sber/agent
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ййй

Browse Source
This commit is contained in:
zosimovaa
2026-03-27 15:51:10 +03:00
parent 15586f9a8c
commit 51378c5d66
1234 changed files with 95644 additions and 543076 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/pipeline_setup_v3/cases/suite_01_synthetic/code_qa_eval.yaml
+1 -1
View File
@@ -2,7 +2,7 @@ defaults:
runner: code_qa_eval
mode: router_rag
input:
repo_path: "../../../pipeline_setup/suite_01_synthetic/fixtures/code_qa_repo"
repo_path: "../../fixtures/code_qa_repo"
project_id: code_qa_repo
cases:
tests/pipeline_setup_v3/cases/suite_02_pipeline/full_chain_smoke.yaml
+1 -1
View File
@@ -2,7 +2,7 @@ defaults:
runner: runtime
mode: full_chain
input:
repo_path: "../../../pipeline_setup/suite_01_synthetic/fixtures/code_qa_repo"
repo_path: "../../fixtures/code_qa_repo"
project_id: code_qa_repo_full_chain
cases:
tests/pipeline_setup_v3/cases/suite_02_pipeline/router_rag_smoke.yaml
+1 -1
View File
@@ -2,7 +2,7 @@ defaults:
runner: runtime
mode: router_rag
input:
repo_path: "../../../pipeline_setup/suite_01_synthetic/fixtures/code_qa_repo"
repo_path: "../../fixtures/code_qa_repo"
project_id: code_qa_repo_runtime
cases:
tests/pipeline_setup_v3/cases/suite_03_docs/full_chain_docs_intents_v1.yaml
+275
View File
@@ -0,0 +1,275 @@
defaults:
runner: runtime
mode: full_chain
input:
rag_session_id: "fb7e0690-e737-44ca-acc6-e38a0687a5ee"
cases:
# DOCUMENTATION_EXPLAIN
- id: docs-v1-explain-system-flow-telegram-loop
query: "Объясни как работает цикл отправки уведомлений в Telegram"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["logic.telegram_notification_loop"]
layers_include: ["D1_DOCUMENT_CATALOG", "D4_WORKFLOW_INDEX", "D5_RELATION_GRAPH"]
llm:
non_empty: true
contains_all: ["TelegramNotifyWorker", "интервал"]
answer_mode: answered
- id: docs-v1-explain-component-control-channel
query: "Объясни как устроен компонент TelegramControlChannel"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: COMPONENT_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["TelegramControlChannel"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D5_RELATION_GRAPH"]
llm:
non_empty: true
contains_all: ["control plane", "/health"]
answer_mode: answered
- id: docs-v1-explain-api-health-endpoint
query: "Объясни что делает API метод /health"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: API_METHOD_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D4_WORKFLOW_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_all: ["HTTP 200", "HTTP 503"]
answer_mode: answered
- id: docs-v1-explain-entity-runtime-health
query: "Что такое сущность runtime health?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: ENTITY_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["WorkerHealth", "WorkerStatus"]
doc_scope_contains: ["domain.runtime_health"]
layers_include: ["D1_DOCUMENT_CATALOG", "D3_ENTITY_CATALOG", "D5_RELATION_GRAPH"]
llm:
non_empty: true
contains_all: ["наблюдаемости", "telegram_notify"]
answer_mode: answered
# DISCOVERY folded into DOCUMENTATION_EXPLAIN
- id: docs-v1-discovery-find-documents-telegram-api
query: "Найди документацию по Telegram API и control plane"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX"]
llm:
non_empty: true
contains_all: ["/health", "/send"]
answer_mode: answered
- id: docs-v1-discovery-find-by-entity-telegramnotifyworker
query: "Какие документы связаны с сущностью TelegramNotifyWorker?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["TelegramNotifyWorker"]
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG"]
llm:
non_empty: true
contains_all: ["logic.telegram_notification_loop", "api.health_endpoint"]
answer_mode: answered
- id: docs-v1-discovery-find-by-topic-health
query: "Где в документации описан health runtime?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG"]
llm:
non_empty: true
contains_all: ["api.health_endpoint", "domain.runtime_health"]
answer_mode: answered
# NAVIGATION folded into DOCUMENTATION_EXPLAIN
- id: docs-v1-navigation-related-docs-health-endpoint
query: "Какие документы связаны с endpoint /health?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG"]
llm:
non_empty: true
contains_all: ["architecture.telegram_notify_app", "domain.runtime_health"]
answer_mode: answered
- id: docs-v1-navigation-parent-child-architecture
query: "Покажи навигацию от overview архитектуры к дочерним документам"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["architecture.telegram_notify_app"]
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG"]
llm:
non_empty: true
contains_all: ["api.health_endpoint", "logic.telegram_notification_loop"]
answer_mode: answered
- id: docs-v1-navigation-workflow-send-message
query: "По каким документам идти, чтобы понять сценарий ручной отправки сообщения?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG"]
llm:
non_empty: true
contains_all: ["api.send_message_endpoint", "logic.telegram_notification_loop"]
answer_mode: answered
# OPENAPI_GENERATION
- id: docs-v1-openapi-method-health-endpoint
