Skip to content

armgit / camel

Public Imported from GitHub
0

🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org

Fix flaky `test_chat_agent_batch_stream_same_output` (#212)

Zecheng Zhang9103fb2c0ef1
diff --git a/test/agents/test_chat_agent.py b/test/agents/test_chat_agent.py
index 5905e0abf..2e278f2f5 100644
--- a/test/agents/test_chat_agent.py
+++ b/test/agents/test_chat_agent.py
@@ -83,31 +83,25 @@ def test_chat_agent_multiple_return_messages(n):
@pytest.mark.model_backend
-def test_chat_agent_batch_stream_same_output():
+def test_chat_agent_stream_output():
system_msg = BaseMessage("Assistant", RoleType.ASSISTANT, meta_dict=None,
content="You are a helpful assistant.")
user_msg = BaseMessage(role_name="User", role_type=RoleType.USER,
meta_dict=dict(), content="Tell me a joke.")
- batch_model_config = ChatGPTConfig(temperature=0, n=2, stream=False)
- batch_assistant = ChatAgent(system_msg, model_config=batch_model_config)
- batch_assistant.reset()
- batch_assistant_response = batch_assistant.step(user_msg)
stream_model_config = ChatGPTConfig(temperature=0, n=2, stream=True)
stream_assistant = ChatAgent(system_msg, model_config=stream_model_config)
stream_assistant.reset()
stream_assistant_response = stream_assistant.step(user_msg)
- for i in range(len(batch_assistant_response.msgs)):
- batch_msg_content = batch_assistant_response.msgs[i].content
- stream_msg_content = stream_assistant_response.msgs[i].content
- assert batch_msg_content == stream_msg_content
+ for msg in stream_assistant_response.msgs:
+ assert len(msg.content) > 0
- batch_usage = batch_assistant_response.info["usage"]
stream_usage = stream_assistant_response.info["usage"]
- for key in batch_usage:
- assert key in stream_usage
- assert stream_usage[key] == batch_usage[key]
+ assert stream_usage["completion_tokens"] > 0
+ assert stream_usage["prompt_tokens"] > 0
+ assert stream_usage["total_tokens"] == stream_usage[
+ "completion_tokens"] + stream_usage["prompt_tokens"]
@pytest.mark.model_backend