
+            cp -f "${dir_}/config_cpu.pbtxt" "${dir_}/config.pbtxt"
+        fi
     fi
 done
 

deploy/hps/sdk/pipelines/3d_bev_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/OCR/server/model_repo/ocr/1/model.py

@@ -20,6 +20,7 @@
 from paddlex_hps_server import (
     BaseTritonPythonModel,
     app_common,
+    logging,
     protocol,
     schemas,
     utils,
@@ -167,10 +168,13 @@ def run_batch(self, inputs, log_ids, batch_id):
 
     def _group_inputs(self, inputs):
         def _to_hashable(obj):
-            if isinstance(obj, list):
-                return tuple(obj)
-            elif isinstance(obj, dict):
-                return tuple(sorted(obj.items()))
+            if isinstance(obj, dict):
+                return tuple(
+                    (_to_hashable(k), _to_hashable(v))
+                    for k, v in sorted(obj.items(), key=lambda x: repr(x[0]))
+                )
+            elif isinstance(obj, list):
+                return tuple(_to_hashable(x) for x in obj)
             else:
                 return obj
 
@@ -231,12 +235,20 @@ def _preprocess(self, input, log_id):
             else self.app_config.visualize
         )
 
-        file_bytes = utils.get_raw_bytes(input.file)
-        images, data_info = utils.file_to_images(
-            file_bytes,
-            file_type,
-            max_num_imgs=self.context["max_num_input_imgs"],
-        )
+        try:
+            file_bytes = utils.get_raw_bytes(input.file)
+            images, data_info = utils.file_to_images(
+                file_bytes,
+                file_type,
+                max_num_imgs=self.context["max_num_input_imgs"],
+            )
+        except Exception as e:
+            logging.error("Failed to get input file bytes: %s", e)
+            return protocol.create_aistudio_output_without_result(
+                422,
+                "Input file is invalid",
+                log_id=log_id,
+            )
 
         return images, data_info, visualize_enabled
 

deploy/hps/sdk/pipelines/OCR/version.txt

@@ -0,0 +1 @@
+0.2.5

deploy/hps/sdk/pipelines/PP-ChatOCRv3-doc/version.txt

@@ -0,0 +1 @@
+0.3.2

deploy/hps/sdk/pipelines/PP-ChatOCRv4-doc/version.txt

@@ -0,0 +1 @@
+0.4.2

deploy/hps/sdk/pipelines/PP-DocTranslation/version.txt

@@ -0,0 +1 @@
+0.1.2

deploy/hps/sdk/pipelines/PP-ShiTuV2/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/PP-StructureV3/server/model_repo/layout-parsing/1/model.py

@@ -19,6 +19,7 @@
 from paddlex_hps_server import (
     BaseTritonPythonModel,
     app_common,
+    logging,
     protocol,
     schemas,
     utils,
@@ -200,10 +201,13 @@ def run_batch(self, inputs, log_ids, batch_id):
 
     def _group_inputs(self, inputs):
         def _to_hashable(obj):
-            if isinstance(obj, list):
-                return tuple(obj)
-            elif isinstance(obj, dict):
-                return tuple(sorted(obj.items()))
+            if isinstance(obj, dict):
+                return tuple(
+                    (_to_hashable(k), _to_hashable(v))
+                    for k, v in sorted(obj.items(), key=lambda x: repr(x[0]))
+                )
+            elif isinstance(obj, list):
+                return tuple(_to_hashable(x) for x in obj)
             else:
                 return obj
 
@@ -286,20 +290,32 @@ def _preprocess(self, input, log_id):
             else self.app_config.visualize
         )
 
-        file_bytes = utils.get_raw_bytes(input.file)
-        images, data_info = utils.file_to_images(
-            file_bytes,
-            file_type,
-            max_num_imgs=self.context["max_num_input_imgs"],
-        )
+        try:
+            file_bytes = utils.get_raw_bytes(input.file)
+            images, data_info = utils.file_to_images(
+                file_bytes,
+                file_type,
+                max_num_imgs=self.context["max_num_input_imgs"],
+            )
+        except Exception as e:
+            logging.error("Failed to get input file bytes: %s", e)
+            return protocol.create_aistudio_output_without_result(
+                422,
+                "Input file is invalid",
+                log_id=log_id,
+            )
 
         return images, data_info, visualize_enabled
 
     def _postprocess(self, images, data_info, visualize_enabled, preds, log_id, input):
         layout_parsing_results: List[Dict[str, Any]] = []
         for i, (img, item) in enumerate(zip(images, preds)):
             pruned_res = app_common.prune_result(item.json["res"])
-            md_data = item.markdown
+            # XXX
+            md_data = item._to_markdown(
+                pretty=input.prettifyMarkdown,
+                show_formula_number=input.showFormulaNumber,
+            )
             md_text = md_data["markdown_texts"]
             md_imgs = app_common.postprocess_images(
                 md_data["markdown_images"],

deploy/hps/sdk/pipelines/PP-StructureV3/version.txt

@@ -0,0 +1 @@
+0.3.5

deploy/hps/sdk/pipelines/PaddleOCR-VL/server/model_repo/concatenate-pages/1/model.py

