mirror of
https://github.com/danswer-ai/danswer.git
synced 2025-04-01 16:38:11 +02:00
Combining Search and Chat Backend (#3273)
* k * k * fix slack issues * rebase * k
This commit is contained in:
Yuhong Sun
GitHub
parent
13f6e8a6b4
commit
aa1c4c635a
backend
alembic/versions
danswer
ee/danswer
chat
danswerbot/slack/handlers
db
server
query_and_chat
query_history
reporting
tests
integration
regression/answer_quality
unit/danswer/llm/answering
@ -0,0 +1,36 @@
|
||||
"""Combine Search and Chat
|
||||
|
||||
Revision ID: 9f696734098f
|
||||
Revises: a8c2065484e6
|
||||
Create Date: 2024-11-27 15:32:19.694972
|
||||
|
||||
"""
|
||||
from alembic import op
|
||||
import sqlalchemy as sa
|
||||
|
||||
# revision identifiers, used by Alembic.
|
||||
revision = "9f696734098f"
|
||||
down_revision = "a8c2065484e6"
|
||||
branch_labels = None
|
||||
depends_on = None
|
||||
|
||||
|
||||
def upgrade() -> None:
|
||||
op.alter_column("chat_session", "description", nullable=True)
|
||||
op.drop_column("chat_session", "one_shot")
|
||||
op.drop_column("slack_channel_config", "response_type")
|
||||
|
||||
|
||||
def downgrade() -> None:
|
||||
op.execute("UPDATE chat_session SET description = '' WHERE description IS NULL")
|
||||
op.alter_column("chat_session", "description", nullable=False)
|
||||
op.add_column(
|
||||
"chat_session",
|
||||
sa.Column("one_shot", sa.Boolean(), nullable=False, server_default=sa.false()),
|
||||
)
|
||||
op.add_column(
|
||||
"slack_channel_config",
|
||||
sa.Column(
|
||||
"response_type", sa.String(), nullable=False, server_default="citations"
|
||||
),
|
||||
)
|
||||
@ -2,20 +2,79 @@ import re
|
||||
from typing import cast
|
||||
from uuid import UUID
|
||||
|
||||
from fastapi import HTTPException
|
||||
from fastapi.datastructures import Headers
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.auth.users import is_user_admin
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import LlmDoc
|
||||
from danswer.chat.models import PersonaOverrideConfig
|
||||
from danswer.chat.models import ThreadMessage
|
||||
from danswer.configs.constants import DEFAULT_PERSONA_ID
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.context.search.models import InferenceSection
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.db.chat import create_chat_session
|
||||
from danswer.db.chat import get_chat_messages_by_session
|
||||
from danswer.db.llm import fetch_existing_doc_sets
|
||||
from danswer.db.llm import fetch_existing_tools
|
||||
from danswer.db.models import ChatMessage
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import Prompt
|
||||
from danswer.db.models import Tool
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_prompts_by_ids
|
||||
from danswer.llm.answering.models import PreviousMessage
|
||||
from danswer.natural_language_processing.utils import BaseTokenizer
|
||||
from danswer.server.query_and_chat.models import CreateChatMessageRequest
|
||||
from danswer.tools.tool_implementations.custom.custom_tool import (
|
||||
build_custom_tools_from_openapi_schema_and_headers,
|
||||
)
|
||||
from danswer.utils.logger import setup_logger
|
||||
|
||||
logger = setup_logger()
|
||||
|
||||
|
||||
def prepare_chat_message_request(
|
||||
message_text: str,
|
||||
user: User | None,
|
||||
persona_id: int | None,
|
||||
# Does the question need to have a persona override
|
||||
persona_override_config: PersonaOverrideConfig | None,
|
||||
prompt: Prompt | None,
|
||||
message_ts_to_respond_to: str | None,
|
||||
retrieval_details: RetrievalDetails | None,
|
||||
rerank_settings: RerankingDetails | None,
|
||||
db_session: Session,
|
||||
) -> CreateChatMessageRequest:
|
||||
# Typically used for one shot flows like SlackBot or non-chat API endpoint use cases
|
||||
new_chat_session = create_chat_session(
|
||||
db_session=db_session,
|
||||
description=None,
|
||||
user_id=user.id if user else None,
|
||||
# If using an override, this id will be ignored later on
|
||||
persona_id=persona_id or DEFAULT_PERSONA_ID,
|
||||
danswerbot_flow=True,
|
||||
slack_thread_id=message_ts_to_respond_to,
|
||||
)
|
||||
|
||||
return CreateChatMessageRequest(
|
||||
chat_session_id=new_chat_session.id,
|
||||
parent_message_id=None, # It's a standalone chat session each time
|
||||
message=message_text,
|
||||
file_descriptors=[], # Currently SlackBot/answer api do not support files in the context
|
||||
prompt_id=prompt.id if prompt else None,
|
||||
# Can always override the persona for the single query, if it's a normal persona
|
||||
# then it will be treated the same
|
||||
persona_override_config=persona_override_config,
|
||||
search_doc_ids=None,
|
||||
retrieval_options=retrieval_details,
|
||||
rerank_settings=rerank_settings,
|
||||
)
|
||||
|
||||
|
||||
def llm_doc_from_inference_section(inference_section: InferenceSection) -> LlmDoc:
|
||||
return LlmDoc(
|
||||
document_id=inference_section.center_chunk.document_id,
|
||||
@ -35,6 +94,45 @@ def llm_doc_from_inference_section(inference_section: InferenceSection) -> LlmDo
|
||||
)
|
||||
|
||||
|
||||
def combine_message_thread(
|
||||
messages: list[ThreadMessage],
|
||||
max_tokens: int | None,
|
||||
llm_tokenizer: BaseTokenizer,
|
||||
) -> str:
|
||||
"""Used to create a single combined message context from threads"""
|
||||
if not messages:
|
||||
return ""
|
||||
|
||||
message_strs: list[str] = []
|
||||
total_token_count = 0
|
||||
|
||||
for message in reversed(messages):
|
||||
if message.role == MessageType.USER:
|
||||
role_str = message.role.value.upper()
|
||||
if message.sender:
|
||||
role_str += " " + message.sender
|
||||
else:
|
||||
# Since other messages might have the user identifying information
|
||||
# better to use Unknown for symmetry
|
||||
role_str += " Unknown"
|
||||
else:
|
||||
role_str = message.role.value.upper()
|
||||
|
||||
msg_str = f"{role_str}:\n{message.message}"
|
||||
message_token_count = len(llm_tokenizer.encode(msg_str))
|
||||
|
||||
if (
|
||||
max_tokens is not None
|
||||
and total_token_count + message_token_count > max_tokens
|
||||
):
|
||||
break
|
||||
|
||||
message_strs.insert(0, msg_str)
|
||||
total_token_count += message_token_count
|
||||
|
||||
return "\n\n".join(message_strs)
|
||||
|
||||
|
||||
def create_chat_chain(
|
||||
chat_session_id: UUID,
|
||||
db_session: Session,
|
||||
@ -197,3 +295,71 @@ def extract_headers(
|
||||
if lowercase_key in headers:
|
||||
extracted_headers[lowercase_key] = headers[lowercase_key]
|
||||
return extracted_headers
|
||||
|
||||
|
||||
def create_temporary_persona(
|
||||
persona_config: PersonaOverrideConfig, db_session: Session, user: User | None = None
|
||||
) -> Persona:
|
||||
if not is_user_admin(user):
|
||||
raise HTTPException(
|
||||
status_code=403,
|
||||
detail="User is not authorized to create a persona in one shot queries",
|
||||
)
|
||||
|
||||
"""Create a temporary Persona object from the provided configuration."""
|
||||
persona = Persona(
|
||||
name=persona_config.name,
|
||||
description=persona_config.description,
|
||||
num_chunks=persona_config.num_chunks,
|
||||
llm_relevance_filter=persona_config.llm_relevance_filter,
|
||||
llm_filter_extraction=persona_config.llm_filter_extraction,
|
||||
recency_bias=persona_config.recency_bias,
|
||||
llm_model_provider_override=persona_config.llm_model_provider_override,
|
||||
llm_model_version_override=persona_config.llm_model_version_override,
|
||||
)
|
||||
|
||||
if persona_config.prompts:
|
||||
persona.prompts = [
|
||||
Prompt(
|
||||
name=p.name,
|
||||
description=p.description,
|
||||
system_prompt=p.system_prompt,
|
||||
task_prompt=p.task_prompt,
|
||||
include_citations=p.include_citations,
|
||||
datetime_aware=p.datetime_aware,
|
||||
)
|
||||
for p in persona_config.prompts
|
||||
]
|
||||
elif persona_config.prompt_ids:
|
||||
persona.prompts = get_prompts_by_ids(
|
||||
db_session=db_session, prompt_ids=persona_config.prompt_ids
|
||||
)
|
||||
|
||||
persona.tools = []
|
||||
if persona_config.custom_tools_openapi:
|
||||
for schema in persona_config.custom_tools_openapi:
|
||||
tools = cast(
|
||||
list[Tool],
|
||||
build_custom_tools_from_openapi_schema_and_headers(schema),
|
||||
)
|
||||
persona.tools.extend(tools)
|
||||
|
||||
if persona_config.tools:
|
||||
tool_ids = [tool.id for tool in persona_config.tools]
|
||||
persona.tools.extend(
|
||||
fetch_existing_tools(db_session=db_session, tool_ids=tool_ids)
|
||||
)
|
||||
|
||||
if persona_config.tool_ids:
|
||||
persona.tools.extend(
|
||||
fetch_existing_tools(
|
||||
db_session=db_session, tool_ids=persona_config.tool_ids
|
||||
)
|
||||
)
|
||||
|
||||
fetched_docs = fetch_existing_doc_sets(
|
||||
db_session=db_session, doc_ids=persona_config.document_set_ids
|
||||
)
|
||||
persona.document_sets = fetched_docs
|
||||
|
||||
return persona
|
||||
|
||||
@ -4,12 +4,14 @@ from enum import Enum
|
||||
from typing import Any
|
||||
|
||||
from pydantic import BaseModel
|
||||
from pydantic import Field
|
||||
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.context.search.enums import QueryFlow
|
||||
from danswer.context.search.enums import RecencyBiasSetting
|
||||
from danswer.context.search.enums import SearchType
|
||||
from danswer.context.search.models import RetrievalDocs
|
||||
from danswer.context.search.models import SearchResponse
|
||||
from danswer.tools.tool_implementations.custom.base_tool_types import ToolResultType
|
||||
|
||||
|
||||
@ -118,20 +120,6 @@ class StreamingError(BaseModel):
|
||||
stack_trace: str | None = None
|
||||
|
||||
|
||||
class DanswerQuote(BaseModel):
|
||||
# This is during inference so everything is a string by this point
|
||||
quote: str
|
||||
document_id: str
|
||||
link: str | None
|
||||
source_type: str
|
||||
semantic_identifier: str
|
||||
blurb: str
|
||||
|
||||
|
||||
class DanswerQuotes(BaseModel):
|
||||
quotes: list[DanswerQuote]
|
||||
|
||||
|
||||
class DanswerContext(BaseModel):
|
||||
content: str
|
||||
document_id: str
|
||||
@ -147,14 +135,20 @@ class DanswerAnswer(BaseModel):
|
||||
answer: str | None
|
||||
|
||||
|
||||
class QAResponse(SearchResponse, DanswerAnswer):
|
||||
quotes: list[DanswerQuote] | None
|
||||
contexts: list[DanswerContexts] | None
|
||||
predicted_flow: QueryFlow
|
||||
predicted_search: SearchType
|
||||
eval_res_valid: bool | None = None
|
||||
class ThreadMessage(BaseModel):
|
||||
message: str
|
||||
sender: str | None = None
|
||||
role: MessageType = MessageType.USER
|
||||
|
||||
|
||||
class ChatDanswerBotResponse(BaseModel):
|
||||
answer: str | None = None
|
||||
citations: list[CitationInfo] | None = None
|
||||
docs: QADocsResponse | None = None
|
||||
llm_selected_doc_indices: list[int] | None = None
|
||||
error_msg: str | None = None
|
||||
chat_message_id: int | None = None
|
||||
answer_valid: bool = True # Reflexion result, default True if Reflexion not run
|
||||
|
||||
|
||||
class FileChatDisplay(BaseModel):
|
||||
@ -166,9 +160,41 @@ class CustomToolResponse(BaseModel):
|
||||
tool_name: str
|
||||
|
||||
|
||||
class ToolConfig(BaseModel):
|
||||
id: int
|
||||
|
||||
|
||||
class PromptOverrideConfig(BaseModel):
|
||||
name: str
|
||||
description: str = ""
|
||||
system_prompt: str
|
||||
task_prompt: str = ""
|
||||
include_citations: bool = True
|
||||
datetime_aware: bool = True
|
||||
|
||||
|
||||
class PersonaOverrideConfig(BaseModel):
|
||||
name: str
|
||||
description: str
|
||||
search_type: SearchType = SearchType.SEMANTIC
|
||||
num_chunks: float | None = None
|
||||
llm_relevance_filter: bool = False
|
||||
llm_filter_extraction: bool = False
|
||||
recency_bias: RecencyBiasSetting = RecencyBiasSetting.AUTO
|
||||
llm_model_provider_override: str | None = None
|
||||
llm_model_version_override: str | None = None
|
||||
|
||||
prompts: list[PromptOverrideConfig] = Field(default_factory=list)
|
||||
prompt_ids: list[int] = Field(default_factory=list)
|
||||
|
||||
document_set_ids: list[int] = Field(default_factory=list)
|
||||
tools: list[ToolConfig] = Field(default_factory=list)
|
||||
tool_ids: list[int] = Field(default_factory=list)
|
||||
custom_tools_openapi: list[dict[str, Any]] = Field(default_factory=list)
|
||||
|
||||
|
||||
AnswerQuestionPossibleReturn = (
|
||||
DanswerAnswerPiece
|
||||
| DanswerQuotes
|
||||
| CitationInfo
|
||||
| DanswerContexts
|
||||
| FileChatDisplay
|
||||
|
||||
@ -7,10 +7,13 @@ from typing import cast
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.chat.chat_utils import create_chat_chain
|
||||
from danswer.chat.chat_utils import create_temporary_persona
|
||||
from danswer.chat.models import AllCitations
|
||||
from danswer.chat.models import ChatDanswerBotResponse
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import CustomToolResponse
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerContexts
|
||||
from danswer.chat.models import FileChatDisplay
|
||||
from danswer.chat.models import FinalUsedContextDocsResponse
|
||||
from danswer.chat.models import LLMRelevanceFilterResponse
|
||||
@ -102,6 +105,7 @@ from danswer.tools.tool_implementations.internet_search.internet_search_tool imp
|
||||
from danswer.tools.tool_implementations.search.search_tool import (
|
||||
FINAL_CONTEXT_DOCUMENTS_ID,
|
||||
)
|
||||
from danswer.tools.tool_implementations.search.search_tool import SEARCH_DOC_CONTENT_ID
|
||||
from danswer.tools.tool_implementations.search.search_tool import (
|
||||
SEARCH_RESPONSE_SUMMARY_ID,
|
||||
)
|
||||
@ -113,8 +117,10 @@ from danswer.tools.tool_implementations.search.search_tool import (
|
||||
from danswer.tools.tool_runner import ToolCallFinalResult
|
||||
from danswer.utils.logger import setup_logger
|
||||
from danswer.utils.long_term_log import LongTermLogger
|
||||
from danswer.utils.timing import log_function_time
|
||||
from danswer.utils.timing import log_generator_function_time
|
||||
|
||||
|
||||
logger = setup_logger()
|
||||
|
||||
|
||||
@ -256,6 +262,7 @@ def _get_force_search_settings(
|
||||
ChatPacket = (
|
||||
StreamingError
|
||||
| QADocsResponse
|
||||
| DanswerContexts
|
||||
| LLMRelevanceFilterResponse
|
||||
| FinalUsedContextDocsResponse
|
||||
| ChatMessageDetail
|
||||
@ -286,6 +293,8 @@ def stream_chat_message_objects(
|
||||
custom_tool_additional_headers: dict[str, str] | None = None,
|
||||
is_connected: Callable[[], bool] | None = None,
|
||||
enforce_chat_session_id_for_search_docs: bool = True,
|
||||
bypass_acl: bool = False,
|
||||
include_contexts: bool = False,
|
||||
) -> ChatPacketStream:
|
||||
"""Streams in order:
|
||||
1. [conditional] Retrieved documents if a search needs to be run
|
||||
@ -322,17 +331,31 @@ def stream_chat_message_objects(
|
||||
metadata={"user_id": str(user_id), "chat_session_id": str(chat_session_id)}
|
||||
)
|
||||
|
||||
# use alternate persona if alternative assistant id is passed in
|
||||
if alternate_assistant_id is not None:
|
||||
# Allows users to specify a temporary persona (assistant) in the chat session
|
||||
# this takes highest priority since it's user specified
|
||||
persona = get_persona_by_id(
|
||||
alternate_assistant_id,
|
||||
user=user,
|
||||
db_session=db_session,
|
||||
is_for_edit=False,
|
||||
)
|
||||
elif new_msg_req.persona_override_config:
|
||||
# Certain endpoints allow users to specify arbitrary persona settings
|
||||
# this should never conflict with the alternate_assistant_id
|
||||
persona = persona = create_temporary_persona(
|
||||
db_session=db_session,
|
||||
persona_config=new_msg_req.persona_override_config,
|
||||
user=user,
|
||||
)
|
||||
else:
|
||||
persona = chat_session.persona
|
||||
|
||||
if not persona:
|
||||
raise RuntimeError("No persona specified or found for chat session")
|
||||
|
||||
# If a prompt override is specified via the API, use that with highest priority
|
||||
# but for saving it, we are just mapping it to an existing prompt
|
||||
prompt_id = new_msg_req.prompt_id
|
||||
if prompt_id is None and persona.prompts:
|
||||
prompt_id = sorted(persona.prompts, key=lambda x: x.id)[-1].id
|
||||
@ -555,19 +578,34 @@ def stream_chat_message_objects(
|
||||
reserved_message_id=reserved_message_id,
|
||||
)
|
||||
|
||||
if not final_msg.prompt:
|
||||
raise RuntimeError("No Prompt found")
|
||||
|
||||
prompt_config = (
|
||||
PromptConfig.from_model(
|
||||
final_msg.prompt,
|
||||
prompt_override=(
|
||||
new_msg_req.prompt_override or chat_session.prompt_override
|
||||
),
|
||||
prompt_override = new_msg_req.prompt_override or chat_session.prompt_override
|
||||
if new_msg_req.persona_override_config:
|
||||
prompt_config = PromptConfig(
|
||||
system_prompt=new_msg_req.persona_override_config.prompts[
|
||||
0
|
||||
].system_prompt,
|
||||
task_prompt=new_msg_req.persona_override_config.prompts[0].task_prompt,
|
||||
datetime_aware=new_msg_req.persona_override_config.prompts[
|
||||
0
|
||||
].datetime_aware,
|
||||
include_citations=new_msg_req.persona_override_config.prompts[
|
||||
0
|
||||
].include_citations,
|
||||
)
|
||||
if not persona
|
||||
else PromptConfig.from_model(persona.prompts[0])
|
||||
)
|
||||
elif prompt_override:
|
||||
if not final_msg.prompt:
|
||||
raise ValueError(
|
||||
"Prompt override cannot be applied, no base prompt found."
