highperfocused/danswer
1
0
Fork 0
mirror of https://github.com/danswer-ai/danswer.git synced 2025-10-10 13:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

rename of documents to verified_reranked_documents

Browse Source
This commit is contained in:
joachim-danswer
2025-01-31 12:16:17 -08:00
committed by Evan Lohn
parent d53dd1e356
commit 732861a940
19 changed files with 44 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
backend/onyx/agents/agent_search/deep_search_a/initial/general_sub_answers/nodes/format_initial_sub_answers.py
View File

@@ -20,12 +20,12 @@ def format_initial_sub_answers(
logger.info(f"--------{now_start}--------INGEST ANSWERS---") logger.info(f"--------{now_start}--------INGEST ANSWERS---")
documents = [] documents = []
context_documents = [] context_documents = []
cited_docs = [] cited_documents = []
answer_results = state.answer_results if hasattr(state, "answer_results") else [] answer_results = state.answer_results if hasattr(state, "answer_results") else []
for answer_result in answer_results: for answer_result in answer_results:
documents.extend(answer_result.documents) documents.extend(answer_result.verified_reranked_documents)
context_documents.extend(answer_result.context_documents) context_documents.extend(answer_result.context_documents)
cited_docs.extend(answer_result.cited_docs) cited_documents.extend(answer_result.cited_documents)
now_end = datetime.now() now_end = datetime.now()
logger.debug( logger.debug(
@@ -35,9 +35,9 @@ def format_initial_sub_answers(
return DecompAnswersUpdate( return DecompAnswersUpdate(
# Deduping is done by the documents operator for the main graph # Deduping is done by the documents operator for the main graph
# so we might not need to dedup here # so we might not need to dedup here
documents=dedup_inference_sections(documents, []), verified_reranked_documents=dedup_inference_sections(documents, []),
context_documents=dedup_inference_sections(context_documents, []), context_documents=dedup_inference_sections(context_documents, []),
cited_documents=dedup_inference_sections(cited_docs, []), cited_documents=dedup_inference_sections(cited_documents, []),
sub_question_results=answer_results, sub_question_results=answer_results,
log_messages=[ log_messages=[
f"{now_start} -- Main - Ingest initial processed sub questions, Time taken: {now_end - now_start}" f"{now_start} -- Main - Ingest initial processed sub questions, Time taken: {now_end - now_start}"

4
backend/onyx/agents/agent_search/deep_search_a/initial/generate_individual_sub_answer/nodes/format_sub_answer.py
View File

@@ -18,9 +18,9 @@ def format_sub_answer(state: AnswerQuestionState) -> AnswerQuestionOutput:
verified_high_quality=state.answer_quality, verified_high_quality=state.answer_quality,
answer=state.answer, answer=state.answer,
expanded_retrieval_results=state.expanded_retrieval_results, expanded_retrieval_results=state.expanded_retrieval_results,
documents=state.documents, verified_reranked_documents=state.verified_reranked_documents,
context_documents=state.context_documents, context_documents=state.context_documents,
cited_docs=state.cited_docs, cited_documents=state.cited_documents,
sub_question_retrieval_stats=state.sub_question_retrieval_stats, sub_question_retrieval_stats=state.sub_question_retrieval_stats,
) )
], ],

6
backend/onyx/agents/agent_search/deep_search_a/initial/generate_individual_sub_answer/nodes/generate_sub_answer.py
View File

