swing deploy api

Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/14103985
* tester

* bug fixed: download_file not import

* input example analyzer

* api: get_task_input_examples

* update chat

* 43 task

* fix decode_base64

* json format

* schema

MANIFEST.in

@@ -1,2 +1,3 @@
 recursive-include modelscope/configs *.py *.cu *.h *.cpp
 recursive-include modelscope/cli/template *.tpl
+recursive-include modelscope/utils *.json

modelscope/pipeline_inputs.py

@@ -82,6 +82,34 @@ def check_input_type(input_type, input):
 
 
 TASK_INPUTS = {
+    Tasks.image_text_retrieval: {
+        InputKeys.IMAGE: InputType.IMAGE,
+        InputKeys.TEXT: InputType.TEXT
+    },
+    Tasks.general_recognition: {
+        InputKeys.IMAGE: InputType.IMAGE,
+        InputKeys.TEXT: InputType.TEXT
+    },
+    Tasks.video_depth_estimation: {
+        InputKeys.IMAGE: InputType.IMAGE,
+        InputKeys.TEXT: InputType.TEXT
+    },
+    Tasks.indoor_layout_estimation:
+    InputType.IMAGE,
+    Tasks.image_demoireing:
+    InputType.IMAGE,
+    Tasks.panorama_depth_estimation:
+    InputType.IMAGE,
+    Tasks.video_depth_estimation:
+    InputType.VIDEO,
+    Tasks.animal_recognition:
+    InputType.IMAGE,
+    Tasks.motion_generation:
+    InputType.TEXT,
+    Tasks.video_panoptic_segmentation:
+    InputType.VIDEO,
+
+
 
     Tasks.task_template: {
         'image': InputType.IMAGE,

modelscope/pipelines/nlp/text_generation_pipeline.py

@@ -297,12 +297,11 @@ class ChatGLM6bV2TextGenerationPipeline(Pipeline):
 class QWenChatPipeline(Pipeline):
 
     def __init__(self, model: Union[Model, str], **kwargs):
-        from modelscope.models.nlp import (QWenConfig, QWenForTextGeneration,
-                                           QWenTokenizer)
+        from modelscope import AutoModelForCausalLM, AutoTokenizer
         torch_dtype = kwargs.get('torch_dtype', torch.bfloat16)
         device_map = kwargs.get('device_map', 'auto')
         use_max_memory = kwargs.get('use_max_memory', False)
-        quantization_config = kwargs.get('quantization_config', None)
+        revision = kwargs.get('model_revision', 'v.1.0.5')
 
         if use_max_memory:
             max_memory = f'{int(torch.cuda.mem_get_info()[0] / 1024 ** 3) - 2}GB'
@@ -310,31 +309,24 @@ class QWenChatPipeline(Pipeline):
             max_memory = {i: max_memory for i in range(n_gpus)}
         else:
             max_memory = None
+        if torch_dtype == 'bf16' or torch_dtype == torch.bfloat16:
+            bf16 = True
+        else:
+            bf16 = False
 
         if isinstance(model, str):
-            model_dir = snapshot_download(
-                model) if not os.path.exists(model) else model
-
-            config = read_config(model_dir)
-            model_config = QWenConfig.from_pretrained(model_dir)
-            model_config.torch_dtype = torch_dtype
-
-            model = QWenForTextGeneration.from_pretrained(
-                model_dir,
-                cfg_dict=config,
-                config=model_config,
+            self.tokenizer = AutoTokenizer.from_pretrained(
+                model, revision=revision, trust_remote_code=True)
+            self.model = AutoModelForCausalLM.from_pretrained(
+                model,
                 device_map=device_map,
-                torch_dtype=torch_dtype,
-                quantization_config=quantization_config,
-                max_memory=max_memory)
-            model.generation_config = GenerationConfig.from_pretrained(
-                model_dir)
+                revision=revision,
+                trust_remote_code=True,
+                fp16=bf16).eval()
+            self.model.generation_config = GenerationConfig.from_pretrained(
+                model, trust_remote_code=True)  # 可指定不同的生成长度、top_p等相关超参
 