|
||||
)
|
||||
prompt_config = PromptConfig.from_model(
|
||||
final_msg.prompt,
|
||||
prompt_override=prompt_override,
|
||||
)
|
||||
elif final_msg.prompt:
|
||||
prompt_config = PromptConfig.from_model(final_msg.prompt)
|
||||
else:
|
||||
prompt_config = PromptConfig.from_model(persona.prompts[0])
|
||||
|
||||
answer_style_config = AnswerStyleConfig(
|
||||
citation_config=CitationConfig(
|
||||
all_docs_useful=selected_db_search_docs is not None
|
||||
@ -587,11 +625,13 @@ def stream_chat_message_objects(
|
||||
answer_style_config=answer_style_config,
|
||||
document_pruning_config=document_pruning_config,
|
||||
retrieval_options=retrieval_options or RetrievalDetails(),
|
||||
rerank_settings=new_msg_req.rerank_settings,
|
||||
selected_sections=selected_sections,
|
||||
chunks_above=new_msg_req.chunks_above,
|
||||
chunks_below=new_msg_req.chunks_below,
|
||||
full_doc=new_msg_req.full_doc,
|
||||
latest_query_files=latest_query_files,
|
||||
bypass_acl=bypass_acl,
|
||||
),
|
||||
internet_search_tool_config=InternetSearchToolConfig(
|
||||
answer_style_config=answer_style_config,
|
||||
@ -737,6 +777,8 @@ def stream_chat_message_objects(
|
||||
response=custom_tool_response.tool_result,
|
||||
tool_name=custom_tool_response.tool_name,
|
||||
)
|
||||
elif packet.id == SEARCH_DOC_CONTENT_ID and include_contexts:
|
||||
yield cast(DanswerContexts, packet.response)
|
||||
|
||||
elif isinstance(packet, StreamStopInfo):
|
||||
pass
|
||||
@ -845,3 +887,30 @@ def stream_chat_message(
|
||||
)
|
||||
for obj in objects:
|
||||
yield get_json_line(obj.model_dump())
|
||||
|
||||
|
||||
@log_function_time()
|
||||
def gather_stream_for_slack(
|
||||
packets: ChatPacketStream,
|
||||
) -> ChatDanswerBotResponse:
|
||||
response = ChatDanswerBotResponse()
|
||||
|
||||
answer = ""
|
||||
for packet in packets:
|
||||
if isinstance(packet, DanswerAnswerPiece) and packet.answer_piece:
|
||||
answer += packet.answer_piece
|
||||
elif isinstance(packet, QADocsResponse):
|
||||
response.docs = packet
|
||||
elif isinstance(packet, StreamingError):
|
||||
response.error_msg = packet.error
|
||||
elif isinstance(packet, ChatMessageDetail):
|
||||
response.chat_message_id = packet.message_id
|
||||
elif isinstance(packet, LLMRelevanceFilterResponse):
|
||||
response.llm_selected_doc_indices = packet.llm_selected_doc_indices
|
||||
elif isinstance(packet, AllCitations):
|
||||
response.citations = packet.citations
|
||||
|
||||
if answer:
|
||||
response.answer = answer
|
||||
|
||||
return response
|
||||
|
||||
@ -522,3 +522,6 @@ API_KEY_HASH_ROUNDS = (
|
||||
|
||||
POD_NAME = os.environ.get("POD_NAME")
|
||||
POD_NAMESPACE = os.environ.get("POD_NAMESPACE")
|
||||
|
||||
|
||||
DEV_MODE = os.environ.get("DEV_MODE", "").lower() == "true"
|
||||
|
||||
@ -31,6 +31,8 @@ DISABLED_GEN_AI_MSG = (
|
||||
"You can still use Danswer as a search engine."
|
||||
)
|
||||
|
||||
DEFAULT_PERSONA_ID = 0
|
||||
|
||||
# Postgres connection constants for application_name
|
||||
POSTGRES_WEB_APP_NAME = "web"
|
||||
POSTGRES_INDEXER_APP_NAME = "indexer"
|
||||
|
||||
@ -4,11 +4,8 @@ import os
|
||||
# Danswer Slack Bot Configs
|
||||
#####
|
||||
DANSWER_BOT_NUM_RETRIES = int(os.environ.get("DANSWER_BOT_NUM_RETRIES", "5"))
|
||||
DANSWER_BOT_ANSWER_GENERATION_TIMEOUT = int(
|
||||
os.environ.get("DANSWER_BOT_ANSWER_GENERATION_TIMEOUT", "90")
|
||||
)
|
||||
# How much of the available input context can be used for thread context
|
||||
DANSWER_BOT_TARGET_CHUNK_PERCENTAGE = 512 * 2 / 3072
|
||||
MAX_THREAD_CONTEXT_PERCENTAGE = 512 * 2 / 3072
|
||||
# Number of docs to display in "Reference Documents"
|
||||
DANSWER_BOT_NUM_DOCS_TO_DISPLAY = int(
|
||||
os.environ.get("DANSWER_BOT_NUM_DOCS_TO_DISPLAY", "5")
|
||||
@ -47,17 +44,6 @@ DANSWER_BOT_DISPLAY_ERROR_MSGS = os.environ.get(
|
||||
DANSWER_BOT_RESPOND_EVERY_CHANNEL = (
|
||||
os.environ.get("DANSWER_BOT_RESPOND_EVERY_CHANNEL", "").lower() == "true"
|
||||
)
|
||||
# Add a second LLM call post Answer to verify if the Answer is valid
|
||||
# Throws out answers that don't directly or fully answer the user query
|
||||
# This is the default for all DanswerBot channels unless the channel is configured individually
|
||||
# Set/unset by "Hide Non Answers"
|
||||
ENABLE_DANSWERBOT_REFLEXION = (
|
||||
os.environ.get("ENABLE_DANSWERBOT_REFLEXION", "").lower() == "true"
|
||||
)
|
||||
# Currently not support chain of thought, probably will add back later
|
||||
DANSWER_BOT_DISABLE_COT = True
|
||||
# if set, will default DanswerBot to use quotes and reference documents
|
||||
DANSWER_BOT_USE_QUOTES = os.environ.get("DANSWER_BOT_USE_QUOTES", "").lower() == "true"
|
||||
|
||||
# Maximum Questions Per Minute, Default Uncapped
|
||||
DANSWER_BOT_MAX_QPM = int(os.environ.get("DANSWER_BOT_MAX_QPM") or 0) or None
|
||||
|
||||
@ -16,7 +16,7 @@ from slack_sdk.models.blocks import SectionBlock
|
||||
from slack_sdk.models.blocks.basic_components import MarkdownTextObject
|
||||
from slack_sdk.models.blocks.block_elements import ImageElement
|
||||
|
||||
from danswer.chat.models import DanswerQuote
|
||||
from danswer.chat.models import ChatDanswerBotResponse
|
||||
from danswer.configs.app_configs import DISABLE_GENERATIVE_AI
|
||||
from danswer.configs.app_configs import WEB_DOMAIN
|
||||
from danswer.configs.constants import DocumentSource
|
||||
@ -40,10 +40,7 @@ from danswer.danswerbot.slack.utils import translate_vespa_highlight_to_slack
|
||||
from danswer.db.chat import get_chat_session_by_message_id
|
||||
from danswer.db.engine import get_session_with_tenant
|
||||
from danswer.db.models import ChannelConfig
|
||||
from danswer.db.models import Persona
|
||||
from danswer.one_shot_answer.models import OneShotQAResponse
|
||||
from danswer.utils.text_processing import decode_escapes
|
||||
from danswer.utils.text_processing import replace_whitespaces_w_space
|
||||
|
||||
_MAX_BLURB_LEN = 45
|
||||
|
||||
@ -327,7 +324,7 @@ def _build_sources_blocks(
|
||||
|
||||
|
||||
def _priority_ordered_documents_blocks(
|
||||
answer: OneShotQAResponse,
|
||||
answer: ChatDanswerBotResponse,
|
||||
) -> list[Block]:
|
||||
docs_response = answer.docs if answer.docs else None
|
||||
top_docs = docs_response.top_documents if docs_response else []
|
||||
@ -350,7 +347,7 @@ def _priority_ordered_documents_blocks(
|
||||
|
||||
|
||||
def _build_citations_blocks(
|
||||
answer: OneShotQAResponse,
|
||||
answer: ChatDanswerBotResponse,
|
||||
) -> list[Block]:
|
||||
docs_response = answer.docs if answer.docs else None
|
||||
top_docs = docs_response.top_documents if docs_response else []
|
||||
@ -369,51 +366,8 @@ def _build_citations_blocks(
|
||||
return citations_block
|
||||
|
||||
|
||||
def _build_quotes_block(
|
||||
quotes: list[DanswerQuote],
|
||||
) -> list[Block]:
|
||||
quote_lines: list[str] = []
|
||||
doc_to_quotes: dict[str, list[str]] = {}
|
||||
doc_to_link: dict[str, str] = {}
|
||||
doc_to_sem_id: dict[str, str] = {}
|
||||
for q in quotes:
|
||||
quote = q.quote
|
||||
doc_id = q.document_id
|
||||
doc_link = q.link
|
||||
doc_name = q.semantic_identifier
|
||||
if doc_link and doc_name and doc_id and quote:
|
||||
if doc_id not in doc_to_quotes:
|
||||
doc_to_quotes[doc_id] = [quote]
|
||||
doc_to_link[doc_id] = doc_link
|
||||
doc_to_sem_id[doc_id] = (
|
||||
doc_name
|
||||
if q.source_type != DocumentSource.SLACK.value
|
||||
else "#" + doc_name
|
||||
)
|
||||
else:
|
||||
doc_to_quotes[doc_id].append(quote)
|
||||
|
||||
for doc_id, quote_strs in doc_to_quotes.items():
|
||||
quotes_str_clean = [
|
||||
replace_whitespaces_w_space(q_str).strip() for q_str in quote_strs
|
||||
]
|
||||
longest_quotes = sorted(quotes_str_clean, key=len, reverse=True)[:5]
|
||||
single_quote_str = "\n".join([f"```{q_str}```" for q_str in longest_quotes])
|
||||
link = doc_to_link[doc_id]
|
||||
sem_id = doc_to_sem_id[doc_id]
|
||||
quote_lines.append(
|
||||
f"<{link}|{sem_id}>:\n{remove_slack_text_interactions(single_quote_str)}"
|
||||
)
|
||||
|
||||
if not doc_to_quotes:
|
||||
return []
|
||||
|
||||
return [SectionBlock(text="*Relevant Snippets*\n" + "\n".join(quote_lines))]
|
||||
|
||||
|
||||
def _build_qa_response_blocks(
|
||||
answer: OneShotQAResponse,
|
||||
skip_quotes: bool = False,
|
||||
answer: ChatDanswerBotResponse,
|
||||
process_message_for_citations: bool = False,
|
||||
) -> list[Block]:
|
||||
retrieval_info = answer.docs
|
||||
@ -422,13 +376,10 @@ def _build_qa_response_blocks(
|
||||
raise RuntimeError("Failed to retrieve docs, cannot answer question.")
|
||||
|
||||
formatted_answer = format_slack_message(answer.answer) if answer.answer else None
|
||||
quotes = answer.quotes.quotes if answer.quotes else None
|
||||
|
||||
if DISABLE_GENERATIVE_AI:
|
||||
return []
|
||||
|
||||
quotes_blocks: list[Block] = []
|
||||
|
||||
filter_block: Block | None = None
|
||||
if (
|
||||
retrieval_info.applied_time_cutoff
|
||||
@ -471,16 +422,6 @@ def _build_qa_response_blocks(
|
||||
answer_blocks = [
|
||||
SectionBlock(text=text) for text in _split_text(answer_processed)
|
||||
]
|
||||
if quotes:
|
||||
quotes_blocks = _build_quotes_block(quotes)
|
||||
|
||||
# if no quotes OR `_build_quotes_block()` did not give back any blocks
|
||||
if not quotes_blocks:
|
||||
quotes_blocks = [
|
||||
SectionBlock(
|
||||
text="*Warning*: no sources were quoted for this answer, so it may be unreliable 😔"
|
||||
)
|
||||
]
|
||||
|
||||
response_blocks: list[Block] = []
|
||||
|
||||
@ -489,9 +430,6 @@ def _build_qa_response_blocks(
|
||||
|
||||
response_blocks.extend(answer_blocks)
|
||||
|
||||
if not skip_quotes:
|
||||
response_blocks.extend(quotes_blocks)
|
||||
|
||||
return response_blocks
|
||||
|
||||
|
||||
@ -567,10 +505,9 @@ def build_follow_up_resolved_blocks(
|
||||
|
||||
|
||||
def build_slack_response_blocks(
|
||||
answer: ChatDanswerBotResponse,
|
||||
tenant_id: str | None,
|
||||
message_info: SlackMessageInfo,
|
||||
answer: OneShotQAResponse,
|
||||
persona: Persona | None,
|
||||
channel_conf: ChannelConfig | None,
|
||||
use_citations: bool,
|
||||
feedback_reminder_id: str | None,
|
||||
@ -587,7 +524,6 @@ def build_slack_response_blocks(
|
||||
|
||||
answer_blocks = _build_qa_response_blocks(
|
||||
answer=answer,
|
||||
skip_quotes=persona is not None or use_citations,
|
||||
process_message_for_citations=use_citations,
|
||||
)
|
||||
|
||||
@ -617,8 +553,7 @@ def build_slack_response_blocks(
|
||||
|
||||
citations_blocks = []
|
||||
document_blocks = []
|
||||
if use_citations:
|
||||
# if citations are enabled, only show cited documents
|
||||
if use_citations and answer.citations:
|
||||
citations_blocks = _build_citations_blocks(answer)
|
||||
else:
|
||||
document_blocks = _priority_ordered_documents_blocks(answer)
|
||||
@ -637,4 +572,5 @@ def build_slack_response_blocks(
|
||||
+ web_follow_up_block
|
||||
+ follow_up_block
|
||||
)
|
||||
|
||||
return all_blocks
|
||||
|
||||
@ -1,7 +1,6 @@
|
||||
import functools
|
||||
from collections.abc import Callable
|
||||
from typing import Any
|
||||
from typing import cast
|
||||
from typing import Optional
|
||||
from typing import TypeVar
|
||||
|
||||
@ -9,46 +8,36 @@ from retry import retry
|
||||
from slack_sdk import WebClient
|
||||
from slack_sdk.models.blocks import SectionBlock
|
||||
|
||||
from danswer.chat.chat_utils import prepare_chat_message_request
|
||||
from danswer.chat.models import ChatDanswerBotResponse
|
||||
from danswer.chat.process_message import gather_stream_for_slack
|
||||
from danswer.chat.process_message import stream_chat_message_objects
|
||||
from danswer.configs.app_configs import DISABLE_GENERATIVE_AI
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_ANSWER_GENERATION_TIMEOUT
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_DISABLE_COT
|
||||
from danswer.configs.constants import DEFAULT_PERSONA_ID
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_DISABLE_DOCS_ONLY_ANSWER
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_DISPLAY_ERROR_MSGS
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_NUM_RETRIES
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_TARGET_CHUNK_PERCENTAGE
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_USE_QUOTES
|
||||
from danswer.configs.danswerbot_configs import DANSWER_FOLLOWUP_EMOJI
|
||||
from danswer.configs.danswerbot_configs import DANSWER_REACT_EMOJI
|
||||
from danswer.configs.danswerbot_configs import ENABLE_DANSWERBOT_REFLEXION
|
||||
from danswer.configs.danswerbot_configs import MAX_THREAD_CONTEXT_PERCENTAGE
|
||||
from danswer.context.search.enums import OptionalSearchSetting
|
||||
from danswer.context.search.models import BaseFilters
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.danswerbot.slack.blocks import build_slack_response_blocks
|
||||
from danswer.danswerbot.slack.handlers.utils import send_team_member_message
|
||||
from danswer.danswerbot.slack.handlers.utils import slackify_message_thread
|
||||
from danswer.danswerbot.slack.models import SlackMessageInfo
|
||||
from danswer.danswerbot.slack.utils import respond_in_thread
|
||||
from danswer.danswerbot.slack.utils import SlackRateLimiter
|
||||
from danswer.danswerbot.slack.utils import update_emote_react
|
||||
from danswer.db.engine import get_session_with_tenant
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import SlackBotResponseType
|
||||
from danswer.db.models import SlackChannelConfig
|
||||
from danswer.db.persona import fetch_persona_by_id
|
||||
from danswer.db.search_settings import get_current_search_settings
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_persona_by_id
|
||||
from danswer.db.users import get_user_by_email
|
||||
from danswer.llm.answering.prompts.citations_prompt import (
|
||||
compute_max_document_tokens_for_persona,
|
||||
)
|
||||
from danswer.llm.factory import get_llms_for_persona
|
||||
from danswer.llm.utils import check_number_of_tokens
|
||||
from danswer.llm.utils import get_max_input_tokens
|
||||
from danswer.one_shot_answer.answer_question import get_search_answer
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.one_shot_answer.models import OneShotQAResponse
|
||||
from danswer.server.query_and_chat.models import CreateChatMessageRequest
|
||||
from danswer.utils.logger import DanswerLoggingAdapter
|
||||
|
||||
|
||||
srl = SlackRateLimiter()
|
||||
|
||||
RT = TypeVar("RT") # return type
|
||||
@ -83,16 +72,14 @@ def handle_regular_answer(
|
||||
feedback_reminder_id: str | None,
|
||||
tenant_id: str | None,
|
||||
num_retries: int = DANSWER_BOT_NUM_RETRIES,
|
||||
answer_generation_timeout: int = DANSWER_BOT_ANSWER_GENERATION_TIMEOUT,
|
||||
thread_context_percent: float = DANSWER_BOT_TARGET_CHUNK_PERCENTAGE,
|
||||
thread_context_percent: float = MAX_THREAD_CONTEXT_PERCENTAGE,
|
||||
should_respond_with_error_msgs: bool = DANSWER_BOT_DISPLAY_ERROR_MSGS,
|
||||
disable_docs_only_answer: bool = DANSWER_BOT_DISABLE_DOCS_ONLY_ANSWER,
|
||||
disable_cot: bool = DANSWER_BOT_DISABLE_COT,
|
||||
reflexion: bool = ENABLE_DANSWERBOT_REFLEXION,
|
||||
) -> bool:
|
||||
channel_conf = slack_channel_config.channel_config if slack_channel_config else None
|
||||
|
||||
messages = message_info.thread_messages
|
||||
|
||||
message_ts_to_respond_to = message_info.msg_to_respond
|
||||
is_bot_msg = message_info.is_bot_msg
|
||||
user = None
|
||||
@ -102,9 +89,18 @@ def handle_regular_answer(
|
||||
user = get_user_by_email(message_info.email, db_session)
|
||||
|
||||
document_set_names: list[str] | None = None
|
||||
persona = slack_channel_config.persona if slack_channel_config else None
|
||||
prompt = None
|
||||
if persona:
|
||||
# If no persona is specified, use the default search based persona
|
||||
# This way slack flow always has a persona
|
||||
persona = slack_channel_config.persona if slack_channel_config else None
|
||||
if not persona:
|
||||
with get_session_with_tenant(tenant_id) as db_session:
|
||||
persona = get_persona_by_id(DEFAULT_PERSONA_ID, user, db_session)
|
||||
document_set_names = [
|
||||
document_set.name for document_set in persona.document_sets
|
||||
]
|
||||
prompt = persona.prompts[0] if persona.prompts else None
|
||||
else:
|
||||
document_set_names = [
|
||||
document_set.name for document_set in persona.document_sets
|
||||
]
|
||||
@ -112,6 +108,26 @@ def handle_regular_answer(
|
||||
|
||||
should_respond_even_with_no_docs = persona.num_chunks == 0 if persona else False
|
||||
|
||||
# TODO: Add in support for Slack to truncate messages based on max LLM context
|
||||
# llm, _ = get_llms_for_persona(persona)
|
||||
|
||||
# llm_tokenizer = get_tokenizer(
|
||||
# model_name=llm.config.model_name,
|
||||
# provider_type=llm.config.model_provider,
|
||||
# )
|
||||
|
||||
# # In cases of threads, split the available tokens between docs and thread context
|
||||
# input_tokens = get_max_input_tokens(
|
||||
# model_name=llm.config.model_name,
|
||||
# model_provider=llm.config.model_provider,
|
||||
# )
|
||||
# max_history_tokens = int(input_tokens * thread_context_percent)
|
||||
# combined_message = combine_message_thread(
|
||||
# messages, max_tokens=max_history_tokens, llm_tokenizer=llm_tokenizer
|
||||
# )
|
||||
|
||||
combined_message = slackify_message_thread(messages)
|
||||
|
||||
bypass_acl = False
|
||||
if (
|
||||
slack_channel_config
|
||||
@ -122,13 +138,6 @@ def handle_regular_answer(
|
||||
# with non-public document sets
|
||||
bypass_acl = True
|
||||
|
||||
# figure out if we want to use citations or quotes
|
||||
use_citations = (
|
||||
not DANSWER_BOT_USE_QUOTES
|
||||
if slack_channel_config is None
|
||||
else slack_channel_config.response_type == SlackBotResponseType.CITATIONS
|
||||
)
|
||||
|
||||
if not message_ts_to_respond_to and not is_bot_msg:
|
||||
# if the message is not "/danswer" command, then it should have a message ts to respond to
|
||||
raise RuntimeError(
|
||||
@ -141,75 +150,23 @@ def handle_regular_answer(
|
||||
backoff=2,
|
||||
)
|
||||
@rate_limits(client=client, channel=channel, thread_ts=message_ts_to_respond_to)
|
||||
def _get_answer(new_message_request: DirectQARequest) -> OneShotQAResponse | None:
|
||||
max_document_tokens: int | None = None
|
||||
max_history_tokens: int | None = None
|
||||
|
||||
def _get_slack_answer(
|
||||
new_message_request: CreateChatMessageRequest, danswer_user: User | None
|
||||
) -> ChatDanswerBotResponse:
|
||||
with get_session_with_tenant(tenant_id) as db_session:
|
||||
if len(new_message_request.messages) > 1:
|
||||
if new_message_request.persona_config:
|
||||
raise RuntimeError("Slack bot does not support persona config")
|
||||
elif new_message_request.persona_id is not None:
|
||||
persona = cast(
|
||||
Persona,
|
||||
fetch_persona_by_id(
|
||||
db_session,
|
||||
new_message_request.persona_id,
|
||||
user=None,
|
||||
get_editable=False,
|
||||
),
|
||||
)
|
||||
else:
|
||||
raise RuntimeError(
|
||||
"No persona id provided, this should never happen."