query: "Сгенерируй OpenAPI для метода /health по документации"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_METHOD_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
filters_contain:
doc_type: api_method
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_all: ["paths:", "/health", "get:"]
answer_mode: structured_spec
- id: docs-v1-openapi-fragment-send-request
query: "Сгенерируй request schema для endpoint /send по документации"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_FRAGMENT_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
filters_contain:
doc_type: api_method
doc_scope_contains: ["api.send_message_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_all: ["message", "type:", "string"]
answer_mode: structured_spec
- id: docs-v1-openapi-fragment-control-actions-response
query: "Собери response schema для /actions/{action} по документации"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_FRAGMENT_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
filters_contain:
doc_type: api_method
doc_scope_contains: ["api.control_actions_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_all: ["status", "detail"]
answer_mode: structured_spec
# NEGATIVE / NOT FOUND
- id: docs-v1-discovery-find-by-entity-nonexistent
query: "Какие документы связаны с сущностью PaymentTransaction?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: false
entity_candidates_contain: ["PaymentTransaction"]
layers_include: ["D3_ENTITY_CATALOG", "D1_DOCUMENT_CATALOG", "D5_RELATION_GRAPH"]
llm:
non_empty: true
contains_all: ["не найдена"]
excludes: ["вероятно", "исходя из названия"]
pipeline:
answer_mode: degraded
- id: docs-v1-openapi-method-nonexistent
query: "Сгенерируй OpenAPI для endpoint /orders по документации"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_METHOD_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: false
filters_contain:
doc_type: api_method
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_all: ["не найден"]
excludes: ["paths:", "/orders", "post:"]
pipeline:
answer_mode: degraded
tests/pipeline_setup_v3/cases/suite_03_docs/full_chain_docs_intents_v2.yaml
+271
View File
@@ -0,0 +1,271 @@
defaults:
runner: runtime
mode: full_chain
input:
rag_session_id: "fb7e0690-e737-44ca-acc6-e38a0687a5ee"
cases:
# ============================================================
# DOCUMENTATION_EXPLAIN / SYSTEM_FLOW_EXPLAIN
# ============================================================
- id: docs-v2-system-flow-telegram-loop
query: "Объясни как работает цикл отправки уведомлений в Telegram"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["logic.telegram_notification_loop"]
layers_include: ["D1_DOCUMENT_CATALOG", "D4_WORKFLOW_INDEX", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["TelegramNotifyWorker", "интервал", "degraded"]
answer_mode: answered
- id: docs-v2-system-flow-health-check
query: "Как происходит получение health состояния runtime?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint", "domain.runtime_health"]
layers_include: ["D1_DOCUMENT_CATALOG", "D4_WORKFLOW_INDEX", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["/health", "HTTP 200", "HTTP 503", "payload"]
answer_mode: answered
# ============================================================
# DOCUMENTATION_EXPLAIN / COMPONENT_EXPLAIN
# ============================================================
- id: docs-v2-component-control-channel
query: "Объясни как работает компонент TelegramControlChannel"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: COMPONENT_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["TelegramControlChannel"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["/health", "/actions/{action}", "/send"]
answer_mode: answered
- id: docs-v2-component-runtime-manager
query: "Какую роль в системе играет RuntimeManager?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: COMPONENT_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["RuntimeManager"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["runtime", "регистр", "TelegramNotifyModule"]
answer_mode: answered
# ============================================================
# DOCUMENTATION_EXPLAIN / API_METHOD_EXPLAIN
# ============================================================
- id: docs-v2-api-health-endpoint
query: "Объясни что делает endpoint /health"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: API_METHOD_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D4_WORKFLOW_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["HTTP 200", "HTTP 503", "payload", "components"]
answer_mode: answered
- id: docs-v2-api-send-endpoint
query: "Как работает endpoint /send?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: API_METHOD_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.send_message_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D4_WORKFLOW_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["message", "HTTP 400", "HTTP 500", "TelegramSendService"]
answer_mode: answered
# ============================================================
# DOCUMENTATION_EXPLAIN / ENTITY_EXPLAIN
# ============================================================
- id: docs-v2-entity-runtime-health