-        self.model = model
-        self.model.eval()
-        self.tokenizer = QWenTokenizer.from_pretrained(self.model.model_dir)
-
-        super().__init__(model=model, **kwargs)
+        super().__init__(model=self.model, **kwargs)
         # skip pipeline model placement
         self._model_prepare = True
 
@@ -345,12 +337,19 @@ class QWenChatPipeline(Pipeline):
         return inputs
 
     # define the forward pass
-    def forward(self, inputs: str, **forward_params) -> Dict[str, Any]:
-        history = forward_params.get('history', None)
+    def forward(self, inputs: Union[Dict, str],
+                **forward_params) -> Dict[str, Any]:
+        if isinstance(inputs, Dict):
+            text = inputs.get('text', None)
+            history = inputs.get('history', None)
+        else:
+            text = inputs
+            history = forward_params.get('history', None)
         system = forward_params.get('system', 'You are a helpful assistant.')
         append_history = forward_params.get('append_history', True)
-        return self.model.chat(self.tokenizer, inputs, history, system,
-                               append_history)
+        res = self.model.chat(self.tokenizer, text, history, system,
+                              append_history)
+        return {'response': res[0], 'history': res[1]}
 
     # format the outputs from pipeline
     def postprocess(self, input, **kwargs) -> Dict[str, Any]:
@@ -362,12 +361,11 @@ class QWenChatPipeline(Pipeline):
 class QWenTextGenerationPipeline(Pipeline):
 
     def __init__(self, model: Union[Model, str], **kwargs):
-        from modelscope.models.nlp import (QWenConfig, QWenForTextGeneration,
-                                           QWenTokenizer)
+        from modelscope import AutoModelForCausalLM, AutoTokenizer
         torch_dtype = kwargs.get('torch_dtype', torch.bfloat16)
         device_map = kwargs.get('device_map', 'auto')
         use_max_memory = kwargs.get('use_max_memory', False)
-        quantization_config = kwargs.get('quantization_config', None)
+        revision = kwargs.get('model_revision', 'v.1.0.4')
 
         if use_max_memory:
             max_memory = f'{int(torch.cuda.mem_get_info()[0] / 1024 ** 3) - 2}GB'
@@ -375,31 +373,27 @@ class QWenTextGenerationPipeline(Pipeline):
             max_memory = {i: max_memory for i in range(n_gpus)}
         else:
             max_memory = None
+        if torch_dtype == 'bf16' or torch_dtype == torch.bfloat16:
+            bf16 = True
+        else:
+            bf16 = False
 
         if isinstance(model, str):
-            model_dir = snapshot_download(
-                model) if not os.path.exists(model) else model
-
-            config = read_config(model_dir)
-            model_config = QWenConfig.from_pretrained(model_dir)
-            model_config.torch_dtype = torch_dtype
-
-            model = QWenForTextGeneration.from_pretrained(
-                model_dir,
-                cfg_dict=config,
-                config=model_config,
+            self.model = AutoModelForCausalLM.from_pretrained(
+                model,
                 device_map=device_map,
-                torch_dtype=torch_dtype,
-                quantization_config=quantization_config,
-                max_memory=max_memory)
-            model.generation_config = GenerationConfig.from_pretrained(
-                model_dir)
+                revision=revision,
+                trust_remote_code=True,
+                bf16=bf16).eval()
+            self.tokenizer = AutoTokenizer.from_pretrained(
+                model, revision=revision, trust_remote_code=True)
+            self.model.generation_config = GenerationConfig.from_pretrained(
+                model)
+        else:
+            self.model = model
+            self.tokenizer = kwargs.get('tokenizer', None)
 
-        self.model = model
-        self.model.eval()
-        self.tokenizer = QWenTokenizer.from_pretrained(self.model.model_dir)
-
-        super().__init__(model=model, **kwargs)
+        super().__init__(model=self.model, **kwargs)
         # skip pipeline model placement
         self._model_prepare = True
 
@@ -411,10 +405,12 @@ class QWenTextGenerationPipeline(Pipeline):
 
     # define the forward pass
     def forward(self, inputs: str, **forward_params) -> Dict[str, Any]:
+        inputs = self.tokenizer(inputs, return_tensors='pt').to('cuda:0')
         return {
             OutputKeys.TEXT:
-            self.model.chat(self.tokenizer, inputs,
-                            history=None)[OutputKeys.RESPONSE]
+            self.tokenizer.decode(
+                self.model.generate(**inputs).cpu()[0],
+                skip_special_tokens=True)
         }
 