|
||||
)
|
||||
|
||||
llm, _ = get_llms_for_persona(persona)
|
||||
|
||||
# In cases of threads, split the available tokens between docs and thread context
|
||||
input_tokens = get_max_input_tokens(
|
||||
model_name=llm.config.model_name,
|
||||
model_provider=llm.config.model_provider,
|
||||
)
|
||||
max_history_tokens = int(input_tokens * thread_context_percent)
|
||||
|
||||
remaining_tokens = input_tokens - max_history_tokens
|
||||
|
||||
query_text = new_message_request.messages[0].message
|
||||
if persona:
|
||||
max_document_tokens = compute_max_document_tokens_for_persona(
|
||||
persona=persona,
|
||||
actual_user_input=query_text,
|
||||
max_llm_token_override=remaining_tokens,
|
||||
)
|
||||
else:
|
||||
max_document_tokens = (
|
||||
remaining_tokens
|
||||
- 512 # Needs to be more than any of the QA prompts
|
||||
- check_number_of_tokens(query_text)
|
||||
)
|
||||
|
||||
if DISABLE_GENERATIVE_AI:
|
||||
return None
|
||||
|
||||
# This also handles creating the query event in postgres
|
||||
answer = get_search_answer(
|
||||
query_req=new_message_request,
|
||||
user=user,
|
||||
max_document_tokens=max_document_tokens,
|
||||
max_history_tokens=max_history_tokens,
|
||||
packets = stream_chat_message_objects(
|
||||
new_msg_req=new_message_request,
|
||||
user=danswer_user,
|
||||
db_session=db_session,
|
||||
answer_generation_timeout=answer_generation_timeout,
|
||||
enable_reflexion=reflexion,
|
||||
bypass_acl=bypass_acl,
|
||||
use_citations=use_citations,
|
||||
danswerbot_flow=True,
|
||||
)
|
||||
|
||||
if not answer.error_msg:
|
||||
return answer
|
||||
else:
|
||||
raise RuntimeError(answer.error_msg)
|
||||
answer = gather_stream_for_slack(packets)
|
||||
|
||||
if answer.error_msg:
|
||||
raise RuntimeError(answer.error_msg)
|
||||
|
||||
return answer
|
||||
|
||||
try:
|
||||
# By leaving time_cutoff and favor_recent as None, and setting enable_auto_detect_filters
|
||||
@ -239,26 +196,24 @@ def handle_regular_answer(
|
||||
enable_auto_detect_filters=auto_detect_filters,
|
||||
)
|
||||
|
||||
# Always apply reranking settings if it exists, this is the non-streaming flow
|
||||
with get_session_with_tenant(tenant_id) as db_session:
|
||||
saved_search_settings = get_current_search_settings(db_session)
|
||||
|
||||
# This includes throwing out answer via reflexion
|
||||
answer = _get_answer(
|
||||
DirectQARequest(
|
||||
messages=messages,
|
||||
multilingual_query_expansion=saved_search_settings.multilingual_expansion
|
||||
if saved_search_settings
|
||||
else None,
|
||||
prompt_id=prompt.id if prompt else None,
|
||||
persona_id=persona.id if persona is not None else 0,
|
||||
retrieval_options=retrieval_details,
|
||||
chain_of_thought=not disable_cot,
|
||||
rerank_settings=RerankingDetails.from_db_model(saved_search_settings)
|
||||
if saved_search_settings
|
||||
else None,
|
||||
answer_request = prepare_chat_message_request(
|
||||
message_text=combined_message,
|
||||
user=user,
|
||||
persona_id=persona.id,
|
||||
# This is not used in the Slack flow, only in the answer API
|
||||
persona_override_config=None,
|
||||
prompt=prompt,
|
||||
message_ts_to_respond_to=message_ts_to_respond_to,
|
||||
retrieval_details=retrieval_details,
|
||||
rerank_settings=None, # Rerank customization supported in Slack flow
|
||||
db_session=db_session,
|
||||
)
|
||||
|
||||
answer = _get_slack_answer(
|
||||
new_message_request=answer_request, danswer_user=user
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
logger.exception(
|
||||
f"Unable to process message - did not successfully answer "
|
||||
@ -359,7 +314,7 @@ def handle_regular_answer(
|
||||
top_docs = retrieval_info.top_documents
|
||||
if not top_docs and not should_respond_even_with_no_docs:
|
||||
logger.error(
|
||||
f"Unable to answer question: '{answer.rephrase}' - no documents found"
|
||||
f"Unable to answer question: '{combined_message}' - no documents found"
|
||||
)
|
||||
# Optionally, respond in thread with the error message
|
||||
# Used primarily for debugging purposes
|
||||
@ -380,18 +335,18 @@ def handle_regular_answer(
|
||||
)
|
||||
return True
|
||||
|
||||
only_respond_with_citations_or_quotes = (
|
||||
only_respond_if_citations = (
|
||||
channel_conf
|
||||
and "well_answered_postfilter" in channel_conf.get("answer_filters", [])
|
||||
)
|
||||
has_citations_or_quotes = bool(answer.citations or answer.quotes)
|
||||
|
||||
if (
|
||||
only_respond_with_citations_or_quotes
|
||||
and not has_citations_or_quotes
|
||||
only_respond_if_citations
|
||||
and not answer.citations
|
||||
and not message_info.bypass_filters
|
||||
):
|
||||
logger.error(
|
||||
f"Unable to find citations or quotes to answer: '{answer.rephrase}' - not answering!"
|
||||
f"Unable to find citations to answer: '{answer.answer}' - not answering!"
|
||||
)
|
||||
# Optionally, respond in thread with the error message
|
||||
# Used primarily for debugging purposes
|
||||
@ -409,9 +364,8 @@ def handle_regular_answer(
|
||||
tenant_id=tenant_id,
|
||||
message_info=message_info,
|
||||
answer=answer,
|
||||
persona=persona,
|
||||
channel_conf=channel_conf,
|
||||
use_citations=use_citations,
|
||||
use_citations=True, # No longer supporting quotes
|
||||
feedback_reminder_id=feedback_reminder_id,
|
||||
)
|
||||
|
||||
|
||||
@ -1,8 +1,33 @@
|
||||
from slack_sdk import WebClient
|
||||
|
||||
from danswer.chat.models import ThreadMessage
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.danswerbot.slack.utils import respond_in_thread
|
||||
|
||||
|
||||
def slackify_message_thread(messages: list[ThreadMessage]) -> str:
|
||||
# Note: this does not handle extremely long threads, every message will be included
|
||||
# with weaker LLMs, this could cause issues with exceeeding the token limit
|
||||
if not messages:
|
||||
return ""
|
||||
|
||||
message_strs: list[str] = []
|
||||
for message in messages:
|
||||
if message.role == MessageType.USER:
|
||||
message_text = (
|
||||
f"{message.sender or 'Unknown User'} said in Slack:\n{message.message}"
|
||||
)
|
||||
elif message.role == MessageType.ASSISTANT:
|
||||
message_text = f"AI said in Slack:\n{message.message}"
|
||||
else:
|
||||
message_text = (
|
||||
f"{message.role.value.upper()} said in Slack:\n{message.message}"
|
||||
)
|
||||
message_strs.append(message_text)
|
||||
|
||||
return "\n\n".join(message_strs)
|
||||
|
||||
|
||||
def send_team_member_message(
|
||||
client: WebClient,
|
||||
channel: str,
|
||||
|
||||
@ -19,6 +19,8 @@ from slack_sdk.socket_mode.request import SocketModeRequest
|
||||
from slack_sdk.socket_mode.response import SocketModeResponse
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.chat.models import ThreadMessage
|
||||
from danswer.configs.app_configs import DEV_MODE
|
||||
from danswer.configs.app_configs import POD_NAME
|
||||
from danswer.configs.app_configs import POD_NAMESPACE
|
||||
from danswer.configs.constants import DanswerRedisLocks
|
||||
@ -74,7 +76,6 @@ from danswer.db.slack_bot import fetch_slack_bots
|
||||
from danswer.key_value_store.interface import KvKeyNotFoundError
|
||||
from danswer.natural_language_processing.search_nlp_models import EmbeddingModel
|
||||
from danswer.natural_language_processing.search_nlp_models import warm_up_bi_encoder
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.redis.redis_pool import get_redis_client
|
||||
from danswer.server.manage.models import SlackBotTokens
|
||||
from danswer.utils.logger import setup_logger
|
||||
@ -250,7 +251,7 @@ class SlackbotHandler:
|
||||
nx=True,
|
||||
ex=TENANT_LOCK_EXPIRATION,
|
||||
)
|
||||
if not acquired:
|
||||
if not acquired and not DEV_MODE:
|
||||
logger.debug(f"Another pod holds the lock for tenant {tenant_id}")
|
||||
continue
|
||||
|
||||
|
||||
@ -1,6 +1,6 @@
|
||||
from pydantic import BaseModel
|
||||
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.chat.models import ThreadMessage
|
||||
|
||||
|
||||
class SlackMessageInfo(BaseModel):
|
||||
|
||||
@ -30,13 +30,13 @@ from danswer.configs.danswerbot_configs import (
|
||||
from danswer.connectors.slack.utils import make_slack_api_rate_limited
|
||||
from danswer.connectors.slack.utils import SlackTextCleaner
|
||||
from danswer.danswerbot.slack.constants import FeedbackVisibility
|
||||
from danswer.danswerbot.slack.models import ThreadMessage
|
||||
from danswer.db.engine import get_session_with_tenant
|
||||
from danswer.db.users import get_user_by_email
|
||||
from danswer.llm.exceptions import GenAIDisabledException
|
||||
from danswer.llm.factory import get_default_llms
|
||||
from danswer.llm.utils import dict_based_prompt_to_langchain_prompt
|
||||
from danswer.llm.utils import message_to_string
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.prompts.miscellaneous_prompts import SLACK_LANGUAGE_REPHRASE_PROMPT
|
||||
from danswer.utils.logger import setup_logger
|
||||
from danswer.utils.telemetry import optional_telemetry
|
||||
|
||||
@ -145,16 +145,10 @@ def get_chat_sessions_by_user(
|
||||
user_id: UUID | None,
|
||||
deleted: bool | None,
|
||||
db_session: Session,
|
||||
only_one_shot: bool = False,
|
||||
limit: int = 50,
|
||||
) -> list[ChatSession]:
|
||||
stmt = select(ChatSession).where(ChatSession.user_id == user_id)
|
||||
|
||||
if only_one_shot:
|
||||
stmt = stmt.where(ChatSession.one_shot.is_(True))
|
||||
else:
|
||||
stmt = stmt.where(ChatSession.one_shot.is_(False))
|
||||
|
||||
stmt = stmt.order_by(desc(ChatSession.time_created))
|
||||
|
||||
if deleted is not None:
|
||||
@ -226,12 +220,11 @@ def delete_messages_and_files_from_chat_session(
|
||||
|
||||
def create_chat_session(
|
||||
db_session: Session,
|
||||
description: str,
|
||||
description: str | None,
|
||||
user_id: UUID | None,
|
||||
persona_id: int | None, # Can be none if temporary persona is used
|
||||
llm_override: LLMOverride | None = None,
|
||||
prompt_override: PromptOverride | None = None,
|
||||
one_shot: bool = False,
|
||||
danswerbot_flow: bool = False,
|
||||
slack_thread_id: str | None = None,
|
||||
) -> ChatSession:
|
||||
@ -241,7 +234,6 @@ def create_chat_session(
|
||||
description=description,
|
||||
llm_override=llm_override,
|
||||
prompt_override=prompt_override,
|
||||
one_shot=one_shot,
|
||||
danswerbot_flow=danswerbot_flow,
|
||||
slack_thread_id=slack_thread_id,
|
||||
)
|
||||
@ -287,8 +279,6 @@ def duplicate_chat_session_for_user_from_slack(
|
||||
description="",
|
||||
llm_override=chat_session.llm_override,
|
||||
prompt_override=chat_session.prompt_override,
|
||||
# Chat sessions from Slack should put people in the chat UI, not the search
|
||||
one_shot=False,
|
||||
# Chat is in UI now so this is false
|
||||
danswerbot_flow=False,
|
||||
# Maybe we want this in the future to track if it was created from Slack
|
||||
|
||||
@ -1,6 +1,5 @@
|
||||
import datetime
|
||||
import json
|
||||
from enum import Enum as PyEnum
|
||||
from typing import Any
|
||||
from typing import Literal
|
||||
from typing import NotRequired
|
||||
@ -964,9 +963,8 @@ class ChatSession(Base):
|
||||
persona_id: Mapped[int | None] = mapped_column(
|
||||
ForeignKey("persona.id"), nullable=True
|
||||
)
|
||||
description: Mapped[str] = mapped_column(Text)
|
||||
# One-shot direct answering, currently the two types of chats are not mixed
|
||||
one_shot: Mapped[bool] = mapped_column(Boolean, default=False)
|
||||
description: Mapped[str | None] = mapped_column(Text, nullable=True)
|
||||
# This chat created by DanswerBot
|
||||
danswerbot_flow: Mapped[bool] = mapped_column(Boolean, default=False)
|
||||
# Only ever set to True if system is set to not hard-delete chats
|
||||
deleted: Mapped[bool] = mapped_column(Boolean, default=False)
|
||||
@ -1488,11 +1486,6 @@ class ChannelConfig(TypedDict):
|
||||
show_continue_in_web_ui: NotRequired[bool] # defaults to False
|
||||
|
||||
|
||||
class SlackBotResponseType(str, PyEnum):
|
||||
QUOTES = "quotes"
|
||||
CITATIONS = "citations"
|
||||
|
||||
|
||||
class SlackChannelConfig(Base):
|
||||
__tablename__ = "slack_channel_config"
|
||||
|
||||
@ -1505,9 +1498,6 @@ class SlackChannelConfig(Base):
|
||||
channel_config: Mapped[ChannelConfig] = mapped_column(
|
||||
postgresql.JSONB(), nullable=False
|
||||
)
|
||||
response_type: Mapped[SlackBotResponseType] = mapped_column(
|
||||
Enum(SlackBotResponseType, native_enum=False), nullable=False
|
||||
)
|
||||
|
||||
enable_auto_filters: Mapped[bool] = mapped_column(
|
||||
Boolean, nullable=False, default=False
|
||||
|
||||
@ -10,7 +10,6 @@ from danswer.db.constants import SLACK_BOT_PERSONA_PREFIX
|
||||
from danswer.db.models import ChannelConfig
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import Persona__DocumentSet
|
||||
from danswer.db.models import SlackBotResponseType
|
||||
from danswer.db.models import SlackChannelConfig
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_default_prompt
|
||||
@ -83,7 +82,6 @@ def insert_slack_channel_config(
|
||||
slack_bot_id: int,
|
||||
persona_id: int | None,
|
||||
channel_config: ChannelConfig,
|
||||
response_type: SlackBotResponseType,
|
||||
standard_answer_category_ids: list[int],
|
||||
enable_auto_filters: bool,
|
||||
) -> SlackChannelConfig:
|
||||
@ -115,7 +113,6 @@ def insert_slack_channel_config(
|
||||
slack_bot_id=slack_bot_id,
|
||||
persona_id=persona_id,
|
||||
channel_config=channel_config,
|
||||
response_type=response_type,
|
||||
standard_answer_categories=existing_standard_answer_categories,
|
||||
enable_auto_filters=enable_auto_filters,
|
||||
)
|
||||
@ -130,7 +127,6 @@ def update_slack_channel_config(
|
||||
slack_channel_config_id: int,
|
||||
persona_id: int | None,
|
||||
channel_config: ChannelConfig,
|
||||
response_type: SlackBotResponseType,
|
||||
standard_answer_category_ids: list[int],
|
||||
enable_auto_filters: bool,
|
||||
) -> SlackChannelConfig:
|
||||
@ -170,7 +166,6 @@ def update_slack_channel_config(
|
||||
# will encounter `violates foreign key constraint` errors
|
||||
slack_channel_config.persona_id = persona_id
|
||||
slack_channel_config.channel_config = channel_config
|
||||
slack_channel_config.response_type = response_type
|
||||
slack_channel_config.standard_answer_categories = list(
|
||||
existing_standard_answer_categories
|
||||
)
|
||||
|
||||
@ -18,18 +18,12 @@ from danswer.llm.answering.models import PromptConfig
|
||||
from danswer.llm.answering.prompts.build import AnswerPromptBuilder
|
||||
from danswer.llm.answering.prompts.build import default_build_system_message
|
||||
from danswer.llm.answering.prompts.build import default_build_user_message
|
||||
from danswer.llm.answering.stream_processing.answer_response_handler import (
|
||||
AnswerResponseHandler,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.answer_response_handler import (
