feat: enhance OpenAI-compatible model support with role-specific configurations (#356)

2025-12-26 10:07:51 +08:00 · 2025-03-21 15:44:08 +08:00 · 2025-03-21 15:44:08 +08:00 · bd8f220416
commit bd8f220416
parent 106a169ee9 2e034d95e1
7 changed files with 99 additions and 31 deletions
--- a/.gitignore
+++ b/.gitignore
@ -27,6 +27,7 @@ venv/
 env/
 ENV/
 .env
+.venv

 # IDE
 .idea/
@ -58,3 +59,4 @@ coverage.xml
 owl/camel/types/__pycache__/
 owl/camel/__pycache__/
 owl/camel/utils/__pycache_/
+tmp/
--- a/README.md
+++ b/README.md
@ -364,8 +364,10 @@ python examples/run_qwen_zh.py
 # Run with Deepseek model
 python examples/run_deepseek_zh.py

-# Run with other OpenAI-compatible models
+# Run with other OpenAI-compatible models, supporting different models for different roles
 python examples/run_openai_compatiable_model.py
+# Example with question
+python examples/run_openai_compatiable_model.py "Navigate to Amazon.com and identify one product that is attractive to coders. Please provide me with the product name and price. No need to verify your answer."

 # Run with Azure OpenAI
 python examples/run_azure_openai.py
--- a/README_zh.md
+++ b/README_zh.md
@ -363,8 +363,10 @@ python examples/run_qwen_zh.py
 # 使用 Deepseek 模型运行
 python examples/run_deepseek_zh.py

-# 使用其他 OpenAI 兼容模型运行
+# 使用其他 OpenAI 兼容模型运行，支持不同的 role 使用不同的模型
 python examples/run_openai_compatiable_model.py
+# 带问题的示例
+python examples/run_openai_compatiable_model.py "浏览京东并找出一款对程序员有吸引力的产品。请提供产品名称和价格。"

