feat:add rag awel operator view metadata. (eosphoros-ai#1174)

dbgpt/core/awel/flow/base.py

@@ -112,6 +112,7 @@ def __init__(self, label: str, description: str):
     "output_parser": _CategoryDetail("Output Parser", "Parse the output of LLM model"),
     "common": _CategoryDetail("Common", "The common operator"),
     "agent": _CategoryDetail("Agent", "The agent operator"),
+    "rag": _CategoryDetail("RAG", "The RAG operator"),
 }
 
 
@@ -124,6 +125,7 @@ class OperatorCategory(str, Enum):
     OUTPUT_PARSER = "output_parser"
     COMMON = "common"
     AGENT = "agent"
+    RAG = "rag"
 
     def label(self) -> str:
         """Get the label of the category."""
@@ -163,6 +165,7 @@ class OperatorType(str, Enum):
     "common": _CategoryDetail("Common", "The common resource"),
     "prompt": _CategoryDetail("Prompt", "The prompt resource"),
     "agent": _CategoryDetail("Agent", "The agent resource"),
+    "rag": _CategoryDetail("RAG", "The  resource"),
 }
 
 
@@ -176,6 +179,7 @@ class ResourceCategory(str, Enum):
     COMMON = "common"
     PROMPT = "prompt"
     AGENT = "agent"
+    RAG = "rag"
 
     def label(self) -> str:
         """Get the label of the category."""

dbgpt/core/awel/trigger/http_trigger.py

@@ -1031,3 +1031,54 @@ def __init__(self, key: str = "user_input", **kwargs):
     async def map(self, request_body: CommonLLMHttpRequestBody) -> Dict[str, Any]:
         """Map the request body to response body."""
         return {self._key: request_body.messages}
+
+
+class RequestedParsedOperator(MapOperator[CommonLLMHttpRequestBody, str]):
+    """User input parsed operator."""
+
+    metadata = ViewMetadata(
+        label="Request Body Parsed To String Operator",
+        name="request_body_to_str__parsed_operator",
+        category=OperatorCategory.COMMON,
+        parameters=[
+            Parameter.build_from(
+                "Key",
+                "key",
+                str,
+                optional=True,
+                default="",
+                description="The key of the dict, link 'user_input'",
+            )
+        ],
+        inputs=[
+            IOField.build_from(
+                "Request Body",
+                "request_body",
+                CommonLLMHttpRequestBody,
+                description="The request body of the API endpoint",
+            )
+        ],
+        outputs=[
+            IOField.build_from(
+                "User Input String",
+                "user_input_str",
+                str,
+                description="The user input dict of the API endpoint",
+            )
+        ],
+        description="User input parsed operator",
+    )
+
+    def __init__(self, key: str = "user_input", **kwargs):
+        """Initialize a UserInputParsedOperator."""
+        self._key = key
+        super().__init__(**kwargs)
+
+    async def map(self, request_body: CommonLLMHttpRequestBody) -> str:
+        """Map the request body to response body."""
+        dict_value = request_body.dict()
+        if not self._key or self._key not in dict_value:
+            raise ValueError(
+                f"Prefix key {self._key} is not a valid key of the request body"
+            )
+        return dict_value[self._key]