@@ -44,7 +44,7 @@ def generate_sub_answer(
agent_search_config = cast(AgentSearchConfig, config["metadata"]["config"]) agent_search_config = cast(AgentSearchConfig, config["metadata"]["config"])
question = state.question question = state.question
state.documents state.verified_reranked_documents
level, question_nr = parse_question_id(state.question_id) level, question_nr = parse_question_id(state.question_id)
context_docs = state.context_documents[:AGENT_MAX_ANSWER_CONTEXT_DOCS] context_docs = state.context_documents[:AGENT_MAX_ANSWER_CONTEXT_DOCS]
persona_contextualized_prompt = get_persona_agent_prompt_expressions( persona_contextualized_prompt = get_persona_agent_prompt_expressions(
@@ -107,7 +107,7 @@ def generate_sub_answer(
) )
answer_citation_ids = get_answer_citation_ids(answer_str) answer_citation_ids = get_answer_citation_ids(answer_str)
cited_docs = [ cited_documents = [
context_docs[id] for id in answer_citation_ids if id < len(context_docs) context_docs[id] for id in answer_citation_ids if id < len(context_docs)
] ]
@@ -121,7 +121,7 @@ def generate_sub_answer(
return QAGenerationUpdate( return QAGenerationUpdate(
answer=answer_str, answer=answer_str,
cited_docs=cited_docs, cited_documents=cited_documents,
log_messages=[ log_messages=[
get_langgraph_node_log_string( get_langgraph_node_log_string(
graph_component="initial - generate individual sub answer", graph_component="initial - generate individual sub answer",

2
backend/onyx/agents/agent_search/deep_search_a/initial/generate_individual_sub_answer/nodes/ingest_retrieved_documents.py
View File

@@ -18,7 +18,7 @@ def ingest_retrieved_documents(
return RetrievalIngestionUpdate( return RetrievalIngestionUpdate(
expanded_retrieval_results=state.expanded_retrieval_result.expanded_queries_results, expanded_retrieval_results=state.expanded_retrieval_result.expanded_queries_results,
documents=state.expanded_retrieval_result.reranked_documents, verified_reranked_documents=state.expanded_retrieval_result.verified_reranked_documents,
context_documents=state.expanded_retrieval_result.context_documents, context_documents=state.expanded_retrieval_result.context_documents,
sub_question_retrieval_stats=sub_question_retrieval_stats, sub_question_retrieval_stats=sub_question_retrieval_stats,
) )

6
backend/onyx/agents/agent_search/deep_search_a/initial/generate_individual_sub_answer/states.py
View File

@@ -25,13 +25,15 @@ class QACheckUpdate(LoggerUpdate, BaseModel):
class QAGenerationUpdate(LoggerUpdate, BaseModel): class QAGenerationUpdate(LoggerUpdate, BaseModel):
answer: str = "" answer: str = ""
log_messages: list[str] = [] log_messages: list[str] = []
cited_docs: Annotated[list[InferenceSection], dedup_inference_sections] = [] cited_documents: Annotated[list[InferenceSection], dedup_inference_sections] = []
# answer_stat: AnswerStats # answer_stat: AnswerStats
class RetrievalIngestionUpdate(LoggerUpdate, BaseModel): class RetrievalIngestionUpdate(LoggerUpdate, BaseModel):
expanded_retrieval_results: list[QueryResult] = [] expanded_retrieval_results: list[QueryResult] = []
documents: Annotated[list[InferenceSection], dedup_inference_sections] = [] verified_reranked_documents: Annotated[
list[InferenceSection], dedup_inference_sections
] = []
context_documents: Annotated[list[InferenceSection], dedup_inference_sections] = [] context_documents: Annotated[list[InferenceSection], dedup_inference_sections] = []
sub_question_retrieval_stats: AgentChunkStats = AgentChunkStats() sub_question_retrieval_stats: AgentChunkStats = AgentChunkStats()

6
backend/onyx/agents/agent_search/deep_search_a/initial/generate_initial_answer/nodes/generate_initial_answer.py
View File

@@ -67,13 +67,13 @@ def generate_initial_answer(
question = agent_a_config.search_request.query question = agent_a_config.search_request.query
prompt_enrichment_components = get_prompt_enrichment_components(agent_a_config) prompt_enrichment_components = get_prompt_enrichment_components(agent_a_config)
sub_questions_cited_docs = state.cited_documents sub_questions_cited_documents = state.cited_documents
all_original_question_documents = state.all_original_question_documents all_original_question_documents = state.all_original_question_documents
consolidated_context_docs: list[InferenceSection] = sub_questions_cited_docs consolidated_context_docs: list[InferenceSection] = sub_questions_cited_documents
counter = 0 counter = 0
for original_doc_number, original_doc in enumerate(all_original_question_documents): for original_doc_number, original_doc in enumerate(all_original_question_documents):
if original_doc_number not in sub_questions_cited_docs: if original_doc_number not in sub_questions_cited_documents:
if ( if (
counter <= AGENT_MIN_ORIG_QUESTION_DOCS counter <= AGENT_MIN_ORIG_QUESTION_DOCS
or len(consolidated_context_docs) < AGENT_MAX_ANSWER_CONTEXT_DOCS or len(consolidated_context_docs) < AGENT_MAX_ANSWER_CONTEXT_DOCS

