修改Component.create_output为classmethod方法，增加组件单测init_args (#621)

* create_output变更为classmethod方法，component_tool_eval_schemas移动到case中 * create_output变更为classmethod方法，component_tool_eval_schemas移动到case中 * create_output变更为classmethod方法，component_tool_eval_schemas移动到case中 * create_output变更为classmethod方法，component_tool_eval_schemas移动到case中 --------- Co-authored-by: yepeiwen01 <[email protected]>
baidubce · Nov 27, 2024 · f418a33 · f418a33
1 parent 60d2cd1
commit f418a33
Show file tree

Hide file tree

Showing 4 changed files with 79 additions and 64 deletions.
diff --git a/python/core/component.py b/python/core/component.py
@@ -486,11 +486,12 @@ def _langchain_tool_eval_implement(self, **kwargs):
                 final_result += step.get("text", "")
         return final_result
 
-    def create_output(self, type, text, role="tool", name="", visible_scope="all", raw_data={}, usage={}, metrics={}):
+    @classmethod
+    def create_output(cls, type, text, role="tool", name="", visible_scope="all", raw_data={}, usage={}, metrics={}):
         """create_text_output
 
         Args:
-            type (str): 类型，包括"text", "code", "files", "urls", "oral_text", "references", "image", "chart", "audio"
+            type (str): 类型，包括"text", "code", "files", "urls", "oral_text", "references", "image", "chart", "audio", "plan", "function_call"
             text (str|dict): text字段，可输入str或dict
             role (str, optional): 当前消息来源. Defaults to "tool".
             name (str, optional): 当前yield内容的step name. Defaults to "".

diff --git a/python/tests/component_check.py b/python/tests/component_check.py
@@ -22,7 +22,6 @@
 from appbuilder.utils.json_schema_to_model import json_schema_to_pydantic_model
 from appbuilder.tests.component_schemas import type_to_json_schemas
 from component_tool_eval_cases import component_tool_eval_cases
-from component_tool_eval_schemas import components_tool_eval_output_json_maps
 
 
 class CheckInfo(BaseModel):
@@ -288,7 +287,7 @@ def _check_jsonschema(self, outputs, output_schemas):
         """检查输出格式是否符合对应的json schema
         """
         invalid_details = []
-        if len(self._check_pre_format(outputs)) > 0 :
+        if len(invalid_details:=self._check_pre_format(outputs)) > 0 :
             return invalid_details
 
         for content in outputs["content"]: 
@@ -369,42 +368,40 @@ def _check_text_and_code(self, component_case, output_dict):
     def check(self, component_cls) -> CheckInfo:
         invalid_details = []
         component_cls_name = component_cls.__name__
-        if component_cls_name not in components_tool_eval_output_json_maps:
-            invalid_details.append("{} 没有注册到 components_tool_eval_output_json_maps 中".format(component_cls_name))
+        if component_cls_name not in component_tool_eval_cases:
+            invalid_details.append("{} 没有添加测试case到 component_tool_eval_cases 中".format(component_cls_name))
         else:
-            output_json_schemas = components_tool_eval_output_json_maps[component_cls_name]
-            if component_cls_name not in component_tool_eval_cases:
-                invalid_details.append("{} 没有添加测试case到 component_tool_eval_cases 中".format(component_cls_name))
-            else:
-                component_case = component_tool_eval_cases[component_cls_name]()
-                input_dict = component_case.inputs()
-                component_obj = component_cls()
-
-                try:
-                    stream_output_dict = {"text": "", "oral_text":"", "code": ""}
-                    stream_outputs = component_obj.tool_eval(**input_dict)
-                    for stream_output in stream_outputs: #校验流式输出
-                        iter_invalid_detail = self._check_jsonschema(stream_output.model_dump(), output_json_schemas)
-                        invalid_details.extend(["流式" + error_message for error_message in iter_invalid_detail])
-                        iter_output_dict = self._gather_iter_outputs(stream_output)
-                        stream_output_dict["text"] += iter_output_dict["text"]
-                        stream_output_dict["oral_text"] += iter_output_dict["oral_text"]
-                        stream_output_dict["code"] += iter_output_dict["code"]
-                    if len(invalid_details) == 0:
-                        invalid_details.extend(self._check_text_and_code(component_case, stream_output_dict))
-                except Exception as e:
-                    invalid_details.append("ToolEval执行失败: {}".format(e))
-
-                time.sleep(2)
-                try:
-                    non_stream_outputs = component_obj.non_stream_tool_eval(**input_dict)
-                    non_stream_invalid_details  = self._check_jsonschema(non_stream_outputs.model_dump(), output_json_schemas)  #校验非流式输出
-                    invalid_details.extend(["非流式" + error_message for error_message in non_stream_invalid_details]) 
-                    if len(invalid_details) == 0:
-                        non_stream_output_dict = self._gather_iter_outputs(non_stream_outputs)
-                        invalid_details.extend(self._check_text_and_code(component_case, non_stream_output_dict))
-                except Exception as e:
-                    invalid_details.append(" NonStreamToolEval执行失败: {}".format(e))
+            component_case = component_tool_eval_cases[component_cls_name]()
+            input_dict = component_case.inputs()
+            output_json_schemas = component_case.schemas()
+            init_args = component_case.init_args()
+            component_obj = component_cls(**init_args)
+
+            try: #校验流式输出
+                stream_output_dict = {"text": "", "oral_text":"", "code": ""}
+                stream_outputs = component_obj.tool_eval(**input_dict)
+                for stream_output in stream_outputs: 
+                    iter_invalid_detail = self._check_jsonschema(stream_output.model_dump(), output_json_schemas)
+                    invalid_details.extend(["流式" + error_message for error_message in iter_invalid_detail])
+                    iter_output_dict = self._gather_iter_outputs(stream_output)
+                    stream_output_dict["text"] += iter_output_dict["text"]
+                    stream_output_dict["oral_text"] += iter_output_dict["oral_text"]
+                    stream_output_dict["code"] += iter_output_dict["code"]
+                if len(invalid_details) == 0:
+                    invalid_details.extend(self._check_text_and_code(component_case, stream_output_dict))
+            except Exception as e:
+                invalid_details.append("ToolEval执行失败: {}".format(e))
+
+            time.sleep(2)
+            try: #校验非流式输出
+                non_stream_outputs = component_obj.non_stream_tool_eval(**input_dict)
+                non_stream_invalid_details  = self._check_jsonschema(non_stream_outputs.model_dump(), output_json_schemas)
+                invalid_details.extend(["非流式" + error_message for error_message in non_stream_invalid_details]) 
+                if len(invalid_details) == 0:
+                    non_stream_output_dict = self._gather_iter_outputs(non_stream_outputs)
+                    invalid_details.extend(self._check_text_and_code(component_case, non_stream_output_dict))
+            except Exception as e:
+                invalid_details.append(" NonStreamToolEval执行失败: {}".format(e))
 