dbgpt/core/interface/operators/llm_operator.py

@@ -457,3 +457,42 @@ async def transform_stream(self, output_iter: AsyncIterator[ModelOutput]):
             decoded_unicode = model_output.text.replace("\ufffd", "")
             msg = decoded_unicode.replace("\n", "\\n")
             yield f"data:{msg}\n\n"
+
+
+class StringOutput2ModelOutputOperator(MapOperator[str, ModelOutput]):
+    """Map String to ModelOutput."""
+
+    metadata = ViewMetadata(
+        label="Map String to ModelOutput",
+        name="string_2_model_output_operator",
+        category=OperatorCategory.COMMON,
+        description="Map String to ModelOutput.",
+        parameters=[],
+        inputs=[
+            IOField.build_from(
+                "String",
+                "input_value",
+                str,
+                description="The input value of the operator.",
+            ),
+        ],
+        outputs=[
+            IOField.build_from(
+                "Model Output",
+                "input_value",
+                ModelOutput,
+                description="The input value of the operator.",
+            ),
+        ],
+    )
+
+    def __int__(self, **kwargs):
+        """Create a new operator."""
+        super().__init__(**kwargs)
+
+    async def map(self, input_value: str) -> ModelOutput:
+        """Map the model output to the common response body."""
+        return ModelOutput(
+            text=input_value,
+            error_code=500,
+        )

dbgpt/rag/operators/knowledge.py

@@ -1,26 +1,92 @@
 from typing import Any, List, Optional
 
 from dbgpt.core.awel import MapOperator
+from dbgpt.core.awel.flow import (
+    IOField,
+    OperatorCategory,
+    OptionValue,
+    Parameter,
+    ViewMetadata,
+)
 from dbgpt.core.awel.task.base import IN
 from dbgpt.rag.knowledge.base import Knowledge, KnowledgeType
 from dbgpt.rag.knowledge.factory import KnowledgeFactory
 
 
 class KnowledgeOperator(MapOperator[Any, Any]):
-    """Knowledge Operator."""
+    """Knowledge Factory Operator."""
+
+    metadata = ViewMetadata(
+        label="Knowledge Factory Operator",
+        name="knowledge_operator",
+        category=OperatorCategory.RAG,
+        description="The knowledge operator.",
+        inputs=[
+            IOField.build_from(
+                "knowledge datasource",
+                "knowledge datasource",
+                dict,
+                "knowledge datasource",
+            )
+        ],
+        outputs=[
+            IOField.build_from(
+                "Knowledge",
+                "Knowledge",
+                Knowledge,
+                description="Knowledge",
+            )
+        ],
+        parameters=[
+            Parameter.build_from(
+                label="datasource",
+                name="datasource",
+                type=str,
+                optional=True,
+                default="DOCUMENT",
+                description="datasource",
+            ),
+            Parameter.build_from(
+                label="knowledge_type",
+                name="knowledge type",
+                type=str,
+                optional=True,
+                options=[
+                    OptionValue(
+                        label="DOCUMENT",
+                        name="DOCUMENT",
+                        value=KnowledgeType.DOCUMENT.name,
+                    ),
+                    OptionValue(label="URL", name="URL", value=KnowledgeType.URL.name),
+                    OptionValue(
+                        label="TEXT", name="TEXT", value=KnowledgeType.TEXT.name
+                    ),
+                ],
+                default=KnowledgeType.DOCUMENT.name,
+                description="knowledge type",
+            ),
+        ],
+        documentation_url="https://github.com/openai/openai-python",
+    )
 
     def __init__(
-        self, knowledge_type: Optional[KnowledgeType] = KnowledgeType.DOCUMENT, **kwargs
+        self,
+        datasource: Optional[str] = None,
+        knowledge_type: Optional[str] = KnowledgeType.DOCUMENT.name,
+        **kwargs
     ):
         """Init the query rewrite operator.
         Args:
             knowledge_type: (Optional[KnowledgeType]) The knowledge type.
         """
         super().__init__(**kwargs)
-        self._knowledge_type = knowledge_type
+        self._datasource = datasource
+        self._knowledge_type = KnowledgeType.get_by_value(knowledge_type)
 
     async def map(self, datasource: IN) -> Knowledge:
         """knowledge operator."""
+        if self._datasource:
+            datasource = self._datasource
         return await self.blocking_func_to_async(
             KnowledgeFactory.create, datasource, self._knowledge_type
         )

dbgpt/rag/operators/rewrite.py

@@ -2,13 +2,67 @@
 
 from dbgpt.core import LLMClient
 from dbgpt.core.awel import MapOperator
+from dbgpt.core.awel.flow import IOField, OperatorCategory, Parameter, ViewMetadata
 from dbgpt.core.awel.task.base import IN
 from dbgpt.rag.retriever.rewrite import QueryRewrite
 
 
 class QueryRewriteOperator(MapOperator[Any, Any]):
     """The Rewrite Operator."""
 