     # format the outputs from pipeline

modelscope/preprocessors/image.py

@@ -10,6 +10,7 @@ from PIL import Image, ImageOps
 
 from modelscope.fileio import File
 from modelscope.metainfo import Preprocessors
+from modelscope.pipeline_inputs import InputKeys
 from modelscope.utils.constant import Fields
 from modelscope.utils.type_assert import type_assert
 from .base import Preprocessor
@@ -92,6 +93,10 @@ class LoadImage:
             if len(input.shape) == 2:
                 input = cv2.cvtColor(input, cv2.COLOR_GRAY2BGR)
             img = input[:, :, ::-1]
+        elif isinstance(input, Dict):
+            img = input.get(InputKeys.IMAGE, None)
+            if img:
+                img = np.array(load_image(img))
         else:
             raise TypeError(f'input should be either str, PIL.Image,'
                             f' np.array, but got {type(input)}')
@@ -108,6 +113,10 @@ class LoadImage:
                 img = cv2.cvtColor(input, cv2.COLOR_GRAY2BGR)
             img = input[:, :, ::-1]
             img = Image.fromarray(img.astype('uint8')).convert('RGB')
+        elif isinstance(input, Dict):
+            img = input.get(InputKeys.IMAGE, None)
+            if img:
+                img = load_image(img)
         else:
             raise TypeError(f'input should be either str, PIL.Image,'
                             f' np.array, but got {type(input)}')

modelscope/utils/input_output.py

@@ -3,11 +3,13 @@ import ast
 import base64
 import importlib
 import inspect
+import os
 from io import BytesIO
 from typing import Any
 from urllib.parse import urlparse
 
 import cv2
+import json
 import numpy as np
 
 from modelscope.hub.api import HubApi
@@ -243,6 +245,33 @@ def process_arg_type_annotation(arg, default_value):
         return arg.arg, 'object'
 
 
+def convert_to_value(item):
+    if isinstance(item, ast.Str):
+        return item.s
+    elif hasattr(ast, 'Bytes') and isinstance(item, ast.Bytes):
+        return item.s
+    elif isinstance(item, ast.Tuple):
+        return tuple(convert_to_value(i) for i in item.elts)
+    elif isinstance(item, ast.Num):
+        return item.n
+    elif isinstance(item, ast.Name):
+        result = VariableKey(item=item)
+        constants_lookup = {
+            'True': True,
+            'False': False,
+            'None': None,
+        }
+        return constants_lookup.get(
+            result.name,
+            result,
+        )
+    elif isinstance(item, ast.NameConstant):
+        # None, True, False are nameconstants in python3, but names in 2
+        return item.value
+    else:
+        return UnhandledKeyType()
+
+
 def process_args(args):
     arguments = []
     # name, type, has_default, default
@@ -259,7 +288,7 @@ def process_args(args):
     # process defaults arg.
     for arg, dft in zip(args.args[n_args - n_args_default:], args.defaults):
         # compatible with python3.7 ast.Num no value.
-        value = dft.value if hasattr(dft, 'value') else dft.n
+        value = convert_to_value(dft)
         arg_name, arg_type = process_arg_type_annotation(arg, value)
         arguments.append((arg_name, arg_type, True, value))
 
@@ -398,7 +427,7 @@ meta_type_schema_map = {
 
 def generate_pipeline_parameters_schema(parameters):
     parameters_schema = {'type': 'object', 'properties': {}}
-    if len(parameters) == 0:
+    if parameters is None or len(parameters) == 0:
         return {}
     for param in parameters:
         name, param_type, has_default, default_value = param
@@ -523,16 +552,18 @@ def is_url(url: str):
 
 
 def decode_base64_to_image(content):
-    if content.startswith('http') or content.startswith('oss'):
+    if content.startswith('http') or content.startswith(
+            'oss') or os.path.exists(content):
         return content
 
     from PIL import Image
-    image_file_content = base64.b64decode(content)
+    image_file_content = base64.b64decode(content, '-_')
     return Image.open(BytesIO(image_file_content))
 
