fix(openai_api_compatible): add support for Zhipu AI thinking mode fo… (#2363)

* fix(openai_api_compatible): add support for Zhipu AI thinking mode format

- Add top-level thinking parameter support for Zhipu API compatibility
- Support both enabled and disabled thinking modes via {type: 'enabled'/'disabled'} format
- Maintain backward compatibility with vLLM/SGLang chat_template_kwargs format
- Bump version from 0.0.28 to 0.0.29

* Apply suggestions from code review

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

---------

Co-authored-by: DeiuDesHommies <[email protected]>
Co-authored-by: crazywoola <[email protected]>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Author: 3 people

models/openai_api_compatible/manifest.yaml

@@ -1,4 +1,4 @@
-version: 0.0.28
+version: 0.0.29
 type: plugin
 author: "langgenius"
 name: "openai_api_compatible"

models/openai_api_compatible/models/llm/llm.py

@@ -187,9 +187,16 @@ def _invoke(
                 enable_thinking_value = bool(user_enable_thinking)
 
         if enable_thinking_value is not None:
-            model_parameters.setdefault("chat_template_kwargs", {})["enable_thinking"] = enable_thinking_value
-            # Add From: https://github.com/langgenius/dify-official-plugins/pull/2151
-            model_parameters.setdefault("chat_template_kwargs", {})["thinking"] = enable_thinking_value
+            # Support vLLM/SGLang format (chat_template_kwargs)
+            chat_template_kwargs = model_parameters.setdefault("chat_template_kwargs", {})
+            chat_template_kwargs["enable_thinking"] = enable_thinking_value
+            chat_template_kwargs["thinking"] = enable_thinking_value
+
+            # Support Zhipu AI API format (top-level thinking parameter)
+            # This allows compatibility with Zhipu's official API format: {"thinking": {"type": "enabled/disabled"}}
+            model_parameters["thinking"] = {
+                "type": "enabled" if enable_thinking_value else "disabled"
+            }
 
         # Remove thinking content from assistant messages for better performance.
         with suppress(Exception):