github/DB-GPT
1
0
Fork 0
mirror of https://github.com/csunny/DB-GPT.git synced 2025-09-13 13:10:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

Native data AI application framework based on AWEL+AGENT (#1152)

Browse Source 
Co-authored-by: Fangyin Cheng <staneyffer@gmail.com>
Co-authored-by: lcx01800250 <lcx01800250@alibaba-inc.com>
Co-authored-by: licunxing <864255598@qq.com>
Co-authored-by: Aralhi <xiaoping0501@gmail.com>
Co-authored-by: xuyuan23 <643854343@qq.com>
Co-authored-by: aries_ckt <916701291@qq.com>
Co-authored-by: hzh97 <2976151305@qq.com>
This commit is contained in:
明天
2024-02-07 17:43:27 +08:00
committed by GitHub
parent dbb9ac83b1
commit d5afa6e206
328 changed files with 22606 additions and 3282 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
dbgpt/core/operators/__init__.py
View File

@@ -25,6 +25,9 @@ from dbgpt.core.interface.operators.prompt_operator import ( # noqa: F401
PromptBuilderOperator,
)
# Flow
from dbgpt.core.operators.flow import * # noqa: F401, F403
__ALL__ = [
"BaseLLM",
"LLMBranchOperator",
@@ -42,4 +45,6 @@ __ALL__ = [
"HistoryDynamicPromptBuilderOperator",
"ChatComposerInput",
"ChatHistoryPromptComposerOperator",
"ConversationComposerOperator",
"PromptFormatDictBuilderOperator",
]

10
dbgpt/core/operators/flow/__init__.py Normal file
View File

@@ -0,0 +1,10 @@
"""Flow operators package."""
from dbgpt.core.operators.flow.composer_operator import ( # noqa: F401
ConversationComposerOperator,
PromptFormatDictBuilderOperator,
)
__ALL__ = [
"ConversationComposerOperator",
"PromptFormatDictBuilderOperator",
]