|
||||
CitationResponseHandler,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.answer_response_handler import (
|
||||
DummyAnswerResponseHandler,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.answer_response_handler import (
|
||||
QuotesResponseHandler,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.utils import map_document_id_order
|
||||
from danswer.llm.answering.tool.tool_response_handler import ToolResponseHandler
|
||||
from danswer.llm.interfaces import LLM
|
||||
@ -214,18 +208,23 @@ class Answer:
|
||||
|
||||
search_result = SearchTool.get_search_result(current_llm_call) or []
|
||||
|
||||
answer_handler: AnswerResponseHandler
|
||||
if self.answer_style_config.citation_config:
|
||||
answer_handler = CitationResponseHandler(
|
||||
context_docs=search_result,
|
||||
doc_id_to_rank_map=map_document_id_order(search_result),
|
||||
)
|
||||
elif self.answer_style_config.quotes_config:
|
||||
answer_handler = QuotesResponseHandler(
|
||||
context_docs=search_result,
|
||||
)
|
||||
else:
|
||||
raise ValueError("No answer style config provided")
|
||||
# Quotes are no longer supported
|
||||
# answer_handler: AnswerResponseHandler
|
||||
# if self.answer_style_config.citation_config:
|
||||
# answer_handler = CitationResponseHandler(
|
||||
# context_docs=search_result,
|
||||
# doc_id_to_rank_map=map_document_id_order(search_result),
|
||||
# )
|
||||
# elif self.answer_style_config.quotes_config:
|
||||
# answer_handler = QuotesResponseHandler(
|
||||
# context_docs=search_result,
|
||||
# )
|
||||
# else:
|
||||
# raise ValueError("No answer style config provided")
|
||||
answer_handler = CitationResponseHandler(
|
||||
context_docs=search_result,
|
||||
doc_id_to_rank_map=map_document_id_order(search_result),
|
||||
)
|
||||
|
||||
response_handler_manager = LLMResponseHandlerManager(
|
||||
tool_call_handler, answer_handler, self.is_cancelled
|
||||
|
||||
@ -8,7 +8,6 @@ from pydantic.v1 import BaseModel as BaseModel__v1
|
||||
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import StreamStopInfo
|
||||
from danswer.chat.models import StreamStopReason
|
||||
from danswer.file_store.models import InMemoryChatFile
|
||||
@ -30,7 +29,6 @@ if TYPE_CHECKING:
|
||||
ResponsePart = (
|
||||
DanswerAnswerPiece
|
||||
| CitationInfo
|
||||
| DanswerQuotes
|
||||
| ToolCallKickoff
|
||||
| ToolResponse
|
||||
| ToolCallFinalResult
|
||||
|
||||
@ -9,9 +9,6 @@ from danswer.llm.answering.llm_response_handler import ResponsePart
|
||||
from danswer.llm.answering.stream_processing.citation_processing import (
|
||||
CitationProcessor,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.quotes_processing import (
|
||||
QuotesProcessor,
|
||||
)
|
||||
from danswer.llm.answering.stream_processing.utils import DocumentIdOrderMapping
|
||||
from danswer.utils.logger import setup_logger
|
||||
|
||||
@ -70,28 +67,29 @@ class CitationResponseHandler(AnswerResponseHandler):
|
||||
yield from self.citation_processor.process_token(content)
|
||||
|
||||
|
||||
class QuotesResponseHandler(AnswerResponseHandler):
|
||||
def __init__(
|
||||
self,
|
||||
context_docs: list[LlmDoc],
|
||||
is_json_prompt: bool = True,
|
||||
):
|
||||
self.quotes_processor = QuotesProcessor(
|
||||
context_docs=context_docs,
|
||||
is_json_prompt=is_json_prompt,
|
||||
)
|
||||
# No longer in use, remove later
|
||||
# class QuotesResponseHandler(AnswerResponseHandler):
|
||||
# def __init__(
|
||||
# self,
|
||||
# context_docs: list[LlmDoc],
|
||||
# is_json_prompt: bool = True,
|
||||
# ):
|
||||
# self.quotes_processor = QuotesProcessor(
|
||||
# context_docs=context_docs,
|
||||
# is_json_prompt=is_json_prompt,
|
||||
# )
|
||||
|
||||
def handle_response_part(
|
||||
self,
|
||||
response_item: BaseMessage | None,
|
||||
previous_response_items: list[BaseMessage],
|
||||
) -> Generator[ResponsePart, None, None]:
|
||||
if response_item is None:
|
||||
yield from self.quotes_processor.process_token(None)
|
||||
return
|
||||
# def handle_response_part(
|
||||
# self,
|
||||
# response_item: BaseMessage | None,
|
||||
# previous_response_items: list[BaseMessage],
|
||||
# ) -> Generator[ResponsePart, None, None]:
|
||||
# if response_item is None:
|
||||
# yield from self.quotes_processor.process_token(None)
|
||||
# return
|
||||
|
||||
content = (
|
||||
response_item.content if isinstance(response_item.content, str) else ""
|
||||
)
|
||||
# content = (
|
||||
# response_item.content if isinstance(response_item.content, str) else ""
|
||||
# )
|
||||
|
||||
yield from self.quotes_processor.process_token(content)
|
||||
# yield from self.quotes_processor.process_token(content)
|
||||
|
||||
@ -1,3 +1,4 @@
|
||||
# THIS IS NO LONGER IN USE
|
||||
import math
|
||||
import re
|
||||
from collections.abc import Generator
|
||||
@ -5,11 +6,10 @@ from json import JSONDecodeError
|
||||
from typing import Optional
|
||||
|
||||
import regex
|
||||
from pydantic import BaseModel
|
||||
|
||||
from danswer.chat.models import DanswerAnswer
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerQuote
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import LlmDoc
|
||||
from danswer.configs.chat_configs import QUOTE_ALLOWED_ERROR_PERCENT
|
||||
from danswer.context.search.models import InferenceChunk
|
||||
@ -26,6 +26,20 @@ logger = setup_logger()
|
||||
answer_pattern = re.compile(r'{\s*"answer"\s*:\s*"', re.IGNORECASE)
|
||||
|
||||
|
||||
class DanswerQuote(BaseModel):
|
||||
# This is during inference so everything is a string by this point
|
||||
quote: str
|
||||
document_id: str
|
||||
link: str | None
|
||||
source_type: str
|
||||
semantic_identifier: str
|
||||
blurb: str
|
||||
|
||||
|
||||
class DanswerQuotes(BaseModel):
|
||||
quotes: list[DanswerQuote]
|
||||
|
||||
|
||||
def _extract_answer_quotes_freeform(
|
||||
answer_raw: str,
|
||||
) -> tuple[Optional[str], Optional[list[str]]]:
|
||||
|
||||
@ -1,5 +1,6 @@
|
||||
from typing import Any
|
||||
|
||||
from danswer.chat.models import PersonaOverrideConfig
|
||||
from danswer.configs.app_configs import DISABLE_GENERATIVE_AI
|
||||
from danswer.configs.chat_configs import QA_TIMEOUT
|
||||
from danswer.configs.model_configs import GEN_AI_MODEL_FALLBACK_MAX_TOKENS
|
||||
@ -13,8 +14,11 @@ from danswer.llm.exceptions import GenAIDisabledException
|
||||
from danswer.llm.interfaces import LLM
|
||||
from danswer.llm.override_models import LLMOverride
|
||||
from danswer.utils.headers import build_llm_extra_headers
|
||||
from danswer.utils.logger import setup_logger
|
||||
from danswer.utils.long_term_log import LongTermLogger
|
||||
|
||||
logger = setup_logger()
|
||||
|
||||
|
||||
def _build_extra_model_kwargs(provider: str) -> dict[str, Any]:
|
||||
"""Ollama requires us to specify the max context window.
|
||||
@ -32,11 +36,15 @@ def get_main_llm_from_tuple(
|
||||
|
||||
|
||||
def get_llms_for_persona(
|
||||
persona: Persona,
|
||||
persona: Persona | PersonaOverrideConfig | None,
|
||||
llm_override: LLMOverride | None = None,
|
||||
additional_headers: dict[str, str] | None = None,
|
||||
long_term_logger: LongTermLogger | None = None,
|
||||
) -> tuple[LLM, LLM]:
|
||||
if persona is None:
|
||||
logger.warning("No persona provided, using default LLMs")
|
||||
return get_default_llms()
|
||||
|
||||
model_provider_override = llm_override.model_provider if llm_override else None
|
||||
model_version_override = llm_override.model_version if llm_override else None
|
||||
temperature_override = llm_override.temperature if llm_override else None
|
||||
|
||||
@ -1,456 +0,0 @@
|
||||
from collections.abc import Callable
|
||||
from collections.abc import Iterator
|
||||
from typing import cast
|
||||
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.chat.chat_utils import reorganize_citations
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerContexts
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import DocumentRelevance
|
||||
from danswer.chat.models import LLMRelevanceFilterResponse
|
||||
from danswer.chat.models import QADocsResponse
|
||||
from danswer.chat.models import RelevanceAnalysis
|
||||
from danswer.chat.models import StreamingError
|
||||
from danswer.configs.chat_configs import DISABLE_LLM_DOC_RELEVANCE
|
||||
from danswer.configs.chat_configs import MAX_CHUNKS_FED_TO_CHAT
|
||||
from danswer.configs.chat_configs import QA_TIMEOUT
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.context.search.enums import LLMEvaluationType
|
||||
from danswer.context.search.models import RerankMetricsContainer
|
||||
from danswer.context.search.models import RetrievalMetricsContainer
|
||||
from danswer.context.search.utils import chunks_or_sections_to_search_docs
|
||||
from danswer.context.search.utils import dedupe_documents
|
||||
from danswer.db.chat import create_chat_session
|
||||
from danswer.db.chat import create_db_search_doc
|
||||
from danswer.db.chat import create_new_chat_message
|
||||
from danswer.db.chat import get_or_create_root_message
|
||||
from danswer.db.chat import translate_db_message_to_chat_message_detail
|
||||
from danswer.db.chat import translate_db_search_doc_to_server_search_doc
|
||||
from danswer.db.chat import update_search_docs_table_with_relevance
|
||||
from danswer.db.engine import get_session_context_manager
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_prompt_by_id
|
||||
from danswer.llm.answering.answer import Answer
|
||||
from danswer.llm.answering.models import AnswerStyleConfig
|
||||
from danswer.llm.answering.models import CitationConfig
|
||||
from danswer.llm.answering.models import DocumentPruningConfig
|
||||
from danswer.llm.answering.models import PromptConfig
|
||||
from danswer.llm.answering.models import QuotesConfig
|
||||
from danswer.llm.factory import get_llms_for_persona
|
||||
from danswer.llm.factory import get_main_llm_from_tuple
|
||||
from danswer.natural_language_processing.utils import get_tokenizer
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.one_shot_answer.models import OneShotQAResponse
|
||||
from danswer.one_shot_answer.models import QueryRephrase
|
||||
from danswer.one_shot_answer.qa_utils import combine_message_thread
|
||||
from danswer.one_shot_answer.qa_utils import slackify_message_thread
|
||||
from danswer.secondary_llm_flows.answer_validation import get_answer_validity
|
||||
from danswer.secondary_llm_flows.query_expansion import thread_based_query_rephrase
|
||||
from danswer.server.query_and_chat.models import ChatMessageDetail
|
||||
from danswer.server.utils import get_json_line
|
||||
from danswer.tools.force import ForceUseTool
|
||||
from danswer.tools.models import ToolResponse
|
||||
from danswer.tools.tool_implementations.search.search_tool import SEARCH_DOC_CONTENT_ID
|
||||
from danswer.tools.tool_implementations.search.search_tool import (
|
||||
SEARCH_RESPONSE_SUMMARY_ID,
|
||||
)
|
||||
from danswer.tools.tool_implementations.search.search_tool import SearchResponseSummary
|
||||
from danswer.tools.tool_implementations.search.search_tool import SearchTool
|
||||
from danswer.tools.tool_implementations.search.search_tool import (
|
||||
SECTION_RELEVANCE_LIST_ID,
|
||||
)
|
||||
from danswer.tools.tool_runner import ToolCallKickoff
|
||||
from danswer.utils.logger import setup_logger
|
||||
from danswer.utils.long_term_log import LongTermLogger
|
||||
from danswer.utils.timing import log_generator_function_time
|
||||
from danswer.utils.variable_functionality import fetch_ee_implementation_or_noop
|
||||
|
||||
logger = setup_logger()
|
||||
|
||||
AnswerObjectIterator = Iterator[
|
||||
QueryRephrase
|
||||
| QADocsResponse
|
||||
| LLMRelevanceFilterResponse
|
||||
| DanswerAnswerPiece
|
||||
| DanswerQuotes
|
||||
| DanswerContexts
|
||||
| StreamingError
|
||||
| ChatMessageDetail
|
||||
| CitationInfo
|
||||
| ToolCallKickoff
|
||||
| DocumentRelevance
|
||||
]
|
||||
|
||||
|
||||
def stream_answer_objects(
|
||||
query_req: DirectQARequest,
|
||||
user: User | None,
|
||||
# These need to be passed in because in Web UI one shot flow,
|
||||
# we can have much more document as there is no history.
|
||||
# For Slack flow, we need to save more tokens for the thread context
|
||||
max_document_tokens: int | None,
|
||||
max_history_tokens: int | None,
|
||||
db_session: Session,
|
||||
# Needed to translate persona num_chunks to tokens to the LLM
|
||||
default_num_chunks: float = MAX_CHUNKS_FED_TO_CHAT,
|
||||
timeout: int = QA_TIMEOUT,
|
||||
bypass_acl: bool = False,
|
||||
use_citations: bool = False,
|
||||
danswerbot_flow: bool = False,
|
||||
retrieval_metrics_callback: (
|
||||
Callable[[RetrievalMetricsContainer], None] | None
|
||||
) = None,
|
||||
rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
|
||||
) -> AnswerObjectIterator:
|
||||
"""Streams in order:
|
||||
1. [always] Retrieved documents, stops flow if nothing is found
|
||||
2. [conditional] LLM selected chunk indices if LLM chunk filtering is turned on
|
||||
3. [always] A set of streamed DanswerAnswerPiece and DanswerQuotes at the end
|
||||
or an error anywhere along the line if something fails
|
||||
4. [always] Details on the final AI response message that is created
|
||||
"""
|
||||
user_id = user.id if user is not None else None
|
||||
query_msg = query_req.messages[-1]
|
||||
history = query_req.messages[:-1]
|
||||
|
||||
chat_session = create_chat_session(
|
||||
db_session=db_session,
|
||||
description="", # One shot queries don't need naming as it's never displayed
|
||||
user_id=user_id,
|
||||
persona_id=query_req.persona_id,
|
||||
one_shot=True,
|
||||
danswerbot_flow=danswerbot_flow,
|
||||
)
|
||||
|
||||
# permanent "log" store, used primarily for debugging
|
||||
long_term_logger = LongTermLogger(
|
||||
metadata={"user_id": str(user_id), "chat_session_id": str(chat_session.id)}
|
||||
)
|
||||
|
||||
temporary_persona: Persona | None = None
|
||||
|
||||
if query_req.persona_config is not None:
|
||||
temporary_persona = fetch_ee_implementation_or_noop(
|
||||
"danswer.server.query_and_chat.utils", "create_temporary_persona", None
|
||||
)(db_session=db_session, persona_config=query_req.persona_config, user=user)
|
||||
|
||||
persona = temporary_persona if temporary_persona else chat_session.persona
|
||||
|
||||
try:
|
||||
llm, fast_llm = get_llms_for_persona(
|
||||
persona=persona, long_term_logger=long_term_logger
|
||||
)
|
||||
except ValueError as e:
|
||||
logger.error(
|
||||
f"Failed to initialize LLMs for persona '{persona.name}': {str(e)}"
|
||||
)
|
||||
if "No LLM provider" in str(e):
|
||||
raise ValueError(
|
||||
"Please configure a Generative AI model to use this feature."
|
||||
) from e
|
||||
raise ValueError(
|
||||
"Failed to initialize the AI model. Please check your configuration and try again."