10
backend/onyx/agents/agent_search/deep_search_a/initial/generate_sub_answers/nodes/format_initial_sub_answers.py
View File

@@ -21,19 +21,19 @@ def format_initial_sub_answers(
documents = [] documents = []
context_documents = [] context_documents = []
cited_docs = [] cited_documents = []
answer_results = state.answer_results if hasattr(state, "answer_results") else [] answer_results = state.answer_results if hasattr(state, "answer_results") else []
for answer_result in answer_results: for answer_result in answer_results:
documents.extend(answer_result.documents) documents.extend(answer_result.verified_reranked_documents)
context_documents.extend(answer_result.context_documents) context_documents.extend(answer_result.context_documents)
cited_docs.extend(answer_result.cited_docs) cited_documents.extend(answer_result.cited_documents)
return DecompAnswersUpdate( return DecompAnswersUpdate(
# Deduping is done by the documents operator for the main graph # Deduping is done by the documents operator for the main graph
# so we might not need to dedup here # so we might not need to dedup here
documents=dedup_inference_sections(documents, []), verified_reranked_documents=dedup_inference_sections(documents, []),
context_documents=dedup_inference_sections(context_documents, []), context_documents=dedup_inference_sections(context_documents, []),
cited_documents=dedup_inference_sections(cited_docs, []), cited_documents=dedup_inference_sections(cited_documents, []),
sub_question_results=answer_results, sub_question_results=answer_results,
log_messages=[ log_messages=[
get_langgraph_node_log_string( get_langgraph_node_log_string(

12
backend/onyx/agents/agent_search/deep_search_a/initial/retrieve_orig_question_docs/models.py
View File

@@ -1,20 +1,8 @@
from pydantic import BaseModel from pydantic import BaseModel
from onyx.agents.agent_search.shared_graph_utils.models import AgentChunkStats
from onyx.agents.agent_search.shared_graph_utils.models import QueryResult
from onyx.context.search.models import InferenceSection
### Models ### ### Models ###
class AnswerRetrievalStats(BaseModel): class AnswerRetrievalStats(BaseModel):
answer_retrieval_stats: dict[str, float | int] answer_retrieval_stats: dict[str, float | int]
class QuestionAnswerResults(BaseModel):
question: str
answer: str
quality: str
expanded_retrieval_results: list[QueryResult]
documents: list[InferenceSection]
sub_question_retrieval_stats: list[AgentChunkStats]

8
backend/onyx/agents/agent_search/deep_search_a/main/nodes/generate_refined_answer.py
View File

@@ -69,16 +69,16 @@ def generate_refined_answer(
prompt_enrichment_components.persona_prompts.contextualized_prompt prompt_enrichment_components.persona_prompts.contextualized_prompt
) )
initial_documents = state.documents initial_documents = state.verified_reranked_documents
refined_documents = state.refined_documents refined_documents = state.refined_documents
sub_questions_cited_docs = state.cited_documents sub_questions_cited_documents = state.cited_documents
all_original_question_documents = state.all_original_question_documents all_original_question_documents = state.all_original_question_documents
consolidated_context_docs: list[InferenceSection] = sub_questions_cited_docs consolidated_context_docs: list[InferenceSection] = sub_questions_cited_documents
counter = 0 counter = 0
for original_doc_number, original_doc in enumerate(all_original_question_documents): for original_doc_number, original_doc in enumerate(all_original_question_documents):
if original_doc_number not in sub_questions_cited_docs: if original_doc_number not in sub_questions_cited_documents:
if ( if (
counter <= AGENT_MIN_ORIG_QUESTION_DOCS counter <= AGENT_MIN_ORIG_QUESTION_DOCS
or len(consolidated_context_docs) or len(consolidated_context_docs)