 
 def decode_base64_to_audio(content):
-    if content.startswith('http') or content.startswith('oss'):
+    if content.startswith('http') or content.startswith(
+            'oss') or os.path.exists(content):
         return content
 
     file_content = base64.b64decode(content)
@@ -540,7 +571,8 @@ def decode_base64_to_audio(content):
 
 
 def decode_base64_to_video(content):
-    if content.startswith('http') or content.startswith('oss'):
+    if content.startswith('http') or content.startswith(
+            'oss') or os.path.exists(content):
         return content
 
     file_content = base64.b64decode(content)
@@ -594,13 +626,14 @@ def call_pipeline_with_json(pipeline_info: PipelineInfomation,
         pipeline (Pipeline): The pipeline object.
         body (Dict): The input object, include input and parameters
     """
-    if pipeline_info.is_custom_call:
-        pipeline_inputs = body['input']
-        result = pipeline(**pipeline_inputs)
-    else:
-        pipeline_inputs, parameters = service_base64_input_to_pipeline_input(
-            pipeline_info.task_name, body)
-        result = pipeline(pipeline_inputs, **parameters)
+    # TODO: is_custom_call misjudgment
+    # if pipeline_info.is_custom_call:
+    #     pipeline_inputs = body['input']
+    #     result = pipeline(**pipeline_inputs)
+    # else:
+    pipeline_inputs, parameters = service_base64_input_to_pipeline_input(
+        pipeline_info.task_name, body)
+    result = pipeline(pipeline_inputs, **parameters)
 
     return result
 
@@ -737,6 +770,9 @@ def pipeline_output_to_service_base64_output(task_name, pipeline_output):
     task_outputs = []
     if task_name in TASK_OUTPUTS:
         task_outputs = TASK_OUTPUTS[task_name]
+    # TODO: for batch
+    if isinstance(pipeline_output, list):
+        pipeline_output = pipeline_output[0]
     for key, value in pipeline_output.items():
         if key not in task_outputs:
             continue  # skip the output not defined.
@@ -768,3 +804,77 @@ def pipeline_output_to_service_base64_output(task_name, pipeline_output):
             json_serializable_output[key] = value
 
     return _convert_to_python_type(json_serializable_output)
+
+
+def get_task_input_examples(task):
+    current_work_dir = os.path.dirname(__file__)
+    with open(current_work_dir + '/pipeline_inputs.json', 'r') as f:
+        input_examples = json.load(f)
+    if task in input_examples:
+        return input_examples[task]
+    return None
+
+
+def get_task_schemas(task):
+    current_work_dir = os.path.dirname(__file__)
+    with open(current_work_dir + '/pipeline_schema.json', 'r') as f:
+        schema = json.load(f)
+    if task in schema:
+        return schema[task]
+    return None
+
+
+if __name__ == '__main__':
+    from modelscope.utils.ast_utils import load_index
+    index = load_index()
+    task_schemas = {}
+    for key, value in index['index'].items():
+        reg, task_name, class_name = key
+        if reg == 'PIPELINES' and task_name != 'default':
+            print(
+                f"value['filepath']: {value['filepath']}, class_name: {class_name}"
+            )
+            input, parameters = get_pipeline_input_parameters(
+                value['filepath'], class_name)
+            try:
+                if task_name in TASK_INPUTS and task_name in TASK_OUTPUTS:
+                    # delete the first default input which is defined by task.
+                    # parameters.pop(0)
+                    parameters_schema = generate_pipeline_parameters_schema(
+                        parameters)
+                    input_schema = get_input_schema(task_name, None)
+                    output_schema = get_output_schema(task_name)
+                    schema = {
+                        'input': input_schema,
+                        'parameters': parameters_schema,
+                        'output': output_schema
+                    }
+                else:
+                    logger.warning(
+                        'Task: %s input is defined: %s, output is defined: %s which is not completed'
+                        % (task_name, task_name in TASK_INPUTS, task_name
+                           in TASK_OUTPUTS))
+                    input_schema = None
+                    output_schema = None
+                    if task_name in TASK_INPUTS:
+                        input_schema = get_input_schema(task_name, None)
+                    if task_name in TASK_OUTPUTS:
+                        output_schema = get_output_schema(task_name)
+                    parameters_schema = generate_pipeline_parameters_schema(
+                        parameters)
+                    schema = {
+                        'input': input_schema if input_schema else
+                        parameters_schema,  # all parameter is input
+                        'parameters':
+                        parameters_schema if input_schema else {},
+                        'output': output_schema if output_schema else {
+                            'type': 'object',
+                        },
+                    }
+            except BaseException:
+                continue
+            task_schemas[task_name] = schema
+
+    s = json.dumps(task_schemas)
+    with open('./task_schema.json', 'w') as f:
+        f.write(s)