|
||||
) from e
|
||||
|
||||
llm_tokenizer = get_tokenizer(
|
||||
model_name=llm.config.model_name,
|
||||
provider_type=llm.config.model_provider,
|
||||
)
|
||||
|
||||
# Create a chat session which will just store the root message, the query, and the AI response
|
||||
root_message = get_or_create_root_message(
|
||||
chat_session_id=chat_session.id, db_session=db_session
|
||||
)
|
||||
|
||||
history_str = combine_message_thread(
|
||||
messages=history,
|
||||
max_tokens=max_history_tokens,
|
||||
llm_tokenizer=llm_tokenizer,
|
||||
)
|
||||
|
||||
rephrased_query = query_req.query_override or thread_based_query_rephrase(
|
||||
user_query=query_msg.message,
|
||||
history_str=history_str,
|
||||
)
|
||||
|
||||
# Given back ahead of the documents for latency reasons
|
||||
# In chat flow it's given back along with the documents
|
||||
yield QueryRephrase(rephrased_query=rephrased_query)
|
||||
|
||||
prompt = None
|
||||
if query_req.prompt_id is not None:
|
||||
# NOTE: let the user access any prompt as long as the Persona is shared
|
||||
# with them
|
||||
prompt = get_prompt_by_id(
|
||||
prompt_id=query_req.prompt_id, user=None, db_session=db_session
|
||||
)
|
||||
if prompt is None:
|
||||
if not persona.prompts:
|
||||
raise RuntimeError(
|
||||
"Persona does not have any prompts - this should never happen"
|
||||
)
|
||||
prompt = persona.prompts[0]
|
||||
|
||||
user_message_str = query_msg.message
|
||||
# For this endpoint, we only save one user message to the chat session
|
||||
# However, for slackbot, we want to include the history of the entire thread
|
||||
if danswerbot_flow:
|
||||
# Right now, we only support bringing over citations and search docs
|
||||
# from the last message in the thread, not the entire thread
|
||||
# in the future, we may want to retrieve the entire thread
|
||||
user_message_str = slackify_message_thread(query_req.messages)
|
||||
|
||||
# Create the first User query message
|
||||
new_user_message = create_new_chat_message(
|
||||
chat_session_id=chat_session.id,
|
||||
parent_message=root_message,
|
||||
prompt_id=query_req.prompt_id,
|
||||
message=user_message_str,
|
||||
token_count=len(llm_tokenizer.encode(user_message_str)),
|
||||
message_type=MessageType.USER,
|
||||
db_session=db_session,
|
||||
commit=True,
|
||||
)
|
||||
|
||||
prompt_config = PromptConfig.from_model(prompt)
|
||||
document_pruning_config = DocumentPruningConfig(
|
||||
max_chunks=int(
|
||||
persona.num_chunks if persona.num_chunks is not None else default_num_chunks
|
||||
),
|
||||
max_tokens=max_document_tokens,
|
||||
)
|
||||
|
||||
answer_config = AnswerStyleConfig(
|
||||
citation_config=CitationConfig() if use_citations else None,
|
||||
quotes_config=QuotesConfig() if not use_citations else None,
|
||||
document_pruning_config=document_pruning_config,
|
||||
)
|
||||
|
||||
search_tool = SearchTool(
|
||||
db_session=db_session,
|
||||
user=user,
|
||||
evaluation_type=(
|
||||
LLMEvaluationType.SKIP
|
||||
if DISABLE_LLM_DOC_RELEVANCE
|
||||
else query_req.evaluation_type
|
||||
),
|
||||
persona=persona,
|
||||
retrieval_options=query_req.retrieval_options,
|
||||
prompt_config=prompt_config,
|
||||
llm=llm,
|
||||
fast_llm=fast_llm,
|
||||
pruning_config=document_pruning_config,
|
||||
answer_style_config=answer_config,
|
||||
bypass_acl=bypass_acl,
|
||||
chunks_above=query_req.chunks_above,
|
||||
chunks_below=query_req.chunks_below,
|
||||
full_doc=query_req.full_doc,
|
||||
)
|
||||
|
||||
answer = Answer(
|
||||
question=query_msg.message,
|
||||
answer_style_config=answer_config,
|
||||
prompt_config=PromptConfig.from_model(prompt),
|
||||
llm=get_main_llm_from_tuple(
|
||||
get_llms_for_persona(persona=persona, long_term_logger=long_term_logger)
|
||||
),
|
||||
single_message_history=history_str,
|
||||
tools=[search_tool] if search_tool else [],
|
||||
force_use_tool=(
|
||||
ForceUseTool(
|
||||
tool_name=search_tool.name,
|
||||
args={"query": rephrased_query},
|
||||
force_use=True,
|
||||
)
|
||||
),
|
||||
# for now, don't use tool calling for this flow, as we haven't
|
||||
# tested quotes with tool calling too much yet
|
||||
skip_explicit_tool_calling=True,
|
||||
return_contexts=query_req.return_contexts,
|
||||
skip_gen_ai_answer_generation=query_req.skip_gen_ai_answer_generation,
|
||||
)
|
||||
# won't be any FileChatDisplay responses since that tool is never passed in
|
||||
for packet in cast(AnswerObjectIterator, answer.processed_streamed_output):
|
||||
# for one-shot flow, don't currently do anything with these
|
||||
if isinstance(packet, ToolResponse):
|
||||
# (likely fine that it comes after the initial creation of the search docs)
|
||||
if packet.id == SEARCH_RESPONSE_SUMMARY_ID:
|
||||
search_response_summary = cast(SearchResponseSummary, packet.response)
|
||||
|
||||
top_docs = chunks_or_sections_to_search_docs(
|
||||
search_response_summary.top_sections
|
||||
)
|
||||
|
||||
# Deduping happens at the last step to avoid harming quality by dropping content early on
|
||||
deduped_docs = top_docs
|
||||
if query_req.retrieval_options.dedupe_docs:
|
||||
deduped_docs, dropped_inds = dedupe_documents(top_docs)
|
||||
|
||||
reference_db_search_docs = [
|
||||
create_db_search_doc(server_search_doc=doc, db_session=db_session)
|
||||
for doc in deduped_docs
|
||||
]
|
||||
|
||||
response_docs = [
|
||||
translate_db_search_doc_to_server_search_doc(db_search_doc)
|
||||
for db_search_doc in reference_db_search_docs
|
||||
]
|
||||
|
||||
initial_response = QADocsResponse(
|
||||
rephrased_query=rephrased_query,
|
||||
top_documents=response_docs,
|
||||
predicted_flow=search_response_summary.predicted_flow,
|
||||
predicted_search=search_response_summary.predicted_search,
|
||||
applied_source_filters=search_response_summary.final_filters.source_type,
|
||||
applied_time_cutoff=search_response_summary.final_filters.time_cutoff,
|
||||
recency_bias_multiplier=search_response_summary.recency_bias_multiplier,
|
||||
)
|
||||
|
||||
yield initial_response
|
||||
|
||||
elif packet.id == SEARCH_DOC_CONTENT_ID:
|
||||
yield packet.response
|
||||
|
||||
elif packet.id == SECTION_RELEVANCE_LIST_ID:
|
||||
document_based_response = {}
|
||||
|
||||
if packet.response is not None:
|
||||
for evaluation in packet.response:
|
||||
document_based_response[
|
||||
evaluation.document_id
|
||||
] = RelevanceAnalysis(
|
||||
relevant=evaluation.relevant, content=evaluation.content
|
||||
)
|
||||
|
||||
evaluation_response = DocumentRelevance(
|
||||
relevance_summaries=document_based_response
|
||||
)
|
||||
if reference_db_search_docs is not None:
|
||||
update_search_docs_table_with_relevance(
|
||||
db_session=db_session,
|
||||
reference_db_search_docs=reference_db_search_docs,
|
||||
relevance_summary=evaluation_response,
|
||||
)
|
||||
yield evaluation_response
|
||||
|
||||
else:
|
||||
yield packet
|
||||
|
||||
# Saving Gen AI answer and responding with message info
|
||||
gen_ai_response_message = create_new_chat_message(
|
||||
chat_session_id=chat_session.id,
|
||||
parent_message=new_user_message,
|
||||
prompt_id=query_req.prompt_id,
|
||||
message=answer.llm_answer,
|
||||
token_count=len(llm_tokenizer.encode(answer.llm_answer)),
|
||||
message_type=MessageType.ASSISTANT,
|
||||
error=None,
|
||||
reference_docs=reference_db_search_docs,
|
||||
db_session=db_session,
|
||||
commit=True,
|
||||
)
|
||||
|
||||
msg_detail_response = translate_db_message_to_chat_message_detail(
|
||||
gen_ai_response_message
|
||||
)
|
||||
yield msg_detail_response
|
||||
|
||||
|
||||
@log_generator_function_time()
|
||||
def stream_search_answer(
|
||||
query_req: DirectQARequest,
|
||||
user: User | None,
|
||||
max_document_tokens: int | None,
|
||||
max_history_tokens: int | None,
|
||||
) -> Iterator[str]:
|
||||
with get_session_context_manager() as session:
|
||||
objects = stream_answer_objects(
|
||||
query_req=query_req,
|
||||
user=user,
|
||||
max_document_tokens=max_document_tokens,
|
||||
max_history_tokens=max_history_tokens,
|
||||
db_session=session,
|
||||
)
|
||||
for obj in objects:
|
||||
yield get_json_line(obj.model_dump())
|
||||
|
||||
|
||||
def get_search_answer(
|
||||
query_req: DirectQARequest,
|
||||
user: User | None,
|
||||
max_document_tokens: int | None,
|
||||
max_history_tokens: int | None,
|
||||
db_session: Session,
|
||||
answer_generation_timeout: int = QA_TIMEOUT,
|
||||
enable_reflexion: bool = False,
|
||||
bypass_acl: bool = False,
|
||||
use_citations: bool = False,
|
||||
danswerbot_flow: bool = False,
|
||||
retrieval_metrics_callback: (
|
||||
Callable[[RetrievalMetricsContainer], None] | None
|
||||
) = None,
|
||||
rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
|
||||
) -> OneShotQAResponse:
|
||||
"""Collects the streamed one shot answer responses into a single object"""
|
||||
qa_response = OneShotQAResponse()
|
||||
|
||||
results = stream_answer_objects(
|
||||
query_req=query_req,
|
||||
user=user,
|
||||
max_document_tokens=max_document_tokens,
|
||||
max_history_tokens=max_history_tokens,
|
||||
db_session=db_session,
|
||||
bypass_acl=bypass_acl,
|
||||
use_citations=use_citations,
|
||||
danswerbot_flow=danswerbot_flow,
|
||||
timeout=answer_generation_timeout,
|
||||
retrieval_metrics_callback=retrieval_metrics_callback,
|
||||
rerank_metrics_callback=rerank_metrics_callback,
|
||||
)
|
||||
|
||||
answer = ""
|
||||
for packet in results:
|
||||
if isinstance(packet, QueryRephrase):
|
||||
qa_response.rephrase = packet.rephrased_query
|
||||
if isinstance(packet, DanswerAnswerPiece) and packet.answer_piece:
|
||||
answer += packet.answer_piece
|
||||
elif isinstance(packet, QADocsResponse):
|
||||
qa_response.docs = packet
|
||||
elif isinstance(packet, LLMRelevanceFilterResponse):
|
||||
qa_response.llm_selected_doc_indices = packet.llm_selected_doc_indices
|
||||
elif isinstance(packet, DanswerQuotes):
|
||||
qa_response.quotes = packet
|
||||
elif isinstance(packet, CitationInfo):
|
||||
if qa_response.citations:
|
||||
qa_response.citations.append(packet)
|
||||
else:
|
||||
qa_response.citations = [packet]
|
||||
elif isinstance(packet, DanswerContexts):
|
||||
qa_response.contexts = packet
|
||||
elif isinstance(packet, StreamingError):
|
||||
qa_response.error_msg = packet.error
|
||||
elif isinstance(packet, ChatMessageDetail):
|
||||
qa_response.chat_message_id = packet.message_id
|
||||
|
||||
if answer:
|
||||
qa_response.answer = answer
|
||||
|
||||
if enable_reflexion:
|
||||
# Because follow up messages are explicitly tagged, we don't need to verify the answer
|
||||
if len(query_req.messages) == 1:
|
||||
first_query = query_req.messages[0].message
|
||||
qa_response.answer_valid = get_answer_validity(first_query, answer)
|
||||
else:
|
||||
qa_response.answer_valid = True
|
||||
|
||||
if use_citations and qa_response.answer and qa_response.citations:
|
||||
# Reorganize citation nums to be in the same order as the answer
|
||||
qa_response.answer, qa_response.citations = reorganize_citations(
|
||||
qa_response.answer, qa_response.citations
|
||||
)
|
||||
|
||||
return qa_response
|
||||
@ -1,114 +0,0 @@
|
||||
from typing import Any
|
||||
|
||||
from pydantic import BaseModel
|
||||
from pydantic import Field
|
||||
from pydantic import model_validator
|
||||
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import DanswerContexts
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import QADocsResponse
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.context.search.enums import LLMEvaluationType
|
||||
from danswer.context.search.enums import RecencyBiasSetting
|
||||
from danswer.context.search.enums import SearchType
|
||||
from danswer.context.search.models import ChunkContext
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
|
||||
|
||||
class QueryRephrase(BaseModel):
|
||||
rephrased_query: str
|
||||
|
||||
|
||||
class ThreadMessage(BaseModel):
|
||||
message: str
|
||||
sender: str | None = None
|
||||
role: MessageType = MessageType.USER
|
||||
|
||||
|
||||
class PromptConfig(BaseModel):
|
||||
name: str
|
||||
description: str = ""
|
||||
system_prompt: str
|
||||
task_prompt: str = ""
|
||||
include_citations: bool = True
|
||||
datetime_aware: bool = True
|
||||
|
||||
|
||||
class ToolConfig(BaseModel):
|
||||
id: int
|
||||
|
||||
|
||||
class PersonaConfig(BaseModel):
|
||||
name: str
|
||||
description: str
|
||||
search_type: SearchType = SearchType.SEMANTIC
|
||||
num_chunks: float | None = None
|
||||
llm_relevance_filter: bool = False
|
||||
llm_filter_extraction: bool = False
|
||||
recency_bias: RecencyBiasSetting = RecencyBiasSetting.AUTO
|
||||
llm_model_provider_override: str | None = None
|
||||
llm_model_version_override: str | None = None
|
||||
|
||||
prompts: list[PromptConfig] = Field(default_factory=list)
|
||||
prompt_ids: list[int] = Field(default_factory=list)
|
||||
|
||||
document_set_ids: list[int] = Field(default_factory=list)
|
||||
tools: list[ToolConfig] = Field(default_factory=list)
|
||||
tool_ids: list[int] = Field(default_factory=list)
|
||||
custom_tools_openapi: list[dict[str, Any]] = Field(default_factory=list)
|
||||
|
||||
|
||||
class DirectQARequest(ChunkContext):
|
||||
persona_config: PersonaConfig | None = None
|
||||
persona_id: int | None = None
|
||||
|
||||
messages: list[ThreadMessage]
|
||||
prompt_id: int | None = None
|
||||
multilingual_query_expansion: list[str] | None = None
|
||||
retrieval_options: RetrievalDetails = Field(default_factory=RetrievalDetails)
|
||||
rerank_settings: RerankingDetails | None = None
|
||||
evaluation_type: LLMEvaluationType = LLMEvaluationType.UNSPECIFIED
|
||||
|
||||
chain_of_thought: bool = False
|
||||
return_contexts: bool = False
|
||||
|
||||
# allows the caller to specify the exact search query they want to use
|
||||
# can be used if the message sent to the LLM / query should not be the same
|
||||
# will also disable Thread-based Rewording if specified
|
||||
query_override: str | None = None
|
||||
|
||||
# If True, skips generative an AI response to the search query
|
||||
skip_gen_ai_answer_generation: bool = False
|
||||
|
||||
@model_validator(mode="after")
|
||||
def check_persona_fields(self) -> "DirectQARequest":
|
||||
if (self.persona_config is None) == (self.persona_id is None):
|
||||
raise ValueError("Exactly one of persona_config or persona_id must be set")
|
||||
return self
|
||||
|
||||
@model_validator(mode="after")
|
||||
def check_chain_of_thought_and_prompt_id(self) -> "DirectQARequest":
|
||||
if self.chain_of_thought and self.prompt_id is not None:
|
||||
raise ValueError(
|
||||
"If chain_of_thought is True, prompt_id must be None"
|
||||
"The chain of thought prompt is only for question "
|
||||
"answering and does not accept customizing."