@@ -0,0 +1,67 @@
+# Copyright (c) 2025 PaddlePaddle 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.
+
+from paddlex_hps_server import (
+    BaseTritonPythonModel,
+    app_common,
+    schemas,
+)
+
+
+class TritonPythonModel(BaseTritonPythonModel):
+    @property
+    def pipeline_creation_kwargs(self):
+        return {"initial_predictor": False}
+
+    def get_input_model_type(self):
+        return schemas.paddleocr_vl.ConcatenatePagesRequest
+
+    def get_result_model_type(self):
+        return schemas.paddleocr_vl.ConcatenatePagesResult
+
+    def run(self, input, log_id):
+        def _to_original_result(pruned_res):
+            orig_res = {"res": pruned_res}
+            return orig_res
+
+        original_results = []
+        markdown_images = {}
+        for i, page in enumerate(input.pages):
+            orig_res = _to_original_result(page.prunedResult)
+            original_results.append(orig_res)
+            markdown_images.update(page.markdownImages)
+
+        concatenated_result = self.pipeline.concatenate_pages(
+            original_results,
+            merge_table=input.mergeTable,
+            title_level=input.titleLevel,
+        )
+
+        layout_parsing_result = {}
+        layout_parsing_result["prunedResult"] = app_common.prune_result(
+            concatenated_result.json["res"]
+        )
+        # XXX
+        md_data = concatenated_result._to_markdown(
+            pretty=input.prettifyMarkdown,
+            show_formula_number=input.showFormulaNumber,
+        )
+        layout_parsing_result["markdown"] = dict(
+            text=md_data["markdown_texts"],
+            images=markdown_images,
+        )
+
+        return schemas.paddleocr_vl.ConcatenatePagesResult(
+            layoutParsingResult=layout_parsing_result,
+        )

deploy/hps/sdk/pipelines/PaddleOCR-VL/server/model_repo/concatenate-pages/config.pbtxt

@@ -0,0 +1,22 @@
+backend: "python"
+max_batch_size: 1
+input [
+  {
+    name: "input"
+    data_type: TYPE_STRING
+    dims: [ 1 ]
+  }
+]
+output [
+  {
+    name: "output"
+    data_type: TYPE_STRING
+    dims: [ 1 ]
+  }
+]
+instance_group [
+  {
+      count: 1
+      kind: KIND_CPU
+  }
+]

deploy/hps/sdk/pipelines/PaddleOCR-VL/server/model_repo/layout-parsing/1/model.py

@@ -19,6 +19,7 @@
 from paddlex_hps_server import (
     BaseTritonPythonModel,
     app_common,
+    logging,
     protocol,
     schemas,
     utils,
@@ -123,7 +124,10 @@ def run_batch(self, inputs, log_ids, batch_id):
                             ].useDocOrientationClassify,
                             use_doc_unwarping=inputs_g[0].useDocUnwarping,
                             use_layout_detection=inputs_g[0].useLayoutDetection,
+                            use_polygon_points=inputs_g[0].usePolygonPoints,
                             use_chart_recognition=inputs_g[0].useChartRecognition,
+                            use_seal_recognition=inputs_g[0].useSealRecognition,
+                            use_ocr_for_image_block=inputs_g[0].useOcrForImageBlock,
                             layout_threshold=inputs_g[0].layoutThreshold,
                             layout_nms=inputs_g[0].layoutNms,
                             layout_unclip_ratio=inputs_g[0].layoutUnclipRatio,
@@ -138,6 +142,7 @@ def run_batch(self, inputs, log_ids, batch_id):
                             max_new_tokens=inputs_g[0].maxNewTokens,
                             merge_layout_blocks=inputs_g[0].mergeLayoutBlocks,
                             markdown_ignore_labels=inputs_g[0].markdownIgnoreLabels,
+                            vlm_extra_args=inputs_g[0].vlmExtraArgs,
                         )
                     )
 