 # 使用 Azure OpenAI模型运行
 python examples/run_azure_openai.py
--- a/examples/run_openai_compatiable_model.py
+++ b/examples/run_openai_compatiable_model.py
@ -53,38 +53,56 @@ def construct_society(question: str) -> RolePlaying:
    models = {
        "user": ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI_COMPATIBLE_MODEL,
-            model_type="qwen-max",
-            api_key=os.getenv("QWEN_API_KEY"),
-            url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-            model_config_dict={"temperature": 0.4, "max_tokens": 128000},
+
+            model_type=os.getenv("USER_ROLE_API_MODEL_TYPE", os.getenv("LLM_ROLE_API_MODEL_TYPE", "qwen-max")),
+            api_key=os.getenv("USER_ROLE_API_KEY", os.getenv("LLM_ROLE_API_KEY", os.getenv("QWEN_API_KEY", "Your_Key"))),
+            url=os.getenv("USER_ROLE_API_BASE_URL", os.getenv("LLM_ROLE_API_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")),
+            model_config_dict={
+                "temperature": float(os.getenv("USER_ROLE_API_MODEL_TEMPERATURE", os.getenv("LLM_ROLE_API_MODEL_TEMPERATURE", "0.4"))), 
+                "max_tokens": int(os.getenv("USER_ROLE_API_MODEL_MAX_TOKENS", os.getenv("LLM_ROLE_API_MODEL_MAX_TOKENS", "4096")))
+            },
        ),
        "assistant": ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI_COMPATIBLE_MODEL,
-            model_type="qwen-max",
-            api_key=os.getenv("QWEN_API_KEY"),
-            url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-            model_config_dict={"temperature": 0.4, "max_tokens": 128000},
+            model_type=os.getenv("ASSISTANT_ROLE_API_MODEL_TYPE", os.getenv("LLM_ROLE_API_MODEL_TYPE", "qwen-max")),
+            api_key=os.getenv("ASSISTANT_ROLE_API_KEY", os.getenv("LLM_ROLE_API_KEY", os.getenv("QWEN_API_KEY", "Your_Key"))),
+            url=os.getenv("ASSISTANT_ROLE_API_BASE_URL", os.getenv("LLM_ROLE_API_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")),
+            model_config_dict={
+                "temperature": float(os.getenv("ASSISTANT_ROLE_API_MODEL_TEMPERATURE", os.getenv("LLM_ROLE_API_MODEL_TEMPERATURE", "0.4"))), 
+                "max_tokens": int(os.getenv("ASSISTANT_ROLE_API_MODEL_MAX_TOKENS", os.getenv("LLM_ROLE_API_MODEL_MAX_TOKENS", "4096")))
+            },
+
        ),
        "browsing": ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI_COMPATIBLE_MODEL,
-            model_type="qwen-vl-max",
-            api_key=os.getenv("QWEN_API_KEY"),
-            url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-            model_config_dict={"temperature": 0.4, "max_tokens": 128000},
+
+            model_type=os.getenv("WEB_ROLE_API_BASE_URL", os.getenv("VLLM_ROLE_API_MODEL_TYPE", "qwen-vl-max")),
+            api_key=os.getenv("WEB_ROLE_API_KEY", os.getenv("VLLM_ROLE_API_KEY", os.getenv("QWEN_API_KEY", "Your_Key"))),
+            url=os.getenv("USER_ROLE_API_BASE_URL", os.getenv("VLLM_ROLE_API_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")),
+            model_config_dict={
+                "temperature": float(os.getenv("WEB_ROLE_API_MODEL_TEMPERATURE", os.getenv("VLLM_ROLE_API_MODEL_TEMPERATURE", "0.4"))), 
+                "max_tokens": int(os.getenv("WEB_ROLE_API_MODEL_MAX_TOKENS", os.getenv("VLLM_ROLE_API_MODEL_MAX_TOKENS", "4096")))
+            },
        ),
        "planning": ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI_COMPATIBLE_MODEL,
-            model_type="qwen-max",
-            api_key=os.getenv("QWEN_API_KEY"),
-            url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-            model_config_dict={"temperature": 0.4, "max_tokens": 128000},
+            model_type=os.getenv("PLANNING_ROLE_API_MODEL_TYPE", os.getenv("LLM_ROLE_API_MODEL_TYPE", "qwen-max")),
+            api_key=os.getenv("PLANNING_ROLE_API_KEY", os.getenv("LLM_ROLE_API_KEY", os.getenv("QWEN_API_KEY", "Your_Key"))),
+            url=os.getenv("PLANNING_ROLE_API_BASE_URL", os.getenv("LLM_ROLE_API_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")),
+            model_config_dict={
+                "temperature": float(os.getenv("PLANNING_ROLE_API_MODEL_TEMPERATURE", os.getenv("LLM_ROLE_API_MODEL_TEMPERATURE", "0.4"))), 
+                "max_tokens": int(os.getenv("PLANNING_ROLE_API_MODEL_MAX_TOKENS", os.getenv("LLM_ROLE_API_MODEL_MAX_TOKENS", "4096")))
+            },
        ),
        "image": ModelFactory.create(
            model_platform=ModelPlatformType.OPENAI_COMPATIBLE_MODEL,
-            model_type="qwen-vl-max",
-            api_key=os.getenv("QWEN_API_KEY"),
-            url="https://dashscope.aliyuncs.com/compatible-mode/v1",
-            model_config_dict={"temperature": 0.4, "max_tokens": 128000},
+            model_type=os.getenv("IMAGE_ROLE_API_MODEL_TYPE", os.getenv("VLLM_ROLE_API_MODEL_TYPE", "qwen-vl-max")),
+            api_key=os.getenv("IMAGE_ROLE_API_KEY", os.getenv("VLLM_ROLE_API_KEY", os.getenv("QWEN_API_KEY", "Your_Key"))),
+            url=os.getenv("IMAGE_ROLE_API_BASE_URL", os.getenv("VLLM_ROLE_API_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")),
+            model_config_dict={
+                "temperature": float(os.getenv("IMAGE_ROLE_API_MODEL_TEMPERATURE", os.getenv("VLLM_ROLE_API_MODEL_TEMPERATURE", "0.4"))), 
+                "max_tokens": int(os.getenv("IMAGE_ROLE_API_MODEL_MAX_TOKENS", os.getenv("VLLM_ROLE_API_MODEL_MAX_TOKENS", "4096")))
+            },
        ),
    }

@ -126,13 +144,16 @@ def construct_society(question: str) -> RolePlaying:
    return society


-def main():
-    r"""Main function to run the OWL system with an example question."""
-    # Example research question
-    default_task = "Navigate to Amazon.com and identify one product that is attractive to coders. Please provide me with the product name and price. No need to verify your answer."

-    # Override default task if command line argument is provided
-    task = sys.argv[1] if len(sys.argv) > 1 else default_task
+def main(question: str = "Navigate to Amazon.com and identify one product that is attractive to coders. Please provide me with the product name and price. No need to verify your answer."):
+    r"""Main function to run the OWL system with an example question.
+    Args:
+        question (str): The task or question to be addressed by the society.
+            If not provided, a default question will be used.
+            Defaults to "Navigate to Amazon.com and identify one product that is attractive to coders. Please provide me with the product name and price. No need to verify your answer."
+    Returns:
+        None
+    """