|
||||
)
|
||||
|
||||
return self
|
||||
|
||||
|
||||
class OneShotQAResponse(BaseModel):
|
||||
# This is built piece by piece, any of these can be None as the flow could break
|
||||
answer: str | None = None
|
||||
rephrase: str | None = None
|
||||
quotes: DanswerQuotes | None = None
|
||||
citations: list[CitationInfo] | None = None
|
||||
docs: QADocsResponse | None = None
|
||||
llm_selected_doc_indices: list[int] | None = None
|
||||
error_msg: str | None = None
|
||||
answer_valid: bool = True # Reflexion result, default True if Reflexion not run
|
||||
chat_message_id: int | None = None
|
||||
contexts: DanswerContexts | None = None
|
||||
@ -1,81 +0,0 @@
|
||||
from collections.abc import Generator
|
||||
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.natural_language_processing.utils import BaseTokenizer
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.utils.logger import setup_logger
|
||||
|
||||
logger = setup_logger()
|
||||
|
||||
|
||||
def simulate_streaming_response(model_out: str) -> Generator[str, None, None]:
|
||||
"""Mock streaming by generating the passed in model output, character by character"""
|
||||
for token in model_out:
|
||||
yield token
|
||||
|
||||
|
||||
def combine_message_thread(
|
||||
messages: list[ThreadMessage],
|
||||
max_tokens: int | None,
|
||||
llm_tokenizer: BaseTokenizer,
|
||||
) -> str:
|
||||
"""Used to create a single combined message context from threads"""
|
||||
if not messages:
|
||||
return ""
|
||||
|
||||
message_strs: list[str] = []
|
||||
total_token_count = 0
|
||||
|
||||
for message in reversed(messages):
|
||||
if message.role == MessageType.USER:
|
||||
role_str = message.role.value.upper()
|
||||
if message.sender:
|
||||
role_str += " " + message.sender
|
||||
else:
|
||||
# Since other messages might have the user identifying information
|
||||
# better to use Unknown for symmetry
|
||||
role_str += " Unknown"
|
||||
else:
|
||||
role_str = message.role.value.upper()
|
||||
|
||||
msg_str = f"{role_str}:\n{message.message}"
|
||||
message_token_count = len(llm_tokenizer.encode(msg_str))
|
||||
|
||||
if (
|
||||
max_tokens is not None
|
||||
and total_token_count + message_token_count > max_tokens
|
||||
):
|
||||
break
|
||||
|
||||
message_strs.insert(0, msg_str)
|
||||
total_token_count += message_token_count
|
||||
|
||||
return "\n\n".join(message_strs)
|
||||
|
||||
|
||||
def slackify_message(message: ThreadMessage) -> str:
|
||||
if message.role != MessageType.USER:
|
||||
return message.message
|
||||
|
||||
return f"{message.sender or 'Unknown User'} said in Slack:\n{message.message}"
|
||||
|
||||
|
||||
def slackify_message_thread(messages: list[ThreadMessage]) -> str:
|
||||
if not messages:
|
||||
return ""
|
||||
|
||||
message_strs: list[str] = []
|
||||
for message in messages:
|
||||
if message.role == MessageType.USER:
|
||||
message_text = (
|
||||
f"{message.sender or 'Unknown User'} said in Slack:\n{message.message}"
|
||||
)
|
||||
elif message.role == MessageType.ASSISTANT:
|
||||
message_text = f"DanswerBot said in Slack:\n{message.message}"
|
||||
else:
|
||||
message_text = (
|
||||
f"{message.role.value.upper()} said in Slack:\n{message.message}"
|
||||
)
|
||||
message_strs.append(message_text)
|
||||
|
||||
return "\n\n".join(message_strs)
|
||||
@ -1,4 +1,5 @@
|
||||
from datetime import datetime
|
||||
from enum import Enum
|
||||
from typing import TYPE_CHECKING
|
||||
|
||||
from pydantic import BaseModel
|
||||
@ -15,7 +16,6 @@ from danswer.danswerbot.slack.config import VALID_SLACK_FILTERS
|
||||
from danswer.db.models import AllowedAnswerFilters
|
||||
from danswer.db.models import ChannelConfig
|
||||
from danswer.db.models import SlackBot as SlackAppModel
|
||||
from danswer.db.models import SlackBotResponseType
|
||||
from danswer.db.models import SlackChannelConfig as SlackChannelConfigModel
|
||||
from danswer.db.models import User
|
||||
from danswer.server.features.persona.models import PersonaSnapshot
|
||||
@ -148,6 +148,12 @@ class SlackBotTokens(BaseModel):
|
||||
model_config = ConfigDict(frozen=True)
|
||||
|
||||
|
||||
# TODO No longer in use, remove later
|
||||
class SlackBotResponseType(str, Enum):
|
||||
QUOTES = "quotes"
|
||||
CITATIONS = "citations"
|
||||
|
||||
|
||||
class SlackChannelConfigCreationRequest(BaseModel):
|
||||
slack_bot_id: int
|
||||
# currently, a persona is created for each Slack channel config
|
||||
@ -197,7 +203,6 @@ class SlackChannelConfig(BaseModel):
|
||||
id: int
|
||||
persona: PersonaSnapshot | None
|
||||
channel_config: ChannelConfig
|
||||
response_type: SlackBotResponseType
|
||||
# XXX this is going away soon
|
||||
standard_answer_categories: list[StandardAnswerCategory]
|
||||
enable_auto_filters: bool
|
||||
@ -217,7 +222,6 @@ class SlackChannelConfig(BaseModel):
|
||||
else None
|
||||
),
|
||||
channel_config=slack_channel_config_model.channel_config,
|
||||
response_type=slack_channel_config_model.response_type,
|
||||
# XXX this is going away soon
|
||||
standard_answer_categories=[
|
||||
StandardAnswerCategory.from_model(standard_answer_category_model)
|
||||
|
||||
@ -118,7 +118,6 @@ def create_slack_channel_config(
|
||||
slack_bot_id=slack_channel_config_creation_request.slack_bot_id,
|
||||
persona_id=persona_id,
|
||||
channel_config=channel_config,
|
||||
response_type=slack_channel_config_creation_request.response_type,
|
||||
standard_answer_category_ids=slack_channel_config_creation_request.standard_answer_categories,
|
||||
db_session=db_session,
|
||||
enable_auto_filters=slack_channel_config_creation_request.enable_auto_filters,
|
||||
@ -182,7 +181,6 @@ def patch_slack_channel_config(
|
||||
slack_channel_config_id=slack_channel_config_id,
|
||||
persona_id=persona_id,
|
||||
channel_config=channel_config,
|
||||
response_type=slack_channel_config_creation_request.response_type,
|
||||
standard_answer_category_ids=slack_channel_config_creation_request.standard_answer_categories,
|
||||
enable_auto_filters=slack_channel_config_creation_request.enable_auto_filters,
|
||||
)
|
||||
|
||||
@ -109,6 +109,7 @@ def process_run_in_background(
|
||||
prompt_id=chat_session.persona.prompts[0].id,
|
||||
search_doc_ids=None,
|
||||
retrieval_options=search_tool_retrieval_details, # Adjust as needed
|
||||
rerank_settings=None,
|
||||
query_override=None,
|
||||
regenerate=None,
|
||||
llm_override=None,
|
||||
|
||||
@ -5,12 +5,14 @@ from uuid import UUID
|
||||
from pydantic import BaseModel
|
||||
from pydantic import model_validator
|
||||
|
||||
from danswer.chat.models import PersonaOverrideConfig
|
||||
from danswer.chat.models import RetrievalDocs
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.configs.constants import SearchFeedbackType
|
||||
from danswer.context.search.models import BaseFilters
|
||||
from danswer.context.search.models import ChunkContext
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.context.search.models import SearchDoc
|
||||
from danswer.context.search.models import Tag
|
||||
@ -87,6 +89,8 @@ class CreateChatMessageRequest(ChunkContext):
|
||||
# If search_doc_ids provided, then retrieval options are unused
|
||||
search_doc_ids: list[int] | None
|
||||
retrieval_options: RetrievalDetails | None
|
||||
# Useable via the APIs but not recommended for most flows
|
||||
rerank_settings: RerankingDetails | None = None
|
||||
# allows the caller to specify the exact search query they want to use
|
||||
# will disable Query Rewording if specified
|
||||
query_override: str | None = None
|
||||
@ -102,6 +106,10 @@ class CreateChatMessageRequest(ChunkContext):
|
||||
# allow user to specify an alternate assistnat
|
||||
alternate_assistant_id: int | None = None
|
||||
|
||||
# This takes the priority over the prompt_override
|
||||
# This won't be a type that's passed in directly from the API
|
||||
persona_override_config: PersonaOverrideConfig | None = None
|
||||
|
||||
# used for seeded chats to kick off the generation of an AI answer
|
||||
use_existing_user_message: bool = False
|
||||
|
||||
@ -145,7 +153,7 @@ class RenameChatSessionResponse(BaseModel):
|
||||
|
||||
class ChatSessionDetails(BaseModel):
|
||||
id: UUID
|
||||
name: str
|
||||
name: str | None
|
||||
persona_id: int | None = None
|
||||
time_created: str
|
||||
shared_status: ChatSessionSharedStatus
|
||||
@ -198,14 +206,14 @@ class ChatMessageDetail(BaseModel):
|
||||
|
||||
class SearchSessionDetailResponse(BaseModel):
|
||||
search_session_id: UUID
|
||||
description: str
|
||||
description: str | None
|
||||
documents: list[SearchDoc]
|
||||
messages: list[ChatMessageDetail]
|
||||
|
||||
|
||||
class ChatSessionDetailResponse(BaseModel):
|
||||
chat_session_id: UUID
|
||||
description: str
|
||||
description: str | None
|
||||
persona_id: int | None = None
|
||||
persona_name: str | None
|
||||
messages: list[ChatMessageDetail]
|
||||
|
||||
@ -1,15 +1,11 @@
|
||||
import json
|
||||
from collections.abc import Generator
|
||||
from uuid import UUID
|
||||
|
||||
from fastapi import APIRouter
|
||||
from fastapi import Depends
|
||||
from fastapi import HTTPException
|
||||
from fastapi.responses import StreamingResponse
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.auth.users import current_curator_or_admin_user
|
||||
from danswer.auth.users import current_limited_user
|
||||
from danswer.auth.users import current_user
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.configs.constants import MessageType
|
||||
@ -32,8 +28,6 @@ from danswer.db.search_settings import get_current_search_settings
|
||||
from danswer.db.tag import find_tags
|
||||
from danswer.document_index.factory import get_default_document_index
|
||||
from danswer.document_index.vespa.index import VespaIndex
|
||||
from danswer.one_shot_answer.answer_question import stream_search_answer
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.server.query_and_chat.models import AdminSearchRequest
|
||||
from danswer.server.query_and_chat.models import AdminSearchResponse
|
||||
from danswer.server.query_and_chat.models import ChatSessionDetails
|
||||
@ -41,7 +35,6 @@ from danswer.server.query_and_chat.models import ChatSessionsResponse
|
||||
from danswer.server.query_and_chat.models import SearchSessionDetailResponse
|
||||
from danswer.server.query_and_chat.models import SourceTag
|
||||
from danswer.server.query_and_chat.models import TagResponse
|
||||
from danswer.server.query_and_chat.token_limit import check_token_rate_limits
|
||||
from danswer.utils.logger import setup_logger
|
||||
|
||||
logger = setup_logger()
|
||||
@ -140,7 +133,7 @@ def get_user_search_sessions(
|
||||
|
||||
try:
|
||||
search_sessions = get_chat_sessions_by_user(
|
||||
user_id=user_id, deleted=False, db_session=db_session, only_one_shot=True
|
||||
user_id=user_id, deleted=False, db_session=db_session
|
||||
)
|
||||
except ValueError:
|
||||
raise HTTPException(
|
||||
@ -229,29 +222,3 @@ def get_search_session(
|
||||
],
|
||||
)
|
||||
return response
|
||||
|
||||
|
||||
@basic_router.post("/stream-answer-with-quote")
|
||||
def get_answer_with_quote(
|
||||
query_request: DirectQARequest,
|
||||
user: User = Depends(current_limited_user),
|
||||
_: None = Depends(check_token_rate_limits),
|
||||
) -> StreamingResponse:
|
||||
query = query_request.messages[0].message
|
||||
|
||||
logger.notice(f"Received query for one shot answer with quotes: {query}")
|
||||
|
||||
def stream_generator() -> Generator[str, None, None]:
|
||||
try:
|
||||
for packet in stream_search_answer(
|
||||
query_req=query_request,
|
||||
user=user,
|
||||
max_document_tokens=None,
|
||||
max_history_tokens=0,
|
||||
):
|
||||
yield json.dumps(packet) if isinstance(packet, dict) else packet
|
||||
except Exception as e:
|
||||
logger.exception("Error in search answer streaming")
|
||||
yield json.dumps({"error": str(e)})
|
||||
|
||||
return StreamingResponse(stream_generator(), media_type="application/json")
|
||||
|
||||
@ -13,6 +13,7 @@ from danswer.configs.chat_configs import BING_API_KEY
|
||||
from danswer.configs.model_configs import GEN_AI_TEMPERATURE
|
||||
from danswer.context.search.enums import LLMEvaluationType
|
||||
from danswer.context.search.models import InferenceSection
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.db.llm import fetch_existing_llm_providers
|
||||
from danswer.db.models import Persona
|
||||
@ -102,11 +103,14 @@ class SearchToolConfig(BaseModel):
|
||||
default_factory=DocumentPruningConfig
|
||||
)
|
||||
retrieval_options: RetrievalDetails = Field(default_factory=RetrievalDetails)
|
||||
rerank_settings: RerankingDetails | None = None
|
||||
selected_sections: list[InferenceSection] | None = None
|
||||
chunks_above: int = 0
|
||||
chunks_below: int = 0
|
||||
full_doc: bool = False
|
||||
latest_query_files: list[InMemoryChatFile] | None = None
|
||||
# Use with care, should only be used for DanswerBot in channels with multiple users
|
||||
bypass_acl: bool = False
|
||||
|
||||
|
||||
class InternetSearchToolConfig(BaseModel):
|
||||
@ -170,6 +174,8 @@ def construct_tools(
|
||||
if persona.llm_relevance_filter
|
||||
else LLMEvaluationType.SKIP
|
||||
),
|
||||
rerank_settings=search_tool_config.rerank_settings,
|
||||
bypass_acl=search_tool_config.bypass_acl,
|
||||
)
|
||||
tool_dict[db_tool_model.id] = [search_tool]
|
||||
|
||||
|
||||
@ -19,6 +19,7 @@ from danswer.context.search.enums import QueryFlow
|
||||
from danswer.context.search.enums import SearchType
|
||||
from danswer.context.search.models import IndexFilters
|
||||
from danswer.context.search.models import InferenceSection
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.context.search.models import SearchRequest
|
||||
from danswer.context.search.pipeline import SearchPipeline
|
||||
@ -103,6 +104,7 @@ class SearchTool(Tool):
|
||||
chunks_below: int | None = None,
|
||||
full_doc: bool = False,
|
||||
bypass_acl: bool = False,
|
||||
rerank_settings: RerankingDetails | None = None,
|
||||
) -> None:
|
||||
self.user = user
|
||||
self.persona = persona
|
||||
@ -118,6 +120,9 @@ class SearchTool(Tool):
|
||||
self.bypass_acl = bypass_acl
|
||||
self.db_session = db_session
|
||||
|
||||
# Only used via API
|
||||
self.rerank_settings = rerank_settings
|
||||
|
||||
self.chunks_above = (
|
||||
chunks_above
|
||||
if chunks_above is not None
|
||||
@ -292,6 +297,7 @@ class SearchTool(Tool):
|
||||
self.retrieval_options.offset if self.retrieval_options else None
|
||||
),
|
||||
limit=self.retrieval_options.limit if self.retrieval_options else None,
|
||||
rerank_settings=self.rerank_settings,
|
||||
chunks_above=self.chunks_above,
|
||||
chunks_below=self.chunks_below,
|
||||
full_doc=self.full_doc,
|
||||
|
||||
41
backend/ee/danswer/chat/process_message.py
Normal file
41
backend/ee/danswer/chat/process_message.py
Normal file
@ -0,0 +1,41 @@
|
||||
from danswer.chat.models import AllCitations
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerContexts
|
||||
from danswer.chat.models import LLMRelevanceFilterResponse
|
||||
from danswer.chat.models import QADocsResponse
|
||||
from danswer.chat.models import StreamingError
|
||||
from danswer.chat.process_message import ChatPacketStream
|
||||
from danswer.server.query_and_chat.models import ChatMessageDetail
|
||||
from danswer.utils.timing import log_function_time
|
||||
from ee.danswer.server.query_and_chat.models import OneShotQAResponse
|
||||
|
||||
|
||||
@log_function_time()
|
||||
def gather_stream_for_answer_api(
|
||||
packets: ChatPacketStream,
|
||||
) -> OneShotQAResponse:
|
||||
response = OneShotQAResponse()
|
||||
|
||||
answer = ""
|
||||
for packet in packets:
|
||||
if isinstance(packet, DanswerAnswerPiece) and packet.answer_piece:
|
||||
answer += packet.answer_piece
|
||||
elif isinstance(packet, QADocsResponse):
|
||||
response.docs = packet
|
||||
# Extraneous, provided for backwards compatibility
|
||||
response.rephrase = packet.rephrased_query
|
||||
elif isinstance(packet, StreamingError):
|
||||
response.error_msg = packet.error
|
||||
elif isinstance(packet, ChatMessageDetail):
|
||||
response.chat_message_id = packet.message_id
|
||||
elif isinstance(packet, LLMRelevanceFilterResponse):
|
||||
response.llm_selected_doc_indices = packet.llm_selected_doc_indices
|
||||
elif isinstance(packet, AllCitations):
|
||||
response.citations = packet.citations
|
||||
elif isinstance(packet, DanswerContexts):
|
||||
response.contexts = packet
|
||||
|
||||
if answer:
|
||||
response.answer = answer
|
||||
|
||||
return response
|
||||
@ -155,7 +155,6 @@ def _handle_standard_answers(
|
||||
else 0,
|
||||
danswerbot_flow=True,
|
||||
slack_thread_id=slack_thread_id,
|
||||
one_shot=True,
|
||||
)
|
||||
|
||||
root_message = get_or_create_root_message(
|
||||
|
||||
@ -33,12 +33,7 @@ def get_empty_chat_messages_entries__paginated(
|
||||
|
||||
message_skeletons: list[ChatMessageSkeleton] = []
|
||||
for chat_session in chat_sessions:
|
||||
if chat_session.one_shot:
|
||||
flow_type = FlowType.SEARCH
|
||||
elif chat_session.danswerbot_flow:
|
||||
flow_type = FlowType.SLACK
|
||||
else:
|
||||
flow_type = FlowType.CHAT
|
||||
flow_type = FlowType.SLACK if chat_session.danswerbot_flow else FlowType.CHAT
|
||||
|
||||
for message in chat_session.messages:
|
||||
# Only count user messages
|
||||
|
||||
@ -6,6 +6,7 @@ from fastapi import HTTPException
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.auth.users import current_user
|
||||
from danswer.chat.chat_utils import combine_message_thread
|
||||
from danswer.chat.chat_utils import create_chat_chain
|
||||
from danswer.chat.models import AllCitations
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
@ -16,8 +17,8 @@ from danswer.chat.models import QADocsResponse
|
||||
from danswer.chat.models import StreamingError
|
||||
from danswer.chat.process_message import ChatPacketStream
|
||||
from danswer.chat.process_message import stream_chat_message_objects
|
||||
from danswer.configs.chat_configs import CHAT_TARGET_CHUNK_PERCENTAGE
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_TARGET_CHUNK_PERCENTAGE
|
||||
from danswer.context.search.models import OptionalSearchSetting
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.context.search.models import SavedSearchDoc
|
||||
@ -29,7 +30,6 @@ from danswer.db.models import User
|
||||
from danswer.llm.factory import get_llms_for_persona
|
||||
from danswer.llm.utils import get_max_input_tokens
|
||||
from danswer.natural_language_processing.utils import get_tokenizer
|
||||
from danswer.one_shot_answer.qa_utils import combine_message_thread
|
||||
from danswer.secondary_llm_flows.query_expansion import thread_based_query_rephrase
|
||||
from danswer.server.query_and_chat.models import ChatMessageDetail
|
||||
from danswer.server.query_and_chat.models import CreateChatMessageRequest
|
||||
@ -171,6 +171,8 @@ def handle_simplified_chat_message(
|
||||
prompt_id=None,
|
||||
search_doc_ids=chat_message_req.search_doc_ids,
|
||||
retrieval_options=retrieval_options,
|
||||
# Simple API does not support reranking, hide complexity from user
|
||||
rerank_settings=None,
|
||||
query_override=chat_message_req.query_override,
|
||||
# Currently only applies to search flow not chat
|
||||
chunks_above=0,
|
||||
@ -232,7 +234,6 @@ def handle_send_message_simple_with_history(
|
||||
description="handle_send_message_simple_with_history",
|
||||
user_id=user_id,
|
||||
persona_id=req.persona_id,
|
||||
one_shot=False,
|
||||
)
|
||||
|
||||
llm, _ = get_llms_for_persona(persona=chat_session.persona)
|
||||
@ -245,7 +246,7 @@ def handle_send_message_simple_with_history(
|
||||
input_tokens = get_max_input_tokens(
|
||||
model_name=llm.config.model_name, model_provider=llm.config.model_provider
|
||||
)
|
||||
max_history_tokens = int(input_tokens * DANSWER_BOT_TARGET_CHUNK_PERCENTAGE)
|
||||
max_history_tokens = int(input_tokens * CHAT_TARGET_CHUNK_PERCENTAGE)
|
||||
|
||||
# Every chat Session begins with an empty root message
|
||||
root_message = get_or_create_root_message(
|
||||
@ -293,6 +294,8 @@ def handle_send_message_simple_with_history(
|
||||
prompt_id=req.prompt_id,
|
||||
search_doc_ids=req.search_doc_ids,
|
||||
retrieval_options=retrieval_options,
|
||||
# Simple API does not support reranking, hide complexity from user
|
||||
rerank_settings=None,
|
||||
query_override=rephrased_query,
|
||||
chunks_above=0,
|
||||
chunks_below=0,
|
||||
|
||||
@ -2,7 +2,13 @@ from uuid import UUID
|
||||
|
||||
from pydantic import BaseModel
|
||||
from pydantic import Field
|
||||
from pydantic import model_validator
|
||||
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import DanswerContexts
|