@@ -174,10 +179,13 @@ def run_batch(self, inputs, log_ids, batch_id):
 
     def _group_inputs(self, inputs):
         def _to_hashable(obj):
-            if isinstance(obj, list):
-                return tuple(obj)
-            elif isinstance(obj, dict):
-                return tuple(sorted(obj.items()))
+            if isinstance(obj, dict):
+                return tuple(
+                    (_to_hashable(k), _to_hashable(v))
+                    for k, v in sorted(obj.items(), key=lambda x: repr(x[0]))
+                )
+            elif isinstance(obj, list):
+                return tuple(_to_hashable(x) for x in obj)
             else:
                 return obj
 
@@ -191,7 +199,10 @@ def _hash(input):
                                 input.useDocOrientationClassify,
                                 input.useDocUnwarping,
                                 input.useLayoutDetection,
+                                input.usePolygonPoints,
                                 input.useChartRecognition,
+                                input.useSealRecognition,
+                                input.useOcrForImageBlock,
                                 input.layoutThreshold,
                                 input.layoutNms,
                                 input.layoutUnclipRatio,
@@ -206,6 +217,7 @@ def _hash(input):
                                 input.maxNewTokens,
                                 input.mergeLayoutBlocks,
                                 input.markdownIgnoreLabels,
+                                input.vlmExtraArgs,
                             )
                         ),
                     )
@@ -248,12 +260,20 @@ def _preprocess(self, input, log_id):
             else self.app_config.visualize
         )
 
-        file_bytes = utils.get_raw_bytes(input.file)
-        images, data_info = utils.file_to_images(
-            file_bytes,
-            file_type,
-            max_num_imgs=self.context["max_num_input_imgs"],
-        )
+        try:
+            file_bytes = utils.get_raw_bytes(input.file)
+            images, data_info = utils.file_to_images(
+                file_bytes,
+                file_type,
+                max_num_imgs=self.context["max_num_input_imgs"],
+            )
+        except Exception as e:
+            logging.error("Failed to get input file bytes: %s", e)
+            return protocol.create_aistudio_output_without_result(
+                422,
+                "Input file is invalid",
+                log_id=log_id,
+            )
 
         return images, data_info, visualize_enabled
 

deploy/hps/sdk/pipelines/PaddleOCR-VL/server/pipeline_config.yaml

@@ -83,7 +83,7 @@ SubModules:
     module_name: vl_recognition
     model_name: PaddleOCR-VL-0.9B
     model_dir: null
-    batch_size: 4096
+    batch_size: -1
     genai_config:
       backend: native
 

deploy/hps/sdk/pipelines/PaddleOCR-VL/version.txt

@@ -0,0 +1 @@
+0.2.0

deploy/hps/sdk/pipelines/anomaly_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/doc_preprocessor/version.txt

@@ -0,0 +1 @@
+0.2.1

deploy/hps/sdk/pipelines/doc_understanding/version.txt

@@ -0,0 +1 @@
+0.1.2

deploy/hps/sdk/pipelines/face_recognition/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/formula_recognition/version.txt

@@ -0,0 +1 @@
+0.2.1

deploy/hps/sdk/pipelines/human_keypoint_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/image_classification/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/image_multilabel_classification/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/instance_segmentation/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/layout_parsing/version.txt

@@ -0,0 +1 @@
+0.3.1

deploy/hps/sdk/pipelines/multilingual_speech_recognition/version.txt

@@ -0,0 +1 @@
+0.2.1

deploy/hps/sdk/pipelines/object_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/open_vocabulary_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/open_vocabulary_segmentation/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/pedestrian_attribute_recognition/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/rotated_object_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/seal_recognition/version.txt

@@ -0,0 +1 @@
+0.2.2

deploy/hps/sdk/pipelines/semantic_segmentation/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/small_object_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/table_recognition/version.txt

@@ -0,0 +1 @@
+0.4.1

deploy/hps/sdk/pipelines/table_recognition_v2/version.txt

@@ -0,0 +1 @@
+0.4.1

deploy/hps/sdk/pipelines/ts_anomaly_detection/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/ts_classification/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/ts_forecast/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/vehicle_attribute_recognition/version.txt

@@ -0,0 +1 @@
+0.1.1

deploy/hps/sdk/pipelines/video_classification/version.txt

@@ -0,0 +1 @@
+0.2.1

deploy/hps/sdk/pipelines/video_detection/version.txt

@@ -0,0 +1 @@
+0.2.1