modelscope/utils/pipeline_inputs.json

@@ -0,0 +1,277 @@
+{
+    "action-detection":{
+        "input":{
+            "video":"data/test/videos/action_detection_test_video.mp4"
+        }
+    },
+    "action-recognition":{
+        "input":{
+            "video":"data/test/videos/action_recognition_test_video.mp4"
+        }
+    },
+    "animal-recognition":{
+        "input":{
+            "image":"data/test/images/dogs.jpg"
+        }
+    },
+    "chat":{
+        "input":{
+            "text":"你有什么推荐吗？",
+            "history":[
+                [
+                    "今天天气真好，",
+                    "今天天气真好，出去走走怎么样？"
+                ]
+            ]
+        }
+    },
+    "domain-specific-object-detection":{
+        "input":{
+            "image":"data/test/images/image_traffic_sign.jpg"
+        }
+    },
+    "face-2d-keypoints":{
+        "input":{
+            "image":"data/test/images/face_detection.png"
+        }
+    },
+    "face-attribute-recognition":{
+        "input":{
+            "image":"data/test/images/face_recognition_1.png"
+        }
+    },
+    "facial-expression-recognition":{
+        "input":{
+            "image":"data/test/images/facial_expression_recognition.jpg"
+        }
+    },
+    "general-recognition":{
+        "input":{
+            "image":"data/test/images/dogs.jpg"
+        }
+    },
+    "human-detection":{
+        "input":{
+            "image":"data/test/images/image_detection.jpg"
+        }
+    },
+    "image-captioning":{
+        "input":{
+            "image":"data/test/images/image_captioning.png"
+        }
+    },
+    "image-classification":{
+        "input":{
+            "image":"data/test/images/content_check.jpg"
+        }
+    },
+    "image-demoireing":{
+        "input":{
+            "image":"data/test/images/shop_segmentation.jpg"
+        }
+    },
+    "image-object-detection":{
+        "input":{
+            "image":"data/test/images/image_detection.jpg"
+        }
+    },
+    "image-portrait-stylization":{
+        "input":{
+            "image":"https://modelscope.oss-cn-beijing.aliyuncs.com/test/images/image_cartoon.png"
+        }
+    },
+    "image-segmentation":{
+        "input":{
+            "image":"data/test/images/image_semantic_segmentation.jpg"
+        },
+        "parameters":{
+
+        }
+    },
+    "image-text-retrieval":{
+        "input":{
+            "image":"data/test/images/image_mplug_vqa.jpg",
+            "text":"What is the woman doing?"
+        }
+    },
+    "indoor-layout-estimation":{
+        "input":{
+            "image":"data/test/images/image_traffic_sign.jpg"
+        }
+    },
+    "live-category":{
+        "input":{
+            "video":"data/test/videos/live_category_test_video.mp4"
+        }
+    },
+    "motion-generation":{
+        "input":{
+            "text":"the person walked forward and is picking up his toolbox"
+        }
+    },
+    "named-entity-recognition":{
+        "input":{
+            "text":"这与温岭市新河镇的一个神秘的传说有关。[SEP]地名"
+        }
+    },
+    "nli":{
+        "input":[
+            "四川商务职业学院和四川财经职业学院哪个好？",
+            "四川商务职业学院商务管理在哪个校区？"
+        ],
+        "parameters":{
+
+        }
+    },
+    "ocr-recognition":{
+        "input":{
+            "image":"data/test/images/image_ocr_recognition.jpg"
+        }
+    },
+    "panorama-depth-estimation":{
+        "input":{
+            "image":"data/test/images/panorama_depth_estimation.jpg"
+        }
+    },
+    "semantic-segmentation":{
+        "input":{
+            "image":"data/test/images/image_salient_detection.jpg"
+        }
+    },
+    "shop-segmentation":{
+        "input":{
+            "image":"data/test/images/shop_segmentation.jpg"
+        }
+    },