||||
from danswer.chat.models import PersonaOverrideConfig
|
||||
from danswer.chat.models import QADocsResponse
|
||||
from danswer.chat.models import ThreadMessage
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.context.search.enums import LLMEvaluationType
|
||||
from danswer.context.search.enums import SearchType
|
||||
@ -10,7 +16,6 @@ from danswer.context.search.models import ChunkContext
|
||||
from danswer.context.search.models import RerankingDetails
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.context.search.models import SavedSearchDoc
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from ee.danswer.server.manage.models import StandardAnswer
|
||||
|
||||
|
||||
@ -96,3 +101,48 @@ class ChatBasicResponse(BaseModel):
|
||||
# TODO: deprecate both of these
|
||||
simple_search_docs: list[SimpleDoc] | None = None
|
||||
llm_chunks_indices: list[int] | None = None
|
||||
|
||||
|
||||
class OneShotQARequest(ChunkContext):
|
||||
# Supports simplier APIs that don't deal with chat histories or message edits
|
||||
# Easier APIs to work with for developers
|
||||
persona_override_config: PersonaOverrideConfig | None = None
|
||||
persona_id: int | None = None
|
||||
|
||||
messages: list[ThreadMessage]
|
||||
prompt_id: int | None = None
|
||||
retrieval_options: RetrievalDetails = Field(default_factory=RetrievalDetails)
|
||||
rerank_settings: RerankingDetails | None = None
|
||||
return_contexts: bool = False
|
||||
|
||||
# allows the caller to specify the exact search query they want to use
|
||||
# can be used if the message sent to the LLM / query should not be the same
|
||||
# will also disable Thread-based Rewording if specified
|
||||
query_override: str | None = None
|
||||
|
||||
# If True, skips generative an AI response to the search query
|
||||
skip_gen_ai_answer_generation: bool = False
|
||||
|
||||
@model_validator(mode="after")
|
||||
def check_persona_fields(self) -> "OneShotQARequest":
|
||||
if self.persona_override_config is None and self.persona_id is None:
|
||||
raise ValueError("Exactly one of persona_config or persona_id must be set")
|
||||
elif self.persona_override_config is not None and (
|
||||
self.persona_id is not None or self.prompt_id is not None
|
||||
):
|
||||
raise ValueError(
|
||||
"If persona_override_config is set, persona_id and prompt_id cannot be set"
|
||||
)
|
||||
return self
|
||||
|
||||
|
||||
class OneShotQAResponse(BaseModel):
|
||||
# This is built piece by piece, any of these can be None as the flow could break
|
||||
answer: str | None = None
|
||||
rephrase: str | None = None
|
||||
citations: list[CitationInfo] | None = None
|
||||
docs: QADocsResponse | None = None
|
||||
llm_selected_doc_indices: list[int] | None = None
|
||||
error_msg: str | None = None
|
||||
chat_message_id: int | None = None
|
||||
contexts: DanswerContexts | None = None
|
||||
|
||||
@ -1,38 +1,47 @@
|
||||
import json
|
||||
from collections.abc import Generator
|
||||
|
||||
from fastapi import APIRouter
|
||||
from fastapi import Depends
|
||||
from fastapi import HTTPException
|
||||
from fastapi.responses import StreamingResponse
|
||||
from pydantic import BaseModel
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.auth.users import current_user
|
||||
from danswer.configs.danswerbot_configs import DANSWER_BOT_TARGET_CHUNK_PERCENTAGE
|
||||
from danswer.chat.chat_utils import combine_message_thread
|
||||
from danswer.chat.chat_utils import prepare_chat_message_request
|
||||
from danswer.chat.models import PersonaOverrideConfig
|
||||
from danswer.chat.process_message import ChatPacketStream
|
||||
from danswer.chat.process_message import stream_chat_message_objects
|
||||
from danswer.configs.danswerbot_configs import MAX_THREAD_CONTEXT_PERCENTAGE
|
||||
from danswer.context.search.models import SavedSearchDocWithContent
|
||||
from danswer.context.search.models import SearchRequest
|
||||
from danswer.context.search.pipeline import SearchPipeline
|
||||
from danswer.context.search.utils import dedupe_documents
|
||||
from danswer.context.search.utils import drop_llm_indices
|
||||
from danswer.context.search.utils import relevant_sections_to_indices
|
||||
from danswer.db.chat import get_prompt_by_id
|
||||
from danswer.db.engine import get_session
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_persona_by_id
|
||||
from danswer.llm.answering.prompts.citations_prompt import (
|
||||
compute_max_document_tokens_for_persona,
|
||||
)
|
||||
from danswer.llm.factory import get_default_llms
|
||||
from danswer.llm.factory import get_llms_for_persona
|
||||
from danswer.llm.factory import get_main_llm_from_tuple
|
||||
from danswer.llm.utils import get_max_input_tokens
|
||||
from danswer.one_shot_answer.answer_question import get_search_answer
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.one_shot_answer.models import OneShotQAResponse
|
||||
from danswer.natural_language_processing.utils import get_tokenizer
|
||||
from danswer.server.utils import get_json_line
|
||||
from danswer.utils.logger import setup_logger
|
||||
from ee.danswer.chat.process_message import gather_stream_for_answer_api
|
||||
from ee.danswer.danswerbot.slack.handlers.handle_standard_answers import (
|
||||
oneoff_standard_answers,
|
||||
)
|
||||
from ee.danswer.server.query_and_chat.models import DocumentSearchRequest
|
||||
from ee.danswer.server.query_and_chat.models import OneShotQARequest
|
||||
from ee.danswer.server.query_and_chat.models import OneShotQAResponse
|
||||
from ee.danswer.server.query_and_chat.models import StandardAnswerRequest
|
||||
from ee.danswer.server.query_and_chat.models import StandardAnswerResponse
|
||||
from ee.danswer.server.query_and_chat.utils import create_temporary_persona
|
||||
|
||||
|
||||
logger = setup_logger()
|
||||
@ -125,58 +134,115 @@ def handle_search_request(
|
||||
return DocumentSearchResponse(top_documents=deduped_docs, llm_indices=llm_indices)
|
||||
|
||||
|
||||
@basic_router.post("/answer-with-quote")
|
||||
def get_answer_with_quote(
|
||||
query_request: DirectQARequest,
|
||||
def get_answer_stream(
|
||||
query_request: OneShotQARequest,
|
||||
user: User | None = Depends(current_user),
|
||||
db_session: Session = Depends(get_session),
|
||||
) -> OneShotQAResponse:
|
||||
) -> ChatPacketStream:
|
||||
query = query_request.messages[0].message
|
||||
logger.notice(f"Received query for one shot answer API with quotes: {query}")
|
||||
logger.notice(f"Received query for Answer API: {query}")
|
||||
|
||||
if query_request.persona_config is not None:
|
||||
new_persona = create_temporary_persona(
|
||||
db_session=db_session,
|
||||
persona_config=query_request.persona_config,
|
||||
if (
|
||||
query_request.persona_override_config is None
|
||||
and query_request.persona_id is None
|
||||
):
|
||||
raise KeyError("Must provide persona ID or Persona Config")
|
||||
|
||||
prompt = None
|
||||
if query_request.prompt_id is not None:
|
||||
prompt = get_prompt_by_id(
|
||||
prompt_id=query_request.prompt_id,
|
||||
user=user,
|
||||
db_session=db_session,
|
||||
)
|
||||
persona = new_persona
|
||||
|
||||
persona_info: Persona | PersonaOverrideConfig | None = None
|
||||
if query_request.persona_override_config is not None:
|
||||
persona_info = query_request.persona_override_config
|
||||
elif query_request.persona_id is not None:
|
||||
persona = get_persona_by_id(
|
||||
persona_info = get_persona_by_id(
|
||||
persona_id=query_request.persona_id,
|
||||
user=user,
|
||||
db_session=db_session,
|
||||
is_for_edit=False,
|
||||
)
|
||||
else:
|
||||
raise KeyError("Must provide persona ID or Persona Config")
|
||||
|
||||
llm = get_main_llm_from_tuple(
|
||||
get_default_llms() if not persona else get_llms_for_persona(persona)
|
||||
llm = get_main_llm_from_tuple(get_llms_for_persona(persona_info))
|
||||
|
||||
llm_tokenizer = get_tokenizer(
|
||||
model_name=llm.config.model_name,
|
||||
provider_type=llm.config.model_provider,
|
||||
)
|
||||
|
||||
input_tokens = get_max_input_tokens(
|
||||
model_name=llm.config.model_name, model_provider=llm.config.model_provider
|
||||
)
|
||||
max_history_tokens = int(input_tokens * DANSWER_BOT_TARGET_CHUNK_PERCENTAGE)
|
||||
max_history_tokens = int(input_tokens * MAX_THREAD_CONTEXT_PERCENTAGE)
|
||||
|
||||
remaining_tokens = input_tokens - max_history_tokens
|
||||
|
||||
max_document_tokens = compute_max_document_tokens_for_persona(
|
||||
persona=persona,
|
||||
actual_user_input=query,
|
||||
max_llm_token_override=remaining_tokens,
|
||||
combined_message = combine_message_thread(
|
||||
messages=query_request.messages,
|
||||
max_tokens=max_history_tokens,
|
||||
llm_tokenizer=llm_tokenizer,
|
||||
)
|
||||
|
||||
answer_details = get_search_answer(
|
||||
query_req=query_request,
|
||||
# Also creates a new chat session
|
||||
request = prepare_chat_message_request(
|
||||
message_text=combined_message,
|
||||
user=user,
|
||||
max_document_tokens=max_document_tokens,
|
||||
max_history_tokens=max_history_tokens,
|
||||
persona_id=query_request.persona_id,
|
||||
persona_override_config=query_request.persona_override_config,
|
||||
prompt=prompt,
|
||||
message_ts_to_respond_to=None,
|
||||
retrieval_details=query_request.retrieval_options,
|
||||
rerank_settings=query_request.rerank_settings,
|
||||
db_session=db_session,
|
||||
)
|
||||
|
||||
return answer_details
|
||||
packets = stream_chat_message_objects(
|
||||
new_msg_req=request,
|
||||
user=user,
|
||||
db_session=db_session,
|
||||
include_contexts=query_request.return_contexts,
|
||||
)
|
||||
|
||||
return packets
|
||||
|
||||
|
||||
@basic_router.post("/answer-with-citation")
|
||||
def get_answer_with_citation(
|
||||
request: OneShotQARequest,
|
||||
db_session: Session = Depends(get_session),
|
||||
user: User | None = Depends(current_user),
|
||||
) -> OneShotQAResponse:
|
||||
try:
|
||||
packets = get_answer_stream(request, user, db_session)
|
||||
answer = gather_stream_for_answer_api(packets)
|
||||
|
||||
if answer.error_msg:
|
||||
raise RuntimeError(answer.error_msg)
|
||||
|
||||
return answer
|
||||
except Exception as e:
|
||||
logger.error(f"Error in get_answer_with_citation: {str(e)}", exc_info=True)
|
||||
raise HTTPException(status_code=500, detail="An internal server error occurred")
|
||||
|
||||
|
||||
@basic_router.post("/stream-answer-with-citation")
|
||||
def stream_answer_with_citation(
|
||||
request: OneShotQARequest,
|
||||
db_session: Session = Depends(get_session),
|
||||
user: User | None = Depends(current_user),
|
||||
) -> StreamingResponse:
|
||||
def stream_generator() -> Generator[str, None, None]:
|
||||
try:
|
||||
for packet in get_answer_stream(request, user, db_session):
|
||||
serialized = get_json_line(packet.model_dump())
|
||||
yield serialized
|
||||
except Exception as e:
|
||||
logger.exception("Error in answer streaming")
|
||||
yield json.dumps({"error": str(e)})
|
||||
|
||||
return StreamingResponse(stream_generator(), media_type="application/json")
|
||||
|
||||
|
||||
@basic_router.get("/standard-answer")
|
||||
|
||||
@ -1,85 +0,0 @@
|
||||
from typing import cast
|
||||
|
||||
from fastapi import HTTPException
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from danswer.auth.users import is_user_admin
|
||||
from danswer.db.llm import fetch_existing_doc_sets
|
||||
from danswer.db.llm import fetch_existing_tools
|
||||
from danswer.db.models import Persona
|
||||
from danswer.db.models import Prompt
|
||||
from danswer.db.models import Tool
|
||||
from danswer.db.models import User
|
||||
from danswer.db.persona import get_prompts_by_ids
|
||||
from danswer.one_shot_answer.models import PersonaConfig
|
||||
from danswer.tools.tool_implementations.custom.custom_tool import (
|
||||
build_custom_tools_from_openapi_schema_and_headers,
|
||||
)
|
||||
|
||||
|
||||
def create_temporary_persona(
|
||||
persona_config: PersonaConfig, db_session: Session, user: User | None = None
|
||||
) -> Persona:
|
||||
if not is_user_admin(user):
|
||||
raise HTTPException(
|
||||
status_code=403,
|
||||
detail="User is not authorized to create a persona in one shot queries",
|
||||
)
|
||||
|
||||
"""Create a temporary Persona object from the provided configuration."""
|
||||
persona = Persona(
|
||||
name=persona_config.name,
|
||||
description=persona_config.description,
|
||||
num_chunks=persona_config.num_chunks,
|
||||
llm_relevance_filter=persona_config.llm_relevance_filter,
|
||||
llm_filter_extraction=persona_config.llm_filter_extraction,
|
||||
recency_bias=persona_config.recency_bias,
|
||||
llm_model_provider_override=persona_config.llm_model_provider_override,
|
||||
llm_model_version_override=persona_config.llm_model_version_override,
|
||||
)
|
||||
|
||||
if persona_config.prompts:
|
||||
persona.prompts = [
|
||||
Prompt(
|
||||
name=p.name,
|
||||
description=p.description,
|
||||
system_prompt=p.system_prompt,
|
||||
task_prompt=p.task_prompt,
|
||||
include_citations=p.include_citations,
|
||||
datetime_aware=p.datetime_aware,
|
||||
)
|
||||
for p in persona_config.prompts
|
||||
]
|
||||
elif persona_config.prompt_ids:
|
||||
persona.prompts = get_prompts_by_ids(
|
||||
db_session=db_session, prompt_ids=persona_config.prompt_ids
|
||||
)
|
||||
|
||||
persona.tools = []
|
||||
if persona_config.custom_tools_openapi:
|
||||
for schema in persona_config.custom_tools_openapi:
|
||||
tools = cast(
|
||||
list[Tool],
|
||||
build_custom_tools_from_openapi_schema_and_headers(schema),
|
||||
)
|
||||
persona.tools.extend(tools)
|
||||
|
||||
if persona_config.tools:
|
||||
tool_ids = [tool.id for tool in persona_config.tools]
|
||||
persona.tools.extend(
|
||||
fetch_existing_tools(db_session=db_session, tool_ids=tool_ids)
|
||||
)
|
||||
|
||||
if persona_config.tool_ids:
|
||||
persona.tools.extend(
|
||||
fetch_existing_tools(
|
||||
db_session=db_session, tool_ids=persona_config.tool_ids
|
||||
)
|
||||
)
|
||||
|
||||
fetched_docs = fetch_existing_doc_sets(
|
||||
db_session=db_session, doc_ids=persona_config.document_set_ids
|
||||
)
|
||||
persona.document_sets = fetched_docs
|
||||
|
||||
return persona
|
||||
@ -179,13 +179,7 @@ class QuestionAnswerPairSnapshot(BaseModel):
|
||||
|
||||
|
||||
def determine_flow_type(chat_session: ChatSession) -> SessionType:
|
||||
return (
|
||||
SessionType.SLACK
|
||||
if chat_session.danswerbot_flow
|
||||
else SessionType.SEARCH
|
||||
if chat_session.one_shot
|
||||
else SessionType.CHAT
|
||||
)
|
||||
return SessionType.SLACK if chat_session.danswerbot_flow else SessionType.CHAT
|
||||
|
||||
|
||||
def fetch_and_process_chat_session_history_minimal(
|
||||
|
||||
@ -9,7 +9,6 @@ from danswer.auth.schemas import UserStatus
|
||||
|
||||
class FlowType(str, Enum):
|
||||
CHAT = "chat"
|
||||
SEARCH = "search"
|
||||
SLACK = "slack"
|
||||
|
||||
|
||||
|
||||
@ -8,8 +8,6 @@ from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.file_store.models import FileDescriptor
|
||||
from danswer.llm.override_models import LLMOverride
|
||||
from danswer.llm.override_models import PromptOverride
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.server.query_and_chat.models import ChatSessionCreationRequest
|
||||
from danswer.server.query_and_chat.models import CreateChatMessageRequest
|
||||
from tests.integration.common_utils.constants import API_SERVER_URL
|
||||
@ -68,6 +66,7 @@ class ChatSessionManager:
|
||||
prompt_id=prompt_id,
|
||||
search_doc_ids=search_doc_ids or [],
|
||||
retrieval_options=retrieval_options,
|
||||
rerank_settings=None, # Can be added if needed
|
||||
query_override=query_override,
|
||||
regenerate=regenerate,
|
||||
llm_override=llm_override,
|
||||
@ -87,30 +86,6 @@ class ChatSessionManager:
|
||||
|
||||
return ChatSessionManager.analyze_response(response)
|
||||
|
||||
@staticmethod
|
||||
def get_answer_with_quote(
|
||||
persona_id: int,
|
||||
message: str,
|
||||
user_performing_action: DATestUser | None = None,
|
||||
) -> StreamedResponse:
|
||||
direct_qa_request = DirectQARequest(
|
||||
messages=[ThreadMessage(message=message)],
|
||||
prompt_id=None,
|
||||
persona_id=persona_id,
|
||||
)
|
||||
|
||||
response = requests.post(
|
||||
f"{API_SERVER_URL}/query/stream-answer-with-quote",
|
||||
json=direct_qa_request.model_dump(),
|
||||
headers=user_performing_action.headers
|
||||
if user_performing_action
|
||||
else GENERAL_HEADERS,
|
||||
stream=True,
|
||||
)
|
||||
response.raise_for_status()
|
||||
|
||||
return ChatSessionManager.analyze_response(response)
|
||||
|
||||
@staticmethod
|
||||
def analyze_response(response: Response) -> StreamedResponse:
|
||||
response_data = [
|
||||
|
||||
@ -1,25 +0,0 @@
|
||||
from tests.integration.common_utils.managers.chat import ChatSessionManager
|
||||
from tests.integration.common_utils.managers.llm_provider import LLMProviderManager
|
||||
from tests.integration.common_utils.managers.user import UserManager
|
||||
from tests.integration.common_utils.test_models import DATestUser
|
||||
|
||||
|
||||
def test_send_message_simple_with_history(reset: None) -> None:
|
||||
admin_user: DATestUser = UserManager.create(name="admin_user")
|
||||
LLMProviderManager.create(user_performing_action=admin_user)
|
||||
|
||||
test_chat_session = ChatSessionManager.create(user_performing_action=admin_user)
|
||||
|
||||
response = ChatSessionManager.get_answer_with_quote(
|
||||
persona_id=test_chat_session.persona_id,
|
||||
message="Hello, this is a test.",
|
||||
user_performing_action=admin_user,
|
||||
)
|
||||
|
||||
assert (
|
||||
response.tool_name is not None
|
||||
), "Tool name should be specified (always search)"
|
||||
assert (
|
||||
response.relevance_summaries is not None
|
||||
), "Relevance summaries should be present for all search streams"
|
||||
assert len(response.full_message) > 0, "Response message should not be empty"
|
||||
@ -1,16 +1,16 @@
|
||||
import requests
|
||||
from retry import retry
|
||||
|
||||
from danswer.chat.models import ThreadMessage
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.configs.constants import MessageType
|
||||
from danswer.connectors.models import InputType
|
||||
from danswer.context.search.enums import OptionalSearchSetting
|
||||
from danswer.context.search.models import IndexFilters
|
||||
from danswer.context.search.models import OptionalSearchSetting
|
||||
from danswer.context.search.models import RetrievalDetails
|
||||
from danswer.db.enums import IndexingStatus
|
||||
from danswer.one_shot_answer.models import DirectQARequest
|
||||
from danswer.one_shot_answer.models import ThreadMessage
|
||||
from danswer.server.documents.models import ConnectorBase
|
||||
from ee.danswer.server.query_and_chat.models import OneShotQARequest
|
||||
from tests.regression.answer_quality.cli_utils import get_api_server_host_port
|
||||
|
||||
GENERAL_HEADERS = {"Content-Type": "application/json"}
|
||||
@ -37,7 +37,7 @@ def get_answer_from_query(
|
||||
|
||||
messages = [ThreadMessage(message=query, sender=None, role=MessageType.USER)]
|
||||
|
||||
new_message_request = DirectQARequest(
|
||||
new_message_request = OneShotQARequest(
|
||||
messages=messages,
|
||||
prompt_id=0,
|
||||
persona_id=0,
|
||||
@ -47,12 +47,11 @@ def get_answer_from_query(
|
||||
filters=filters,
|
||||
enable_auto_detect_filters=False,
|
||||
),
|
||||
chain_of_thought=False,
|
||||
return_contexts=True,
|
||||
skip_gen_ai_answer_generation=only_retrieve_docs,
|
||||
)
|
||||
|
||||
url = _api_url_builder(env_name, "/query/answer-with-quote/")
|
||||
url = _api_url_builder(env_name, "/query/answer-with-citation/")
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
}
|
||||
|
||||
@ -1,352 +0,0 @@
|
||||
import json
|
||||
from datetime import datetime
|
||||
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import LlmDoc
|
||||
from danswer.configs.constants import DocumentSource
|
||||
from danswer.llm.answering.stream_processing.quotes_processing import (
|
||||
QuotesProcessor,
|
||||
)
|
||||
|
||||
mock_docs = [
|
||||
LlmDoc(
|
||||
document_id=f"doc_{int(id/2)}",
|
||||
content="Document is a doc",
|
||||
blurb=f"Document #{id}",
|
||||
semantic_identifier=f"Doc {id}",
|
||||
source_type=DocumentSource.WEB,
|
||||
metadata={},
|
||||
updated_at=datetime.now(),
|
||||
link=f"https://{int(id/2)}.com" if int(id / 2) % 2 == 0 else None,
|
||||
source_links={0: "https://mintlify.com/docs/settings/broken-links"},
|
||||
match_highlights=[],
|
||||
)
|
||||
for id in range(10)
|
||||
]
|
||||
|
||||
|
||||
def _process_tokens(
|
||||
processor: QuotesProcessor, tokens: list[str]
|
||||
) -> tuple[str, list[str]]:
|
||||
"""Process a list of tokens and return the answer and quotes.