         if len(invalid_details) > 0:
             return CheckInfo(

diff --git a/python/tests/component_tool_eval_cases.py b/python/tests/component_tool_eval_cases.py
@@ -11,8 +11,27 @@
 # 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 appbuilder.core.component import Component
+from appbuilder.tests.component_schemas import text_schema, url_schema, image_schema, code_schema, file_schema, oral_text_schema, references_schema, chart_schema, audio_schema, plan_schema, function_call_schema
 
-class AnimalRecognitionCase:
+class Case():
+    def init_args(self):
+        return {}
+
+    def inputs(self):
+        return NotImplementedError()
+
+    def outputs(self):
+        return {}
+
+    def schemas(self):
+        return NotImplementedError()
+
+    def envs(self):
+        return {}
+
+
+class AnimalRecognitionCase(Case):
     def inputs(self):
         return {
             "img_name": "",
@@ -24,7 +43,10 @@ def inputs(self):
     def outputs(self):
         return {"text": ["熊猫"]}
 
-class ASRCase:
+    def schemas(self):
+        return [text_schema]
+
+class ASRCase(Case):
     def inputs(self):
         return {
             "file_url": "https://bj.bcebos.com/v1/appbuilder/asr_test.pcm?authorization=bce-auth-v1" \
@@ -34,11 +56,17 @@ def inputs(self):
     def outputs(self):
         return {"text": ["北京科技馆"]}
 
-class TreeMindCase:
+    def schemas(self):
+        return [text_schema]
+
+class TreeMindCase(Case):
     def inputs(self):
         return {"query": "生成一份年度总结的思维导图"}
+
+    def schemas(self):
+        return [text_schema, url_schema]
 
-class ImageUnderstandCase:
+class ImageUnderstandCase(Case):
     def inputs(self):
         return {
             "img_url": "https://bj.bcebos.com/v1/appbuilder/animal_recognize_test.png?" \
@@ -50,6 +78,16 @@ def inputs(self):
     def outputs(self):
         return {"text": ["熊猫"]}
 
+    def schemas(self):
+        return [text_schema]
+
+class Text2ImageCase(Case):
+    def inputs(self):
+        return {"query": "生成一张熊猫图片"}
+
+    def schemas(self):
+        return [url_schema]
+
 component_tool_eval_cases = {
     "AnimalRecognition": AnimalRecognitionCase,
     "ImageUnderstand": ImageUnderstandCase,

diff --git a/python/tests/component_tool_eval_schemas.py b/python/tests/component_tool_eval_schemas.py