+    "text-classification":{
+        "input":{
+            "text":"i like this wonderful place"
+        },
+        "parameters":{
+
+        }
+    },
+    "text-driven-segmentation":{
+        "input":{
+            "image":"data/test/images/text_driven_segmentation.jpg",
+            "text":"bear"
+        }
+    },
+    "text-generation":{
+        "input":{
+            "text":"蒙古国的首都是乌兰巴托（Ulaanbaatar）\n冰岛的首都是雷克雅未克（Reykjavik）\n埃塞俄比亚的首都是"
+        },
+        "parameters":{
+
+        }
+    },
+    "text-ranking":{
+        "input":{
+            "source_sentence":[
+                "how long it take to get a master's degree"
+            ],
+            "sentences_to_compare":[
+                "On average, students take about 18 to 24 months to complete a master's degree.",
+                "On the other hand, some students prefer to go at a slower pace and choose to take several years to complete their studies.",
+                "It can take anywhere from two semesters"
+            ]
+        }
+    },
+    "text-summarization":{
+        "input":{
+            "text":"five-time world champion michelle kwan withdrew from the #### us figure skating championships on wednesday , but will petition us skating officials for the chance to compete at the #### turin olympics ."
+        }
+    },
+    "text-to-video-synthesis":{
+        "input":{
+            "text":"A panda eating bamboo on a rock."
+        }
+    },
+    "translation":{
+        "input":{
+            "text":"声明补充说，沃伦的同事都深感震惊，并且希望他能够投案自首。"
+        }
+    },
+    "video-captioning":{
+        "input":{
+            "video":"data/test/videos/video_caption_and_qa_test.mp4"
+        }
+    },
+    "video-category":{
+        "input":{
+            "video":"data/test/videos/video_category_test_video.mp4"
+        }
+    },
+    "video-depth-estimation":{
+        "input":{
+            "video":"data/test/videos/video_depth_estimation.mp4"
+        }
+    },
+    "video-embedding":{
+        "input":{
+            "video":"data/test/videos/action_recognition_test_video.mp4"
+        }
+    },
+    "video-multi-object-tracking":{
+        "input":{
+            "video":"data/test/videos/MOT17-03-partial.mp4"
+        }
+    },
+    "video-panoptic-segmentation":{
+        "input":{
+            "video":"data/test/videos/kitti-step_testing_image_02_0000.mp4"
+        }
+    },
+    "video-question-answering":{
+        "input":{
+            "video":"data/test/videos/video_caption_and_qa_test.mp4",
+            "text":"How many people are there?"
+        }
+    },
+    "video-summarization":{
+        "input":{
+            "text":"data/test/videos/video_category_test_video.mp4"
+        }
+    },
+    "visual-entailment":{
+        "input":{
+            "image":"data/test/images/dogs.jpg",
+            "text":"there are two birds."
+        }
+    },
+    "visual-grounding":{
+        "input":{
+            "image":"data/test/images/visual_grounding.png",
+            "text":"a blue turtle-like pokemon with round head"
+        }
+    },
+    "visual-question-answering":{
+        "input":{
+            "image":"data/test/images/image_mplug_vqa.jpg",
+            "text":"What is the woman doing?"
+        }
+    },
+    "word-segmentation":{
+        "input":{
+            "text":"今天天气不错，适合出去游玩"
+        }
+    },
+    "zero-shot-classification":{
+        "input":{
+            "text":"全新突破 解放军运20版空中加油机曝光"
+        },
+        "parameters":{
+            "candidate_labels":[
+                "文化",
+                "体育",
+                "娱乐",
+                "财经",
+                "家居",
+                "汽车",
+                "教育",
+                "科技",
+                "军事"
+            ]
+        }
+    }
+}