query: "Что такое runtime health?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: ENTITY_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["domain.runtime_health"]
layers_include: ["D3_ENTITY_CATALOG", "D1_DOCUMENT_CATALOG", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["наблюдаемости", "WorkerHealth", "WorkerStatus", "telegram_notify"]
answer_mode: answered
- id: docs-v2-entity-worker-status
query: "Объясни сущность WorkerStatus"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: ENTITY_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
entity_candidates_contain: ["WorkerStatus"]
layers_include: ["D3_ENTITY_CATALOG", "D1_DOCUMENT_CATALOG", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["ok", "degraded", "unhealthy"]
answer_mode: answered
# ============================================================
# DOCUMENTATION_EXPLAIN / RELATED_DOCS_EXPLAIN
# ============================================================
- id: docs-v2-related-docs-health
query: "Какие документы связаны с endpoint /health?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["domain.runtime_health", "architecture.telegram_notify_app"]
answer_mode: answered
- id: docs-v2-related-docs-architecture-overview
query: "Какие дочерние документы есть у overview архитектуры сервиса?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: RELATED_DOCS_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["architecture.telegram_notify_app"]
layers_include: ["D5_RELATION_GRAPH", "D1_DOCUMENT_CATALOG", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["api.health_endpoint", "api.send_message_endpoint", "logic.telegram_notification_loop"]
answer_mode: answered
# ============================================================
# OPENAPI_GENERATION / OPENAPI_METHOD_GENERATE
# ============================================================
- id: docs-v2-openapi-health
query: "Сгенерируй OpenAPI для endpoint /health по документации"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_METHOD_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.health_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["paths:", "/health", "get:"]
answer_mode: structured_spec
# ============================================================
# OPENAPI_GENERATION / OPENAPI_FRAGMENT_GENERATE
# ============================================================
- id: docs-v2-openapi-fragment-send-request
query: "Сгенерируй request schema для endpoint /send"
expected:
router:
intent: OPENAPI_GENERATION
sub_intent: OPENAPI_FRAGMENT_GENERATE
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
doc_scope_contains: ["api.send_message_endpoint"]
layers_include: ["D1_DOCUMENT_CATALOG", "D2_FACT_INDEX", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["message", "string"]
answer_mode: structured_spec
# ============================================================
# GENERAL_QA / GENERIC_QA
# ============================================================
- id: docs-v2-general-overview
query: "Что вообще описано в документации по этому сервису?"
expected:
router:
intent: GENERAL_QA
sub_intent: GENERIC_QA
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: true
layers_include: ["D1_DOCUMENT_CATALOG", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["архитектура", "API", "worker", "runtime"]
answer_mode: answered
# ============================================================
# NEGATIVE / DEGRADED
# ============================================================
- id: docs-v2-negative-missing-entity
query: "Что такое сущность PaymentTransaction?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: ENTITY_EXPLAIN
graph_id: DocsQAGraph
conversation_mode: START
retrieval:
non_empty: false
entity_candidates_contain: ["PaymentTransaction"]
layers_include: ["D3_ENTITY_CATALOG", "D1_DOCUMENT_CATALOG", "D5_RELATION_GRAPH", "D0_DOC_CHUNKS"]
llm:
non_empty: true
contains_any: ["не найдена", "нет в документации"]
pipeline:
answer_mode: degraded
tests/pipeline_setup_v3/cases/suite_03_docs/full_chain_docs_intents_v3-3x.yaml
+220
View File
@@ -0,0 +1,220 @@
# full_chain_docs_intents_v3.yaml
defaults:
runner: runtime
mode: pre_llm
input:
rag_session_id: "fb7e0690-e737-44ca-acc6-e38a0687a5ee"
cases:
# ============================================================
# SYSTEM FLOW (расширено)
# ============================================================
- id: v3-system-flow-telegram-loop-1
query: "Объясни цикл отправки уведомлений"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
- id: v3-system-flow-telegram-loop-2
query: "Как работает процесс отправки уведомлений в Telegram?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
- id: v3-system-flow-telegram-loop-3
query: "Опиши workflow Telegram уведомлений"
expected:
router:
intent: DOCUMENTATION_EXPLAIN
sub_intent: SYSTEM_FLOW_EXPLAIN
- id: v3-system-flow-health-1
query: "Как работает health check runtime?"
expected:
router:
sub_intent: SYSTEM_FLOW_EXPLAIN
- id: v3-system-flow-health-2
query: "Как происходит проверка состояния runtime?"