4
backend/onyx/agents/agent_search/deep_search_a/main/nodes/ingest_refined_answers.py
View File

@@ -22,12 +22,12 @@ def ingest_refined_answers(
documents = [] documents = []
answer_results = state.answer_results if hasattr(state, "answer_results") else [] answer_results = state.answer_results if hasattr(state, "answer_results") else []
for answer_result in answer_results: for answer_result in answer_results:
documents.extend(answer_result.documents) documents.extend(answer_result.verified_reranked_documents)
return DecompAnswersUpdate( return DecompAnswersUpdate(
# Deduping is done by the documents operator for the main graph # Deduping is done by the documents operator for the main graph
# so we might not need to dedup here # so we might not need to dedup here
documents=dedup_inference_sections(documents, []), verified_reranked_documents=dedup_inference_sections(documents, []),
sub_question_results=answer_results, sub_question_results=answer_results,
log_messages=[ log_messages=[
get_langgraph_node_log_string( get_langgraph_node_log_string(

4
backend/onyx/agents/agent_search/deep_search_a/main/states.py
View File

@@ -102,7 +102,9 @@ class RequireRefinedAnswerUpdate(LoggerUpdate):
class DecompAnswersUpdate(LoggerUpdate): class DecompAnswersUpdate(LoggerUpdate):
documents: Annotated[list[InferenceSection], dedup_inference_sections] = [] verified_reranked_documents: Annotated[
list[InferenceSection], dedup_inference_sections
] = []
context_documents: Annotated[list[InferenceSection], dedup_inference_sections] = [] context_documents: Annotated[list[InferenceSection], dedup_inference_sections] = []
cited_documents: Annotated[ cited_documents: Annotated[
list[InferenceSection], dedup_inference_sections list[InferenceSection], dedup_inference_sections

10
backend/onyx/agents/agent_search/deep_search_a/refinement/consolidate_sub_answers/models.py
View File

@@ -1,18 +1,8 @@
from pydantic import BaseModel from pydantic import BaseModel
from onyx.agents.agent_search.shared_graph_utils.models import AgentChunkStats
from onyx.context.search.models import InferenceSection
### Models ### ### Models ###
class AnswerRetrievalStats(BaseModel): class AnswerRetrievalStats(BaseModel):
answer_retrieval_stats: dict[str, float | int] answer_retrieval_stats: dict[str, float | int]
class QuestionAnswerResults(BaseModel):
question: str
answer: str
quality: str
documents: list[InferenceSection]
sub_question_retrieval_stats: AgentChunkStats

2
backend/onyx/agents/agent_search/deep_search_a/shared/expanded_retrieval/models.py
View File

@@ -7,6 +7,6 @@ from onyx.context.search.models import InferenceSection
class ExpandedRetrievalResult(BaseModel): class ExpandedRetrievalResult(BaseModel):
expanded_queries_results: list[QueryResult] = [] expanded_queries_results: list[QueryResult] = []
reranked_documents: list[InferenceSection] = [] verified_reranked_documents: list[InferenceSection] = []
context_documents: list[InferenceSection] = [] context_documents: list[InferenceSection] = []
sub_question_retrieval_stats: AgentChunkStats = AgentChunkStats() sub_question_retrieval_stats: AgentChunkStats = AgentChunkStats()

2
backend/onyx/agents/agent_search/deep_search_a/shared/expanded_retrieval/nodes/format_results.py
View File

@@ -79,7 +79,7 @@ def format_results(
return ExpandedRetrievalUpdate( return ExpandedRetrievalUpdate(
expanded_retrieval_result=ExpandedRetrievalResult( expanded_retrieval_result=ExpandedRetrievalResult(
expanded_queries_results=state.expanded_retrieval_results, expanded_queries_results=state.expanded_retrieval_results,
reranked_documents=reranked_documents, verified_reranked_documents=reranked_documents,
context_documents=state.reranked_documents, context_documents=state.reranked_documents,
sub_question_retrieval_stats=sub_question_retrieval_stats, sub_question_retrieval_stats=sub_question_retrieval_stats,
), ),