tests/json_call_test.py

@@ -0,0 +1,76 @@
+import os
+
+import json
+
+from modelscope.hub.api import HubApi
+from modelscope.hub.file_download import model_file_download
+from modelscope.hub.utils.utils import get_cache_dir
+from modelscope.pipelines import pipeline
+from modelscope.utils.config import Config
+from modelscope.utils.constant import ModelFile
+from modelscope.utils.input_output import (
+    call_pipeline_with_json, get_pipeline_information_by_pipeline,
+    get_task_input_examples, pipeline_output_to_service_base64_output)
+
+
+class ModelJsonTest:
+
+    def __init__(self):
+        self.api = HubApi()
+
+    def test_single(self, model_id: str, model_revision=None):
+        # get model_revision & task info
+        cache_root = get_cache_dir()
+        configuration_file = os.path.join(cache_root, model_id,
+                                          ModelFile.CONFIGURATION)
+        if not model_revision:
+            model_revision = self.api.list_model_revisions(
+                model_id=model_id)[0]
+        if not os.path.exists(configuration_file):
+
+            configuration_file = model_file_download(
+                model_id=model_id,
+                file_path=ModelFile.CONFIGURATION,
+                revision=model_revision)
+        cfg = Config.from_file(configuration_file)
+        task = cfg.safe_get('task')
+
+        # init pipeline
+        ppl = pipeline(
+            task=task, model=model_id, model_revision=model_revision)
+        pipeline_info = get_pipeline_information_by_pipeline(ppl)
+
+        # call pipeline
+        data = get_task_input_examples(task)
+        print(task, data)
+        infer_result = call_pipeline_with_json(pipeline_info, ppl, data)
+        result = pipeline_output_to_service_base64_output(task, infer_result)
+        return result
+
+
+if __name__ == '__main__':
+    model_list = [
+        'damo/nlp_structbert_nli_chinese-base',
+        'damo/nlp_structbert_word-segmentation_chinese-base',
+        'damo/nlp_structbert_zero-shot-classification_chinese-base',
+        'damo/cv_unet_person-image-cartoon_compound-models',
+        'damo/nlp_structbert_sentiment-classification_chinese-tiny',
+        'damo/nlp_csanmt_translation_zh2en',
+        'damo/nlp_rom_passage-ranking_chinese-base',
+        'damo/ofa_image-caption_muge_base_zh',
+        'damo/nlp_raner_named-entity-recognition_chinese-base-ecom-50cls',
+        'damo/nlp_structbert_sentiment-classification_chinese-ecommerce-base',
+        'damo/text-to-video-synthesis',
+        'qwen/Qwen-7B',
+        'qwen/Qwen-7B-Chat',
+        'ZhipuAI/ChatGLM-6B',
+    ]
+    tester = ModelJsonTest()
+    for model in model_list:
+        try:
+            res = tester.test_single(model)
+            print(f'\nmodel_id {model} call_pipeline_with_json run ok.\n')
+        except BaseException as e:
+            print(
+                f'\nmodel_id {model} call_pipeline_with_json run failed: {e}.\n'
+            )

tests/utils/case_file_analyzer.py

@@ -62,7 +62,10 @@ class AnalysisTestFile(ast.NodeVisitor):
 
 class AnalysisTestClass(ast.NodeVisitor):
 
-    def __init__(self, test_class_node, builder_function_name) -> None:
+    def __init__(self,
+                 test_class_node,
+                 builder_function_name,
+                 file_analyzer=None) -> None:
         super().__init__()
         self.test_class_node = test_class_node
         self.builder_function_name = builder_function_name
@@ -72,6 +75,44 @@ class AnalysisTestClass(ast.NodeVisitor):
         ]  # class method trainer builder(call build_trainer)
         self.custom_class_method_builder_calls = [
         ]  # the builder call statement
+        self.variables = {}
+
+    def get_variables(self, key: str):
+        if key in self.variables:
+            return self.variables[key]
+        return key
+
+    def get_ast_value(self, statements):
+        if not isinstance(statements, list):
+            statements = [statements]
+        res = []
+        for item in statements:
+            if isinstance(item, ast.Name):
+                res.append(self.get_variables(item.id))
+            elif isinstance(item, ast.Attribute):
+                res.append(self.get_variables(item.value.id))
+            elif isinstance(item, ast.Str):
+                res.append(self.get_variables(item.s))
+            elif isinstance(item, ast.Dict):
+                keys = [i.s for i in item.keys]
+                values = self.get_ast_value(item.values)
+                res.append(dict(zip(keys, values)))
+        return res
+
+    def get_final_variables(self, statement: ast.Assign):
+        if len(statement.targets) == 1 and \
+           isinstance(statement.targets[0], ast.Name):
+            if isinstance(statement.value, ast.Call):
+                if isinstance(statement.value.func, ast.Attribute) and \
+                   isinstance(statement.value.func.value, ast.Name) and \
+                   statement.value.func.value.id == 'Image':
+                    self.variables[str(
+                        statement.targets[0].id)] = self.get_ast_value(
+                            statement.value.args[0])
+            else:
+                self.variables[str(
+                    statement.targets[0].id)] = self.get_ast_value(
+                        statement.value)
 
