Merge pull request #1561 from KennyDizi/main

Support reasoning effort via configuration

Author: mrT23

docs/docs/usage-guide/changing_a_model.md

@@ -204,3 +204,12 @@ custom_model_max_tokens= ...
 
 4. Most reasoning models do not support chat-style inputs (`system` and `user` messages) or temperature settings. 
 To bypass chat templates and temperature controls, set `config.custom_reasoning_model = true` in your configuration file.
+
+## Dedicated parameters
+
+### OpenAI models
+
+[config]
+reasoning_efffort= = "medium" # "low", "medium", "high"
+
+With the OpenAI models that support reasoning effort (eg: o3-mini), you can specify its reasoning effort via `config` section. The default value is `medium`. You can change it to `high` or `low` based on your usage.

pr_agent/algo/__init__.py

@@ -104,3 +104,8 @@
     "o3-mini-2025-01-31",
     "o1-preview"
 ]
+
+SUPPORT_REASONING_EFFORT_MODELS = [
+    "o3-mini",
+    "o3-mini-2025-01-31"
+]

pr_agent/algo/ai_handlers/litellm_ai_handler.py

@@ -6,9 +6,9 @@
 from litellm import acompletion
 from tenacity import retry, retry_if_exception_type, stop_after_attempt
 
-from pr_agent.algo import NO_SUPPORT_TEMPERATURE_MODELS, USER_MESSAGE_ONLY_MODELS
+from pr_agent.algo import NO_SUPPORT_TEMPERATURE_MODELS, SUPPORT_REASONING_EFFORT_MODELS, USER_MESSAGE_ONLY_MODELS
 from pr_agent.algo.ai_handlers.base_ai_handler import BaseAiHandler
-from pr_agent.algo.utils import get_version
+from pr_agent.algo.utils import ReasoningEffort, get_version
 from pr_agent.config_loader import get_settings
 from pr_agent.log import get_logger
 
@@ -101,6 +101,9 @@ def __init__(self):
         # Model that doesn't support temperature argument
         self.no_support_temperature_models = NO_SUPPORT_TEMPERATURE_MODELS
 
+        # Models that support reasoning effort
+        self.support_reasoning_models = SUPPORT_REASONING_EFFORT_MODELS
+
     def prepare_logs(self, response, system, user, resp, finish_reason):
         response_log = response.dict().copy()
         response_log['system'] = system
@@ -228,8 +231,16 @@ async def chat_completion(self, model: str, system: str, user: str, temperature:
 
             # Add temperature only if model supports it
             if model not in self.no_support_temperature_models and not get_settings().config.custom_reasoning_model:
+                get_logger().info(f"Adding temperature with value {temperature} to model {model}.")
                 kwargs["temperature"] = temperature
 
+            # Add reasoning_effort if model supports it
+            if (model in self.support_reasoning_models):
+                supported_reasoning_efforts = [ReasoningEffort.HIGH.value, ReasoningEffort.MEDIUM.value, ReasoningEffort.LOW.value]
+                reasoning_effort = get_settings().config.reasoning_effort if (get_settings().config.reasoning_effort in supported_reasoning_efforts) else ReasoningEffort.MEDIUM.value
+                get_logger().info(f"Adding reasoning_effort with value {reasoning_effort} to model {model}.")
+                kwargs["reasoning_effort"] = reasoning_effort
+
             if get_settings().litellm.get("enable_callbacks", False):
                 kwargs = self.add_litellm_callbacks(kwargs)
 

pr_agent/algo/utils.py

@@ -50,6 +50,11 @@ class PRReviewHeader(str, Enum):
     REGULAR = "## PR Reviewer Guide"
     INCREMENTAL = "## Incremental PR Reviewer Guide"
 
+class ReasoningEffort(str, Enum):
+    HIGH = "high"
+    MEDIUM = "medium"
+    LOW = "low"
+
 
 class PRDescriptionHeader(str, Enum):
     CHANGES_WALKTHROUGH = "### **Changes walkthrough** 📝"

pr_agent/settings/configuration.toml

@@ -48,6 +48,7 @@ ignore_pr_authors = [] # authors to ignore from PR agent when an PR is created
 #
 is_auto_command = false # will be auto-set to true if the command is triggered by an automation
 enable_ai_metadata = false # will enable adding ai metadata
+reasoning_effort = "medium" # "low", "medium", "high"
 # auto approval 💎
 enable_auto_approval=false # Set to true to enable auto-approval of PRs under certain conditions
 auto_approve_for_low_review_effort=-1 # -1 to disable, [1-5] to set the threshold for auto-approval