expected:
router:
sub_intent: SYSTEM_FLOW_EXPLAIN
# ============================================================
# COMPONENT (расширено)
# ============================================================
- id: v3-component-runtime-manager-1
query: "Что делает RuntimeManager?"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
- id: v3-component-runtime-manager-2
query: "Как устроен RuntimeManager?"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
- id: v3-component-runtime-manager-3
query: "Опиши компонент RuntimeManager"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
- id: v3-component-control-channel-1
query: "Как работает TelegramControlChannel?"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
- id: v3-component-control-channel-2
query: "Роль TelegramControlChannel в системе"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
# ============================================================
# API (расширено)
# ============================================================
- id: v3-api-health-1
query: "Что делает /health?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
- id: v3-api-health-2
query: "Как работает health endpoint?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
- id: v3-api-health-3
query: "Опиши endpoint health"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
- id: v3-api-send-1
query: "Что делает /send?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
- id: v3-api-send-2
query: "Как работает send endpoint?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
- id: v3-api-send-3
query: "Опиши endpoint отправки сообщений"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
# ============================================================
# ENTITY (расширено)
# ============================================================
- id: v3-entity-runtime-health-1
query: "Что такое runtime health?"
expected:
router:
sub_intent: ENTITY_EXPLAIN
- id: v3-entity-runtime-health-2
query: "Объясни runtime health"
expected:
router:
sub_intent: ENTITY_EXPLAIN
- id: v3-entity-worker-status-1
query: "Что такое WorkerStatus?"
expected:
router:
sub_intent: ENTITY_EXPLAIN
- id: v3-entity-worker-status-2
query: "Опиши статус воркера"
expected:
router:
sub_intent: ENTITY_EXPLAIN
# ============================================================
# RELATED DOCS (расширено)
# ============================================================
- id: v3-related-health-1
query: "Что связано с /health?"
expected:
router:
sub_intent: RELATED_DOCS_EXPLAIN
- id: v3-related-health-2
query: "Какие документы относятся к health endpoint?"
expected:
router:
sub_intent: RELATED_DOCS_EXPLAIN
- id: v3-related-send-1
query: "Какие документы связаны с /send?"
expected:
router:
sub_intent: RELATED_DOCS_EXPLAIN
- id: v3-related-architecture-1
query: "Какие документы связаны с архитектурой?"
expected:
router:
sub_intent: RELATED_DOCS_EXPLAIN
# ============================================================
# GENERAL (расширено)
# ============================================================
- id: v3-general-1
query: "Что есть в документации?"
expected:
router:
sub_intent: GENERIC_QA
- id: v3-general-2
query: "Какая структура документации?"
expected:
router:
sub_intent: GENERIC_QA
- id: v3-general-3
query: "С чего начать читать документацию?"
expected:
router:
sub_intent: GENERIC_QA
# ============================================================
# NEGATIVE (расширено)
# ============================================================
- id: v3-negative-1
query: "Что такое PaymentTransaction?"
expected:
router:
sub_intent: ENTITY_EXPLAIN
- id: v3-negative-2
query: "Объясни сущность UnknownEntity"
expected:
router:
sub_intent: ENTITY_EXPLAIN
- id: v3-negative-3
query: "Что делает endpoint /unknown?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
tests/pipeline_setup_v3/cases/suite_03_docs/full_chain_docs_intents_v4-llm.yaml
+79
View File
@@ -0,0 +1,79 @@
# full_chain_docs_intents_v4.yaml
defaults:
runner: runtime
mode: full_chain
input:
rag_session_id: "fb7e0690-e737-44ca-acc6-e38a0687a5ee"
cases:
# SYSTEM FLOW
- id: v4-system-flow
query: "Объясни как работает цикл отправки уведомлений"
expected:
router:
sub_intent: SYSTEM_FLOW_EXPLAIN
# COMPONENT
- id: v4-component
query: "Как работает RuntimeManager?"
expected:
router:
sub_intent: COMPONENT_EXPLAIN
# API
- id: v4-api
query: "Что делает endpoint /health?"
expected:
router:
sub_intent: API_METHOD_EXPLAIN
# ENTITY
- id: v4-entity
query: "Что такое runtime health?"
expected:
router:
sub_intent: ENTITY_EXPLAIN
# RELATED DOCS
- id: v4-related
query: "Какие документы связаны с endpoint /health?"
expected:
router:
sub_intent: RELATED_DOCS_EXPLAIN
# OPENAPI METHOD
- id: v4-openapi-method
query: "Сгенерируй OpenAPI для endpoint /health"
expected:
router:
intent: OPENAPI_GENERATION
# OPENAPI FRAGMENT
- id: v4-openapi-fragment
query: "Сгенерируй request schema для endpoint /send"
expected:
router:
sub_intent: OPENAPI_FRAGMENT_GENERATE
# GENERAL
- id: v4-general
query: "Что описано в документации?"
expected:
router:
sub_intent: GENERIC_QA
# NEGATIVE
- id: v4-negative
query: "Что такое PaymentTransaction?"
expected:
router:
sub_intent: ENTITY_EXPLAIN
# EDGE MIXED
- id: v4-mixed
query: "Как работает /send и какие есть связанные документы?"
expected:
router:
intent: DOCUMENTATION_EXPLAIN