feat: switch to gpt5 models #92

configurations/clients/sample/channels.yaml

@@ -18,10 +18,18 @@ channels:
             datasets, as it contains data directly provided by national statistical
             agencies.
         llm_model_config:
-          deployment: "gpt-4.1-2025-04-14"
+          deployment: "gpt-5.2-2025-12-11"
+          reasoningEffort: "none"
+          verbosity: "low"
+          temperature: 1
+          seed: null
       out_of_scope:
         llm_model_config:
-          deployment: "gpt-4.1-2025-04-14"
+          deployment: "gpt-5.2-2025-12-11"
+          reasoningEffort: "none"
+          verbosity: "low"
+          temperature: 1
+          seed: null
         use_general_topics_blacklist: true
         domain: "Statistics, economics and SDMX."
       token_usage:

configurations/clients/sample/tools.yaml

@@ -1,29 +1,54 @@
 _available_datasets_call_id: &available_datasets_call_id "call_EBJJeaOMKeCzm8h378ubURQN"
 _available_terms_tool_call_id: &availableTermsCallId "call_EBJJeaOMKeCzm8h378ubU003"
-_data_query_gpt_41_models: &data-query-gpt-41-models
+_data_query_gpt_5_models: &data-query-gpt-5-models
   datasetsSelectionModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   dimensionsSelectionModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null  
   indicatorsSelectionModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   incompleteQueriesModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   groupExpanderModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   namedEntitiesModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   queryNormalizationModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
   timePeriodModelConfig:
-    deployment: "gpt-4.1-2025-04-14"
-    temperature: 0.0
+    deployment: "gpt-5.2-2025-12-11"
+    reasoningEffort: "none"
+    verbosity: "low"
+    temperature: 1
+    seed: null
 _relevancy_prompts: &relevancy_prompts
   systemMessage: |-
     You are an expert in statistical indicators.
@@ -121,7 +146,7 @@ tools:
         namedEntitiesToRemove: ["Country/Reference area", "Counterpart area/country"]
         prompts:
           relevancyPrompts: *relevancy_prompts
-      llmModels: *data-query-gpt-41-models
+      llmModels: *data-query-gpt-5-models
       attachments:
         customTable:
           enabledStr: "True"

statgpt/admin/settings/background_tasks.py

@@ -11,5 +11,5 @@ class BackgroundTasksSettings(BaseSettings):
 
     max_concurrent: int = Field(5, description="Maximum number of concurrent background tasks")
     task_timeout: float | None = Field(
-        1800.0, description="Timeout in seconds for a single background task (None to disable)"
+        3600.0, description="Timeout in seconds for a single background task (None to disable)"
     )

statgpt/app/schemas/file_rags/base.py

@@ -4,7 +4,7 @@
 
 class BaseRagState(ToolMessageState):
     type: ToolTypes = ToolTypes.FILE_RAG
-    version: RAGVersion
+    version: RAGVersion = RAGVersion.DIAL
 
     response: str = ""  # This is not needed since we have content field
     answered_by: str = ""

statgpt/common/schemas/data_query_tool.py

@@ -3,7 +3,7 @@
 from pydantic import Field, PositiveInt, TypeAdapter, field_validator, model_validator
 from pydantic_core.core_schema import FieldValidationInfo
 
-from statgpt.common.config import LLMModelsEnum
+from statgpt.common.config import LLMModelsEnum, ReasoningEffortEnum, VerbosityEnum
 from statgpt.common.config.utils import replace_env
 
 from .base import BaseYamlModel, SystemUserPrompt
@@ -175,11 +175,21 @@ class HybridSearchConfig(BaseYamlModel):
 
     normalize_model_config: LLMModelConfig = Field(
         description="LLM Model used for normalization",
-        default_factory=lambda: LLMModelConfig(deployment=LLMModelsEnum.GPT_4_1_MINI_2025_04_14),
+        default_factory=lambda: LLMModelConfig(
+            deployment=LLMModelsEnum.GPT_5_MINI_2025_08_07,
+            reasoning_effort=ReasoningEffortEnum.MINIMAL,
+            verbosity=VerbosityEnum.LOW,
+            temperature=1,
+        ),
     )
     harmonize_model_config: LLMModelConfig = Field(
         description="LLM Model used for harmonization",
-        default_factory=lambda: LLMModelConfig(deployment=LLMModelsEnum.GPT_4_1_MINI_2025_04_14),
+        default_factory=lambda: LLMModelConfig(
+            deployment=LLMModelsEnum.GPT_5_MINI_2025_08_07,
+            reasoning_effort=ReasoningEffortEnum.MINIMAL,
+            verbosity=VerbosityEnum.LOW,
+            temperature=1,
+        ),
     )
 