    # Construct and run the society
    society = construct_society(task)
@ -141,7 +162,9 @@ def main():

    # Output the result
    print(f"\033[94mAnswer: {answer}\033[0m")
+    # Output the token count
+    print(f"\033[94mToken count: {token_count}\033[0m")


 if __name__ == "__main__":
-    main()
+    main(sys.argv[1] if len(sys.argv) > 1 else "")
--- a/owl/.env_template
+++ b/owl/.env_template
@ -24,6 +24,45 @@ QWEN_API_KEY='Your_Key'
 # DeepSeek API (https://platform.deepseek.com/api_keys)
 DEEPSEEK_API_KEY='Your_Key'

+# Multi-platform LLM/VLLM API, default values for user assistant planning web image roles
+#  LLM_ROLE_API_BASE_URL=''
+#  LLM_ROLE_API_KEY='Your_Key'
+#  LLM_ROLE_API_MODEL_TYPE=''
+#  LLM_ROLE_API_MODEL_TEMPERATURE='0.0'
+#  LLM_ROLE_API_MODEL_MAX_TOKENS='0'
+#  VLLM_ROLE_API_BASE_URL=''
+#  VLLM_ROLE_API_KEY='Your_Key'
+#  VLLM_ROLE_API_MODEL_TYPE=''
+#  VLLM_ROLE_API_MODEL_TEMPERATURE='0.0'
+#  VLLM_ROLE_API_MODEL_MAX_TOKENS='0'
+
+# Multi-platform LLM/VLLM API for user assistant planning web image roles
+# USER_ROLE_API_BASE_URL=''
+# USER_ROLE_API_KEY='Your_Key'
+# USER_ROLE_API_MODEL_TYPE=''
+# USER_ROLE_API_MODEL_TEMPERATURE='0.8'
+# USER_ROLE_API_MODEL_MAX_TOKENS='4096'
+# ASSISTANT_ROLE_API_BASE_URL=''
+# ASSISTANT_ROLE_API_KEY='Your_Key'
+# ASSISTANT_ROLE_API_MODEL_TYPE=''
+# ASSISTANT_ROLE_API_MODEL_TEMPERATURE='0.2'
+# ASSISTANT_ROLE_API_MODEL_MAX_TOKENS='4096'
+# PLANNING_ROLE_API_BASE_URL=''
+# PLANNING_ROLE_API_KEY='Your_Key'
+# PLANNING_ROLE_API_MODEL_TYPE=''
+# PLANNING_ROLE_API_MODEL_TEMPERATURE='0.4'
+# PLANNING_ROLE_API_MODEL_MAX_TOKENS='8192'
+# WEB_ROLE_API_BASE_URL=''
+# WEB_ROLE_API_KEY='Your_Key'
+# WEB_ROLE_API_MODEL_TYPE=''
+# WEB_ROLE_API_MODEL_TEMPERATURE='0.0'
+# WEB_ROLE_API_MODEL_MAX_TOKENS='0'
+# IMAGE_ROLE_API_BASE_URL=''
+# IMAGE_ROLE_API_KEY='Your_Key'
+# IMAGE_ROLE_API_MODEL_TYPE=''
+# IMAGE_ROLE_API_MODEL_TEMPERATURE='0.0'
+# IMAGE_ROLE_API_MODEL_MAX_TOKENS='0'
+
 #===========================================
 # Tools & Services API
 #===========================================
--- a/owl/webapp.py
+++ b/owl/webapp.py
@ -245,7 +245,7 @@ MODULE_DESCRIPTIONS = {
    "run": "Default mode: Using OpenAI model's default agent collaboration mode, suitable for most tasks.",
    "run_mini": "Using OpenAI model with minimal configuration to process tasks",
    "run_deepseek_zh": "Using deepseek model to process Chinese tasks",
-    "run_openai_compatiable_model": "Using openai compatible model to process tasks",
+    "run_openai_compatiable_model": "Using multiple openai compatible model to process tasks",
    "run_ollama": "Using local ollama model to process tasks",
    "run_qwen_mini_zh": "Using qwen model with minimal configuration to process tasks",
    "run_qwen_zh": "Using qwen model to process tasks",
--- a/owl/webapp_zh.py
+++ b/owl/webapp_zh.py
@ -245,7 +245,7 @@ MODULE_DESCRIPTIONS = {
    "run": "默认模式：使用OpenAI模型的默认的智能体协作模式，适合大多数任务。",
    "run_mini": "使用使用OpenAI模型最小化配置处理任务",
    "run_deepseek_zh": "使用deepseek模型处理中文任务",
-    "run_openai_compatiable_model": "使用openai兼容模型处理任务",
+    "run_openai_compatiable_model": "使用多个openai兼容模型处理任务",
    "run_ollama": "使用本地ollama模型处理任务",
    "run_qwen_mini_zh": "使用qwen模型最小化配置处理任务",
    "run_qwen_zh": "使用qwen模型处理任务",