Add reasoning param to configure LLM instead of max_token

.gitignore

@@ -16,7 +16,7 @@ wheels/
 # Local Files and Directories
 .output_dataset
 output/
-/resources/
+/resources*/
 **/tmp
 
 # Docker

examples/configs/dataset_generator/llm_config.yaml

@@ -1,13 +1,17 @@
 # llm model configuration
 
-# OpenAI LLM
+# LLM provider
+# Accepted values: openai, gemini
 name: openai
 
 # Chat model name
 model: gpt-5-nano-2025-08-07
 
-# Maximum number of tokens the model may return
-max_tokens: 100000
+# (Optional) Reasoning effort parameter of the LLM.
+# If reasoning_effort is not provided, the related Langchain parameter is not set.
+# If reasoning_effort is set, and its level is not found, the level is set to the lowest level for the given model.
+# Since the different LLM have different levels of reasoning, you must refer to the langchain documentation.
+reasoning_effort: minimal
 
 # Environment variable where LLM API key is stored
 api_key_env: OPENAI_API_KEY

src/llm_search_quality_evaluation/dataset_generator/llm/llm_config.py

@@ -2,20 +2,36 @@
 
 import logging
 from pathlib import Path
-from typing import Optional
+from typing import Optional, Literal
 
 import yaml
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 log = logging.getLogger(__name__)
 
 
 class LLMConfig(BaseModel):
-    name: str
+    name: Literal['openai', 'gemini']
     model: str
-    max_tokens: int = Field(default=512, gt=0)
+    reasoning_effort: Optional[str] = Field(default=None, description="The reasoning effort of the model")
     api_key_env: Optional[str] = None
 
+    @model_validator(mode="after")
+    def set_reasoning_effort_defaults(self) -> "LLMConfig":
+        if self.reasoning_effort is not None:
+            if self.name == "openai":
+                default_effort_mode = "minimal"
+                effort_list = ["minimal", "low", "medium", "high"]
+            else:
+                default_effort_mode = "low"
+                effort_list = ["low", "high"]
+
+            if self.reasoning_effort not in effort_list:
+                self.reasoning_effort = default_effort_mode
+
+        return self
+
+
     @classmethod
     def load(cls, path: str | Path = "llm_config.yaml") -> LLMConfig:
         path = Path(path).resolve()

src/llm_search_quality_evaluation/dataset_generator/llm/llm_provider_factory.py

@@ -26,13 +26,19 @@ def build_openai(config: LLMConfig) -> BaseChatModel:
     if not key:
         log.error("OpenAI API key not set %s in the env", config.api_key_env)
         raise ValueError("OpenAI API key not set.")
-    log.debug("Building OpenAI ChatModel using name=%s, model=%s, max_tokens=%s",
-              config.name, config.model, config.max_tokens)
-    return ChatOpenAI(
-        model=config.model,
-        max_tokens=config.max_tokens,  # type: ignore[arg-type]
-        api_key=SecretStr(key),
-    )
+    log.debug("Building OpenAI ChatModel using name=%s, model=%s, reasoning_effort=%s",
+              config.name, config.model, config.reasoning_effort)
+    if config.reasoning_effort is None:
+        return ChatOpenAI(
+            model=config.model,
+            api_key=SecretStr(key),
+        )
+    else:
+        return ChatOpenAI(
+            model=config.model,
+            api_key=SecretStr(key),
+            reasoning_effort=config.reasoning_effort,
+        )
 
 
 def build_gemini(config: LLMConfig) -> BaseChatModel:
@@ -41,12 +47,16 @@ def build_gemini(config: LLMConfig) -> BaseChatModel:
     if not key:
         log.error("Google Gemini API key not set %s in the env", config.api_key_env)
         raise ValueError("Google Gemini API key not set.")
-    log.debug("Building Google Gemini ChatModel using name=%s, model=%s, max_tokens=%s",
-              config.name, config.model, config.max_tokens)
+    log.debug("Building Google Gemini ChatModel using name=%s, model=%s, reasoning_effort=%s",
+              config.name, config.model, config.reasoning_effort)
     return ChatGoogleGenerativeAI(
         model=config.model,
-        max_output_tokens=config.max_tokens,
         google_api_key=key,
+        model_kwargs={
+            "thinking_config": {
+                "thinking_budget": config.reasoning_effort,
+            }
+        }
     )
 
 

tests/resources/llm_config.yaml

@@ -1,4 +1,2 @@
-llm_provider: "openai"
-model: "gpt-3.5-turbo"
-temperature: 0.7
-max_tokens: 1000
+name: openai
+model: gpt-5-nano-2025-08-07