     # ~~~~~~~~~~ Search config ~~~~~~~~~~

statgpt/common/settings/langchain.py

@@ -2,7 +2,12 @@
 from pydantic import Field
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
-from statgpt.common.config.llm_models import EmbeddingModelsEnum, LLMModelsEnum
+from statgpt.common.config.llm_models import (
+    EmbeddingModelsEnum,
+    LLMModelsEnum,
+    ReasoningEffortEnum,
+    VerbosityEnum,
+)
 
 
 class LangChainSettings(BaseSettings):
@@ -38,6 +43,16 @@ class LangChainSettings(BaseSettings):
         description="Default seed for reproducible outputs",
     )
 
+    default_reasoning_effort: ReasoningEffortEnum | None = Field(
+        default=ReasoningEffortEnum.NONE,
+        description="Default reasoning effort for GPT-5 models (none/minimal/low/medium/high/xhigh)",
+    )
+
+    default_verbosity: VerbosityEnum | None = Field(
+        default=VerbosityEnum.LOW,
+        description="Default verbosity for GPT-5 models (low/medium/high). None means use model default.",
+    )
+
     # Debugging settings
     verbose: bool = Field(default=False, description="Enable verbose mode for LangChain")
 

statgpt/common/utils/callbacks.py

@@ -124,29 +124,30 @@ class TokenUsageByModelsCallback(AsyncCallbackHandler):
 
     def __init__(self) -> None:
         super().__init__()
-        self._run_2_deployment: dict[UUID, str] = {}
+        self._deployment_ids: dict[UUID, str] = {}
 
-    async def on_chat_model_start(
+    def on_chat_model_start(  # type: ignore[override]
         self,
         serialized: dict[str, t.Any],
         messages: list[list[t.Any]],
         *,
         run_id: UUID,
         **kwargs: t.Any,
     ) -> None:
-        try:
-            self._run_2_deployment[run_id] = serialized['kwargs']['deployment_name']
-        except (KeyError, TypeError):
-            pass
+        if serialized['id'][-1] == 'AzureChatOpenAI':
+            try:
+                self._deployment_ids[run_id] = serialized['kwargs']['deployment_name']
+            except (KeyError, TypeError):
+                pass
 
-    async def on_llm_end(
+    def on_llm_end(  # type: ignore[override]
         self,
         response: LLMResult,
         *,
         run_id: UUID,
         **kwargs: t.Any,
     ) -> None:
-        deployment_id = self._run_2_deployment.pop(run_id, None)
+        deployment_id = self._deployment_ids.pop(run_id, None)
 
         try:
             generation = response.generations[0][0]

statgpt/common/utils/models.py

@@ -54,13 +54,10 @@ def get_chat_model(
     params.update(model_config.model_dump(mode="json", exclude_none=True, exclude={"deployment"}))
 
     if model_config.deployment.is_gpt_41_family:
-        callback = BrokenResponseInterceptor(regex_pattern=r'\s{5,}')
-        params.setdefault('callbacks', []).append(callback)
+        callback = BrokenResponseInterceptor(regex_pattern=r"\s{5,}")
+        params.setdefault("callbacks", []).append(callback)
 
-    api_key_log = f'{api_key.get_secret_value()[:3]}*****{api_key.get_secret_value()[-2:]}'
-    logger.info(
-        f'creating langchain LLM with the following params: {params}, Api key: {api_key_log}'
-    )
+    logger.info(f"creating langchain LLM with the following params: {params}")
     return AzureChatOpenAI.model_validate(params)
 
 
@@ -78,8 +75,5 @@ def get_embeddings_model(
         max_retries=10,
         api_key=api_key,  # since we use SecretStr, it won't be logged
     )
-    api_key_log = f'{api_key.get_secret_value()[:3]}*****{api_key.get_secret_value()[-2:]}'
-    logger.info(
-        f'creating langchain embeddings with the following params: {params}, Api key: {api_key_log}'
-    )
+    logger.info(f"creating langchain embeddings with the following params: {params}")
     return AzureOpenAIEmbeddings.model_validate(params)