     def visit_FunctionDef(self, node: ast.FunctionDef) -> Any:
         if node.name.startswith('setUp'):
@@ -83,6 +124,7 @@ class AnalysisTestClass(ast.NodeVisitor):
                         self.setup_variables[str(
                             statement.targets[0].attr)] = str(
                                 statement.value.attr)
+                    self.get_final_variables(statement)
         elif node.name.startswith('test_'):
             self.test_methods.append(node)
         else:
@@ -312,6 +354,48 @@ def analysis_trainer_test_suite(test_file, modified_register_modules):
     return tested_trainers
 
 
+def get_test_parameters(test_method, analyzer):
+    for node in ast.walk(test_method):
+        func = None
+        if not isinstance(node, ast.FunctionDef):
+            continue
+        for statement in node.body:
+            if isinstance(statement, ast.Assign):
+                analyzer.get_final_variables(statement)
+            if not func and isinstance(statement, ast.Assign):
+                if isinstance(statement.value, ast.Call) and isinstance(
+                        statement.value.func, ast.Name) and (  # noqa W504
+                            'pipeline' in statement.value.func.id
+                            or 'Pipeline' in statement.value.func.id):
+                    func = statement.targets[0].id
+            if func and isinstance(statement, ast.Assign) and isinstance(
+                    statement.value, ast.Call) and isinstance(
+                        statement.value.func, ast.Name):
+                if statement.value.func.id == func:
+                    inputs = statement.value.args
+                    return analyzer.get_ast_value(inputs)
+
+
+def analysis_pipeline_test_examples(test_file):
+    examples = []
+    with open(test_file, 'rb') as tsf:
+        src = tsf.read()
+    test_root = ast.parse(src, test_file)
+    test_file_analyzer = AnalysisTestFile(
+        test_file, SYSTEM_PIPELINE_BUILDER_FUNCTION_NAME)
+    test_file_analyzer.visit(test_root)
+
+    for test_class in test_file_analyzer.test_classes:
+        test_class_analyzer = AnalysisTestClass(
+            test_class, SYSTEM_PIPELINE_BUILDER_FUNCTION_NAME,
+            test_file_analyzer)
+        test_class_analyzer.visit(test_class)
+        for test_method in test_class_analyzer.test_methods:
+            parameters = get_test_parameters(test_method, test_class_analyzer)
+            examples.append(parameters)
+    return examples
+
+
 def analysis_pipeline_test_suite(test_file, modified_register_modules):
     tested_tasks = []
     with open(test_file, 'rb') as tsf:
@@ -413,7 +497,18 @@ def get_pipelines_trainers_test_info(register_modules):
 
 
 if __name__ == '__main__':
-    test_file = 'tests/pipelines/test_action_detection.py'
-    tasks = analysis_pipeline_test_suite(test_file, None)
+    all_pipeline_cases = [
+        os.path.join(dp, f) for dp, dn, filenames in os.walk(
+            os.path.join(os.getcwd(), 'tests', 'pipelines')) for f in filenames
+        if os.path.splitext(f)[1] == '.py'
+    ]
+    for test_file in all_pipeline_cases:
+        print('\n', test_file)
+        tasks = analysis_pipeline_test_suite(test_file, None)
+        examples = analysis_pipeline_test_examples(test_file)
 
-    print(tasks)
+        from modelsope.metainfo import Tasks
+        for task, example in zip(tasks, examples):
+            task_convert = f't = Tasks.{task}'
+            exec(task_convert)
+            print(t, example)