6
backend/onyx/agents/agent_search/deep_search_a/shared/expanded_retrieval/nodes/kickoff_verification.py
View File

@@ -16,7 +16,7 @@ def kickoff_verification(
state: ExpandedRetrievalState, state: ExpandedRetrievalState,
config: RunnableConfig, config: RunnableConfig,
) -> Command[Literal["verify_documents"]]: ) -> Command[Literal["verify_documents"]]:
documents = state.retrieved_documents retrieved_documents = state.retrieved_documents
verification_question = state.question verification_question = state.question
sub_question_id = state.sub_question_id sub_question_id = state.sub_question_id
@@ -26,13 +26,13 @@ def kickoff_verification(
Send( Send(
node="verify_documents", node="verify_documents",
arg=DocVerificationInput( arg=DocVerificationInput(
doc_to_verify=doc, retrieved_document_to_verify=document,
question=verification_question, question=verification_question,
base_search=False, base_search=False,
sub_question_id=sub_question_id, sub_question_id=sub_question_id,
log_messages=[], log_messages=[],
), ),
) )
for doc in documents for document in retrieved_documents
], ],
) )

6
backend/onyx/agents/agent_search/deep_search_a/shared/expanded_retrieval/nodes/verify_documents.py
View File

@@ -31,8 +31,8 @@ def verify_documents(
""" """
question = state.question question = state.question
doc_to_verify = state.doc_to_verify retrieved_document_to_verify = state.retrieved_document_to_verify
document_content = doc_to_verify.combined_content document_content = retrieved_document_to_verify.combined_content
agent_a_config = cast(AgentSearchConfig, config["metadata"]["config"]) agent_a_config = cast(AgentSearchConfig, config["metadata"]["config"])
fast_llm = agent_a_config.fast_llm fast_llm = agent_a_config.fast_llm
@@ -53,7 +53,7 @@ def verify_documents(
verified_documents = [] verified_documents = []
if isinstance(response.content, str) and "yes" in response.content.lower(): if isinstance(response.content, str) and "yes" in response.content.lower():
verified_documents.append(doc_to_verify) verified_documents.append(retrieved_document_to_verify)
return DocVerificationUpdate( return DocVerificationUpdate(
verified_documents=verified_documents, verified_documents=verified_documents,

2
backend/onyx/agents/agent_search/deep_search_a/shared/expanded_retrieval/states.py
View File

@@ -81,7 +81,7 @@ class ExpandedRetrievalState(
class DocVerificationInput(ExpandedRetrievalInput): class DocVerificationInput(ExpandedRetrievalInput):
doc_to_verify: InferenceSection retrieved_document_to_verify: InferenceSection
class RetrievalInput(ExpandedRetrievalInput): class RetrievalInput(ExpandedRetrievalInput):

4
backend/onyx/agents/agent_search/shared_graph_utils/models.py
View File

@@ -105,9 +105,9 @@ class QuestionAnswerResults(BaseModel):
answer: str answer: str
verified_high_quality: bool verified_high_quality: bool
expanded_retrieval_results: list[QueryResult] expanded_retrieval_results: list[QueryResult]
documents: list[InferenceSection] verified_reranked_documents: list[InferenceSection]
context_documents: list[InferenceSection] context_documents: list[InferenceSection]
cited_docs: list[InferenceSection] cited_documents: list[InferenceSection]
sub_question_retrieval_stats: AgentChunkStats sub_question_retrieval_stats: AgentChunkStats

2
backend/onyx/db/chat.py
View File

@@ -1018,7 +1018,7 @@ def log_agent_sub_question_results(
sub_question = sub_question_answer_result.question sub_question = sub_question_answer_result.question
sub_answer = sub_question_answer_result.answer sub_answer = sub_question_answer_result.answer
sub_document_results = _create_citation_format_list( sub_document_results = _create_citation_format_list(
sub_question_answer_result.documents sub_question_answer_result.verified_reranked_documents
) )
sub_question_object = AgentSubQuestion( sub_question_object = AgentSubQuestion(