246
dbgpt/core/operators/flow/composer_operator.py Normal file
View File

@@ -0,0 +1,246 @@
"""Composer operator for AWEL flow.
Usage cases: Some atomized operators cannot be easily modified on the flow page, we
need provide a way to compose the atomized operators to a new operator.
"""
from typing import Any, Dict, Optional, cast
from dbgpt.core.awel import (
DAG,
BaseOperator,
InputOperator,
JoinOperator,
MapOperator,
SimpleCallDataInputSource,
)
from dbgpt.core.awel.flow import IOField, OperatorCategory, Parameter, ViewMetadata
from dbgpt.core.awel.trigger.http_trigger import CommonLLMHttpRequestBody
from dbgpt.core.interface.llm import ModelRequest
from dbgpt.core.interface.message import MessageStorageItem, StorageConversation
from dbgpt.core.interface.operators.llm_operator import (
MergedRequestBuilderOperator,
RequestBuilderOperator,
)
from dbgpt.core.interface.operators.message_operator import (
BufferedConversationMapperOperator,
PreChatHistoryLoadOperator,
)
from dbgpt.core.interface.operators.prompt_operator import HistoryPromptBuilderOperator
from dbgpt.core.interface.prompt import ChatPromptTemplate
from dbgpt.core.interface.storage import StorageInterface
class ConversationComposerOperator(MapOperator[CommonLLMHttpRequestBody, ModelRequest]):
"""A Composer operator for conversation.
Build for AWEL Flow.
"""
metadata = ViewMetadata(
label="Conversation Composer Operator",
name="conversation_composer_operator",
category=OperatorCategory.CONVERSION,
description="A composer operator for conversation.\nIncluding chat history "
"handling, prompt composing, etc. Output is ModelRequest.",
parameters=[
Parameter.build_from(
"Prompt Template",
"prompt_template",
ChatPromptTemplate,
description="The prompt template for the conversation.",
),
Parameter.build_from(
"Human Message Key",
"human_message_key",
str,
optional=True,
default="user_input",
description="The key for human message in the prompt format dict.",
),
Parameter.build_from(
"History Key",
"history_key",
str,
optional=True,
default="chat_history",
description="The chat history key, with chat history message pass to "
"prompt template.",
),
Parameter.build_from(
"Keep Start Rounds",
"keep_start_rounds",
int,
optional=True,
default=None,
description="The start rounds to keep in the chat history.",
),
Parameter.build_from(
"Keep End Rounds",
"keep_end_rounds",
int,
optional=True,
default=10,
description="The end rounds to keep in the chat history.",
),
Parameter.build_from(
"Conversation Storage",
"storage",
StorageInterface,
optional=True,
default=None,
description="The conversation storage(Not include message detail).",
),
Parameter.build_from(
"Message Storage",
"message_storage",
StorageInterface,
optional=True,
default=None,
description="The message storage.",
),
],
inputs=[
IOField.build_from(
"Common LLM Http Request Body",
"common_llm_http_request_body",
CommonLLMHttpRequestBody,
description="The common LLM http request body.",
)
],
outputs=[
IOField.build_from(
"Model Request",
"model_request",
ModelRequest,
description="The model request with chat history prompt.",
)
],
)
def __init__(
self,
prompt_template: ChatPromptTemplate,
human_message_key: str = "user_input",
history_key: str = "chat_history",
keep_start_rounds: Optional[int] = None,
keep_end_rounds: Optional[int] = None,
storage: Optional[StorageInterface[StorageConversation, Any]] = None,
message_storage: Optional[StorageInterface[MessageStorageItem, Any]] = None,
**kwargs
):
"""Create a new instance of ConversationComposerOperator."""
super().__init__(**kwargs)
self._prompt_template = prompt_template
self._human_message_key = human_message_key
self._history_key = history_key
self._keep_start_rounds = keep_start_rounds
self._keep_end_rounds = keep_end_rounds
self._storage = storage
self._message_storage = message_storage
self._sub_compose_dag = self._build_composer_dag()
async def map(self, input_value: CommonLLMHttpRequestBody) -> ModelRequest:
"""Receive the common LLM http request body, and build the model request."""
end_node: BaseOperator = cast(BaseOperator, self._sub_compose_dag.leaf_nodes[0])
# Sub dag, use the same dag context in the parent dag
return await end_node.call(
call_data=input_value, dag_ctx=self.current_dag_context
)
def _build_composer_dag(self):
with DAG("dbgpt_awel_chat_history_prompt_composer") as composer_dag:
input_task = InputOperator(input_source=SimpleCallDataInputSource())
# Load and store chat history, default use InMemoryStorage.
chat_history_load_task = PreChatHistoryLoadOperator(
storage=self._storage, message_storage=self._message_storage
)
# History transform task, here we keep last 5 round messages
history_transform_task = BufferedConversationMapperOperator(
keep_start_rounds=self._keep_start_rounds,
keep_end_rounds=self._keep_end_rounds,
)
history_prompt_build_task = HistoryPromptBuilderOperator(
prompt=self._prompt_template, history_key=self._history_key
)
prompt_build_task = PromptFormatDictBuilderOperator(
human_message_key=self._human_message_key
)
model_request_build_task: JoinOperator[
ModelRequest
] = MergedRequestBuilderOperator()
# Build composer dag
(
input_task
>> MapOperator(lambda x: x.context)
>> chat_history_load_task
>> history_transform_task
>> history_prompt_build_task
)
input_task >> prompt_build_task >> history_prompt_build_task
input_task >> RequestBuilderOperator() >> model_request_build_task
history_prompt_build_task >> model_request_build_task
return composer_dag
async def after_dag_end(self):
"""Execute after dag end."""
# Should call after_dag_end() of sub dag
await self._sub_compose_dag._after_dag_end()
class PromptFormatDictBuilderOperator(
MapOperator[CommonLLMHttpRequestBody, Dict[str, Any]]
):
"""Prompt format dict builder operator for AWEL flow.
Receive the common LLM http request body, and build the prompt format dict.
"""
metadata = ViewMetadata(
label="Prompt Format Dict Builder Operator",
name="prompt_format_dict_builder_operator",
category=OperatorCategory.CONVERSION,
description="A operator to build prompt format dict from common LLM http "
"request body.",
parameters=[
Parameter.build_from(
"Human Message Key",
"human_message_key",
str,
optional=True,
default="user_input",
description="The key for human message in the prompt format dict.",
)
],
inputs=[
IOField.build_from(
"Common LLM Http Request Body",
"common_llm_http_request_body",
CommonLLMHttpRequestBody,
description="The common LLM http request body.",
)
],
outputs=[
IOField.build_from(
"Prompt Format Dict",
"prompt_format_dict",
Dict[str, Any],
description="The prompt format dict.",
)
],
)
def __init__(self, human_message_key: str = "user_input", **kwargs):
"""Create a new instance of PromptFormatDictBuilderOperator."""
self._human_message_key = human_message_key
super().__init__(**kwargs)
async def map(self, input_value: CommonLLMHttpRequestBody) -> Dict[str, Any]:
"""Build prompt format dict from common LLM http request body."""
extra_data = input_value.context.extra if input_value.context.extra else {}
return {
self._human_message_key: input_value.messages,
**extra_data,
}