+    metadata = ViewMetadata(
+        label="Query Rewrite Operator",
+        name="query_rewrite_operator",
+        category=OperatorCategory.RAG,
+        description="query rewrite operator.",
+        inputs=[
+            IOField.build_from("query_context", "query_context", dict, "query context")
+        ],
+        outputs=[
+            IOField.build_from(
+                "rewritten queries",
+                "queries",
+                List[str],
+                description="rewritten queries",
+            )
+        ],
+        parameters=[
+            Parameter.build_from(
+                "LLM Client",
+                "llm_client",
+                LLMClient,
+                optional=True,
+                default=None,
+                description="The LLM Client.",
+            ),
+            Parameter.build_from(
+                label="model name",
+                name="model_name",
+                type=str,
+                optional=True,
+                default="gpt-3.5-turbo",
+                description="llm model name",
+            ),
+            Parameter.build_from(
+                label="prompt language",
+                name="language",
+                type=str,
+                optional=True,
+                default="en",
+                description="prompt language",
+            ),
+            Parameter.build_from(
+                label="nums",
+                name="nums",
+                type=int,
+                optional=True,
+                default=5,
+                description="rewrite query nums",
+            ),
+        ],
+        documentation_url="https://github.com/openai/openai-python",
+    )
+
     def __init__(
         self,
         llm_client: Optional[LLMClient],

dbgpt/rag/operators/summary.py

@@ -1,12 +1,77 @@
 from typing import Any, Optional
 
 from dbgpt.core import LLMClient
+from dbgpt.core.awel.flow import IOField, OperatorCategory, Parameter, ViewMetadata
 from dbgpt.core.awel.task.base import IN
+from dbgpt.rag.knowledge.base import Knowledge
 from dbgpt.serve.rag.assembler.summary import SummaryAssembler
 from dbgpt.serve.rag.operators.base import AssemblerOperator
 
 
 class SummaryAssemblerOperator(AssemblerOperator[Any, Any]):
+    metadata = ViewMetadata(
+        label="Summary Operator",
+        name="summary_assembler_operator",
+        category=OperatorCategory.RAG,
+        description="The summary assembler operator.",
+        inputs=[
+            IOField.build_from(
+                "Knowledge", "knowledge", Knowledge, "knowledge datasource"
+            )
+        ],
+        outputs=[
+            IOField.build_from(
+                "document summary",
+                "summary",
+                str,
+                description="document summary",
+            )
+        ],
+        parameters=[
+            Parameter.build_from(
+                "LLM Client",
+                "llm_client",
+                LLMClient,
+                optional=True,
+                default=None,
+                description="The LLM Client.",
+            ),
+            Parameter.build_from(
+                label="model name",
+                name="model_name",
+                type=str,
+                optional=True,
+                default="gpt-3.5-turbo",
+                description="llm model name",
+            ),
+            Parameter.build_from(
+                label="prompt language",
+                name="language",
+                type=str,
+                optional=True,
+                default="en",
+                description="prompt language",
+            ),
+            Parameter.build_from(
+                label="max_iteration_with_llm",
+                name="max_iteration_with_llm",
+                type=int,
+                optional=True,
+                default=5,
+                description="prompt language",
+            ),
+            Parameter.build_from(
+                label="concurrency_limit_with_llm",
+                name="concurrency_limit_with_llm",
+                type=int,
+                optional=True,
+                default=3,
+                description="The concurrency limit with llm",
+            ),
+        ],
+        documentation_url="https://github.com/openai/openai-python",
+    )
+
     def __init__(
         self,
         llm_client: Optional[LLMClient],