Skip to content

Build and run agents you can see, understand and trust.

fix(model): honor explicit thinking_enable=False on Ollama and Gemini (#1784)

qbc8ff9182f8b26
diff --git a/scripts/model_examples/openai_chat_call.py b/scripts/model_examples/openai_chat_call.py
index 132ab59a6..660918ffa 100644
--- a/scripts/model_examples/openai_chat_call.py
+++ b/scripts/model_examples/openai_chat_call.py
@@ -34,6 +34,7 @@ async def example_simple_call() -> None:
stream=True,
context_size=200_000,
parameters=OpenAIChatModel.Parameters(
+ thinking_enable=True,
reasoning_effort="low",
),
)
@@ -80,6 +81,7 @@ async def example_tool_call() -> None:
stream=True,
context_size=200_000,
parameters=OpenAIChatModel.Parameters(
+ thinking_enable=True,
reasoning_effort="low",
),
)
@@ -155,6 +157,7 @@ async def example_structured_output() -> None:
stream=True,
context_size=200_000,
parameters=OpenAIChatModel.Parameters(
+ thinking_enable=True,
reasoning_effort="low",
),
)
diff --git a/scripts/model_examples/openai_response_call.py b/scripts/model_examples/openai_response_call.py
index 62c9195f8..c93a30901 100644
--- a/scripts/model_examples/openai_response_call.py
+++ b/scripts/model_examples/openai_response_call.py
@@ -38,7 +38,10 @@ async def example_simple_call() -> None:
model="o1",
stream=True,
context_size=200_000,
- parameters=OpenAIResponseModel.Parameters(reasoning_effort="low"),
+ parameters=OpenAIResponseModel.Parameters(
+ thinking_enable=True,
+ reasoning_effort="low",
+ ),
)
msgs = [
@@ -82,7 +85,10 @@ async def example_tool_call() -> None:
model="o1",
stream=True,
context_size=200_000,
- parameters=OpenAIResponseModel.Parameters(reasoning_effort="low"),
+ parameters=OpenAIResponseModel.Parameters(
+ thinking_enable=True,
+ reasoning_effort="low",
+ ),
)
msgs = [
@@ -158,7 +164,10 @@ async def example_structured_output() -> None:
model="o1",
stream=True,
context_size=200_000,
- parameters=OpenAIResponseModel.Parameters(reasoning_effort="low"),
+ parameters=OpenAIResponseModel.Parameters(
+ thinking_enable=True,
+ reasoning_effort="low",
+ ),
)
msgs = [
diff --git a/scripts/model_examples/xai_call.py b/scripts/model_examples/xai_call.py
index 97e88d249..2d44e4dce 100644
--- a/scripts/model_examples/xai_call.py
+++ b/scripts/model_examples/xai_call.py
@@ -38,7 +38,10 @@ async def example_simple_call() -> None:
model="grok-3-mini",
stream=True,
context_size=131_072,
- parameters=XAIChatModel.Parameters(reasoning_effort="low"),
+ parameters=XAIChatModel.Parameters(
+ thinking_enable=True,
+ reasoning_effort="low",
+ ),
)
msgs = [
@@ -157,7 +160,10 @@ async def example_structured_output() -> None:
model="grok-3-mini",
stream=True,
context_size=131_072,
- parameters=XAIChatModel.Parameters(reasoning_effort="low"),
+ parameters=XAIChatModel.Parameters(
+ thinking_enable=True,
+ reasoning_effort="low",
+ ),
)
msgs = [
diff --git a/src/agentscope/model/_gemini/_model.py b/src/agentscope/model/_gemini/_model.py
index 946e13841..ce84e678f 100644
--- a/src/agentscope/model/_gemini/_model.py
+++ b/src/agentscope/model/_gemini/_model.py
@@ -239,6 +239,11 @@ class GeminiChatModel(ChatModelBase):
"include_thoughts": True,
"thinking_budget": self.parameters.thinking_budget or 1024,
}
+ else:
+ config["thinking_config"] = {
+ "include_thoughts": False,
+ "thinking_budget": 0,
+ }
fmt_tools, fmt_tool_choice = self._format_tools(tools, tool_choice)
diff --git a/src/agentscope/model/_ollama/_model.py b/src/agentscope/model/_ollama/_model.py
index 81b42e4da..d507771dd 100644
--- a/src/agentscope/model/_ollama/_model.py
+++ b/src/agentscope/model/_ollama/_model.py
@@ -175,8 +175,7 @@ class OllamaChatModel(ChatModelBase):
if options:
kwargs["options"] = options
- if self.parameters.thinking_enable:
- kwargs["think"] = True
+ kwargs["think"] = self.parameters.thinking_enable
kwargs.update(generate_kwargs)