|
||||
|
||||
Args:
|
||||
processor: QuotesProcessor instance
|
||||
tokens: List of tokens to process
|
||||
|
||||
Returns:
|
||||
Tuple of (answer_text, list_of_quotes)
|
||||
"""
|
||||
answer = ""
|
||||
quotes: list[str] = []
|
||||
|
||||
# need to add a None to the end to simulate the end of the stream
|
||||
for token in tokens + [None]:
|
||||
for output in processor.process_token(token):
|
||||
if isinstance(output, DanswerAnswerPiece):
|
||||
if output.answer_piece:
|
||||
answer += output.answer_piece
|
||||
elif isinstance(output, DanswerQuotes):
|
||||
quotes.extend(q.quote for q in output.quotes)
|
||||
|
||||
return answer, quotes
|
||||
|
||||
|
||||
def test_process_model_tokens_answer() -> None:
|
||||
tokens_with_quotes = [
|
||||
"{",
|
||||
"\n ",
|
||||
'"answer": "Yes',
|
||||
", Danswer allows",
|
||||
" customized prompts. This",
|
||||
" feature",
|
||||
" is currently being",
|
||||
" developed and implemente",
|
||||
"d to",
|
||||
" improve",
|
||||
" the accuracy",
|
||||
" of",
|
||||
" Language",
|
||||
" Models (",
|
||||
"LL",
|
||||
"Ms) for",
|
||||
" different",
|
||||
" companies",
|
||||
".",
|
||||
" The custom",
|
||||
"ized prompts feature",
|
||||
" woul",
|
||||
"d allow users to ad",
|
||||
"d person",
|
||||
"alized prom",
|
||||
"pts through",
|
||||
" an",
|
||||
" interface or",
|
||||
" metho",
|
||||
"d,",
|
||||
" which would then be used to",
|
||||
" train",
|
||||
" the LLM.",
|
||||
" This enhancement",
|
||||
" aims to make",
|
||||
" Danswer more",
|
||||
" adaptable to",
|
||||
" different",
|
||||
" business",
|
||||
" contexts",
|
||||
" by",
|
||||
" tail",
|
||||
"oring it",
|
||||
" to the specific language",
|
||||
" an",
|
||||
"d terminology",
|
||||
" used within",
|
||||
" a",
|
||||
" company.",
|
||||
" Additionally",
|
||||
",",
|
||||
" Danswer already",
|
||||
" supports creating",
|
||||
" custom AI",
|
||||
" Assistants with",
|
||||
" different",
|
||||
" prom",
|
||||
"pts and backing",
|
||||
" knowledge",
|
||||
" sets",
|
||||
",",
|
||||
" which",
|
||||
" is",
|
||||
" a form",
|
||||
" of prompt",
|
||||
" customization. However, it",
|
||||
"'s important to nLogging Details LiteLLM-Success Call: Noneote that some",
|
||||
" aspects",
|
||||
" of prompt",
|
||||
" customization,",
|
||||
" such as for",
|
||||
" Sl",
|
||||
"ack",
|
||||
"b",
|
||||
"ots, may",
|
||||
" still",
|
||||
" be in",
|
||||
" development or have",
|
||||
' limitations.",',
|
||||
'\n "quotes": [',
|
||||
'\n "We',
|
||||
" woul",
|
||||
"d like to ad",
|
||||
"d customized prompts for",
|
||||
" different",
|
||||
" companies to improve the accuracy of",
|
||||
" Language",
|
||||
" Model",
|
||||
" (LLM)",
|
||||
'.",\n "A',
|
||||
" new",
|
||||
" feature that",
|
||||
" allows users to add personalize",
|
||||
"d prompts.",
|
||||
" This would involve",
|
||||
" creating",
|
||||
" an interface or method for",
|
||||
" users to input",
|
||||
" their",
|
||||
" own",
|
||||
" prom",
|
||||
"pts,",
|
||||
" which would then be used to",
|
||||
' train the LLM.",',
|
||||
'\n "Create',
|
||||
" custom AI Assistants with",
|
||||
" different prompts and backing knowledge",
|
||||
' sets.",',
|
||||
'\n "This',
|
||||
" PR",
|
||||
" fixes",
|
||||
" https",
|
||||
"://github.com/dan",
|
||||
"swer-ai/dan",
|
||||
"swer/issues/1",
|
||||
"584",
|
||||
" by",
|
||||
" setting",
|
||||
" the system",
|
||||
" default",
|
||||
" prompt for",
|
||||
" sl",
|
||||
"ackbots const",
|
||||
"rained by",
|
||||
" ",
|
||||
"document sets",
|
||||
".",
|
||||
" It",
|
||||
" probably",
|
||||
" isn",
|
||||
"'t ideal",
|
||||
" -",
|
||||
" it",
|
||||
" might",
|
||||
" be pref",
|
||||
"erable to be",
|
||||
" able to select",
|
||||
" a prompt for",
|
||||
" the",
|
||||
" slackbot from",
|
||||
" the",
|
||||
" admin",
|
||||
" panel",
|
||||
" -",
|
||||
" but it sol",
|
||||
"ves the immediate problem",
|
||||
" of",
|
||||
" the slack",
|
||||
" listener",
|
||||
" cr",
|
||||
"ashing when",
|
||||
" configure",
|
||||
"d this",
|
||||
' way."\n ]',
|
||||
"\n}",
|
||||
"",
|
||||
]
|
||||
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens_with_quotes)
|
||||
|
||||
s_json = "".join(tokens_with_quotes)
|
||||
j = json.loads(s_json)
|
||||
expected_answer = j["answer"]
|
||||
assert expected_answer == answer
|
||||
# NOTE: no quotes, since the docs don't match the quotes
|
||||
assert len(quotes) == 0
|
||||
|
||||
|
||||
def test_simple_json_answer() -> None:
|
||||
tokens = [
|
||||
"```",
|
||||
"json",
|
||||
"\n",
|
||||
"{",
|
||||
'"answer": "This is a simple ',
|
||||
"answer.",
|
||||
'",\n"',
|
||||
'quotes": []',
|
||||
"\n}",
|
||||
"\n",
|
||||
"```",
|
||||
]
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "This is a simple answer." == answer
|
||||
assert len(quotes) == 0
|
||||
|
||||
|
||||
def test_json_answer_with_quotes() -> None:
|
||||
tokens = [
|
||||
"```",
|
||||
"json",
|
||||
"\n",
|
||||
"{",
|
||||
'"answer": "This ',
|
||||
"is a ",
|
||||
"split ",
|
||||
"answer.",
|
||||
'",\n"',
|
||||
'quotes": []',
|
||||
"\n}",
|
||||
"\n",
|
||||
"```",
|
||||
]
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "This is a split answer." == answer
|
||||
assert len(quotes) == 0
|
||||
|
||||
|
||||
def test_json_answer_with_quotes_one_chunk() -> None:
|
||||
tokens = ['```json\n{"answer": "z",\n"quotes": ["Document"]\n}\n```']
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "z" == answer
|
||||
assert len(quotes) == 1
|
||||
assert quotes[0] == "Document"
|
||||
|
||||
|
||||
def test_json_answer_split_tokens() -> None:
|
||||
tokens = [
|
||||
"```",
|
||||
"json",
|
||||
"\n",
|
||||
"{",
|
||||
'\n"',
|
||||
'answer": "This ',
|
||||
"is a ",
|
||||
"split ",
|
||||
"answer.",
|
||||
'",\n"',
|
||||
'quotes": []',
|
||||
"\n}",
|
||||
"\n",
|
||||
"```",
|
||||
]
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "This is a split answer." == answer
|
||||
assert len(quotes) == 0
|
||||
|
||||
|
||||
def test_lengthy_prefixed_json_with_quotes() -> None:
|
||||
tokens = [
|
||||
"This is my response in json\n\n",
|
||||
"```",
|
||||
"json",
|
||||
"\n",
|
||||
"{",
|
||||
'"answer": "This is a simple ',
|
||||
"answer.",
|
||||
'",\n"',
|
||||
'quotes": ["Document"]',
|
||||
"\n}",
|
||||
"\n",
|
||||
"```",
|
||||
]
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "This is a simple answer." == answer
|
||||
assert len(quotes) == 1
|
||||
assert quotes[0] == "Document"
|
||||
|
||||
|
||||
def test_json_with_lengthy_prefix_and_quotes() -> None:
|
||||
tokens = [
|
||||
"*** Based on the provided documents, there does not appear to be any information ",
|
||||
"directly relevant to answering which documents are my favorite. ",
|
||||
"The documents seem to be focused on describing the Danswer product ",
|
||||
"and its features/use cases. Since I do not have personal preferences ",
|
||||
"for documents, I will provide a general response:\n\n",
|
||||
"```",
|
||||
"json",
|
||||
"\n",
|
||||
"{",
|
||||
'"answer": "This is a simple ',
|
||||
"answer.",
|
||||
'",\n"',
|
||||
'quotes": ["Document"]',
|
||||
"\n}",
|
||||
"\n",
|
||||
"```",
|
||||
]
|
||||
processor = QuotesProcessor(context_docs=mock_docs)
|
||||
answer, quotes = _process_tokens(processor, tokens)
|
||||
|
||||
assert "This is a simple answer." == answer
|
||||
assert len(quotes) == 1
|
||||
assert quotes[0] == "Document"
|
||||
@ -13,15 +13,12 @@ from langchain_core.messages import ToolCallChunk
|
||||
|
||||
from danswer.chat.models import CitationInfo
|
||||
from danswer.chat.models import DanswerAnswerPiece
|
||||
from danswer.chat.models import DanswerQuote
|
||||
from danswer.chat.models import DanswerQuotes
|
||||
from danswer.chat.models import LlmDoc
|
||||
from danswer.chat.models import StreamStopInfo
|
||||
from danswer.chat.models import StreamStopReason
|
||||
from danswer.llm.answering.answer import Answer
|
||||
from danswer.llm.answering.models import AnswerStyleConfig
|
||||
from danswer.llm.answering.models import PromptConfig
|
||||
from danswer.llm.answering.models import QuotesConfig
|
||||
from danswer.llm.interfaces import LLM
|
||||
from danswer.tools.force import ForceUseTool
|
||||
from danswer.tools.models import ToolCallFinalResult
|
||||
@ -284,90 +281,6 @@ def test_answer_with_search_no_tool_calling(
|
||||
mock_search_tool.run.assert_called_once()
|
||||
|
||||
|
||||
def test_answer_with_search_call_quotes_enabled(
|
||||
answer_instance: Answer,
|
||||
mock_search_results: list[LlmDoc],
|
||||
mock_search_tool: MagicMock,
|
||||
) -> None:
|
||||
answer_instance.tools = [mock_search_tool]
|
||||
answer_instance.force_use_tool = ForceUseTool(
|
||||
force_use=False, tool_name="", args=None
|
||||
)
|
||||
answer_instance.answer_style_config.citation_config = None
|
||||
answer_instance.answer_style_config.quotes_config = QuotesConfig()
|
||||
|
||||
# Set up the LLM mock to return search results and then an answer
|
||||
mock_llm = cast(Mock, answer_instance.llm)
|
||||
|
||||
tool_call_chunk = AIMessageChunk(content="")
|
||||
tool_call_chunk.tool_calls = [
|
||||
ToolCall(
|
||||
id="search",
|
||||
name="search",
|
||||
args=DEFAULT_SEARCH_ARGS,
|
||||
)
|
||||
]
|
||||
tool_call_chunk.tool_call_chunks = [
|
||||
ToolCallChunk(
|
||||
id="search",
|
||||
name="search",
|
||||
args=json.dumps(DEFAULT_SEARCH_ARGS),
|
||||
index=0,
|
||||
)
|
||||
]
|
||||
|
||||
# needs to be short due to the "anti-hallucination" check in QuotesProcessor
|
||||
answer_content = "z"
|
||||
quote_content = mock_search_results[0].content
|
||||
mock_llm.stream.side_effect = [
|
||||
[tool_call_chunk],
|
||||
[
|
||||
AIMessageChunk(
|
||||
content=(
|
||||
'{"answer": "'
|
||||
+ answer_content
|
||||
+ '", "quotes": ["'
|
||||
+ quote_content
|
||||
+ '"]}'
|
||||
)
|
||||
),
|
||||
],
|
||||
]
|
||||
|
||||
# Process the output
|
||||
output = list(answer_instance.processed_streamed_output)
|
||||
|
||||
# Assertions
|
||||
assert len(output) == 5
|
||||
assert output[0] == ToolCallKickoff(
|
||||
tool_name="search", tool_args=DEFAULT_SEARCH_ARGS
|
||||
)
|
||||
assert output[1] == ToolResponse(
|
||||
id="final_context_documents",
|
||||
response=mock_search_results,
|
||||
)
|
||||
assert output[2] == ToolCallFinalResult(
|
||||
tool_name="search",
|
||||
tool_args=DEFAULT_SEARCH_ARGS,
|
||||
tool_result=[json.loads(doc.model_dump_json()) for doc in mock_search_results],
|
||||
)
|
||||
assert output[3] == DanswerAnswerPiece(answer_piece=answer_content)
|
||||
assert output[4] == DanswerQuotes(
|
||||
quotes=[
|
||||
DanswerQuote(
|
||||
quote=quote_content,
|
||||
document_id=mock_search_results[0].document_id,
|
||||
link=mock_search_results[0].link,
|
||||
source_type=mock_search_results[0].source_type,
|
||||
semantic_identifier=mock_search_results[0].semantic_identifier,
|
||||
blurb=mock_search_results[0].blurb,
|
||||
)
|
||||
]
|
||||
)
|
||||
|
||||
assert answer_instance.llm_answer == answer_content
|
||||
|
||||
|
||||
def test_is_cancelled(answer_instance: Answer) -> None:
|
||||
# Set up the LLM mock to return multiple chunks
|
||||
mock_llm = Mock()
|
||||
|
||||
@ -6,9 +6,9 @@ import pytest
|
||||
from pytest_mock import MockerFixture
|
||||
|
||||
from danswer.llm.answering.answer import Answer
|
||||
from danswer.llm.answering.answer import AnswerStream
|
||||
from danswer.llm.answering.models import AnswerStyleConfig
|
||||
from danswer.llm.answering.models import PromptConfig
|
||||
from danswer.one_shot_answer.answer_question import AnswerObjectIterator
|
||||
from danswer.tools.force import ForceUseTool
|
||||
from danswer.tools.tool_implementations.search.search_tool import SearchTool
|
||||
from tests.regression.answer_quality.run_qa import _process_and_write_query_results
|
||||
@ -60,7 +60,7 @@ def test_skip_gen_ai_answer_generation_flag(
|
||||
skip_gen_ai_answer_generation=skip_gen_ai_answer_generation,
|
||||
)
|
||||
count = 0
|
||||
for _ in cast(AnswerObjectIterator, answer.processed_streamed_output):
|
||||
for _ in cast(AnswerStream, answer.processed_streamed_output):
|
||||
count += 1
|
||||
assert count == 3 if skip_gen_ai_answer_generation else 4
|
||||
if not skip_gen_ai_answer_generation:
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user