MacBook Air上ChatOllama调用Llama3失败 VSCode报错求助

环境信息 #

• 设备：8GB内存MacBook Air
• 已部署Ollama服务，运行地址http://localhost:11434
• 终端可正常与Llama3交互，但VSCode中通过LangChain调用时抛出异常

错误日志 #

Checkpoint 1
Checkpoint 2
checkpoint 3
<class 'list'>
<class 'langchain_core.messages.human.HumanMessage'>
Input to invoke: [HumanMessage(content='Are you ready?', additional_kwargs={}, response_metadata={})]
Traceback (most recent call last):
  File "/Users/sandeephugar/VSCProjects/purplekicks/manager/llmtest.py", line 156, in <module>
    response = llm.invoke(test_message)
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_core/language_models/chat_models.py", line 286, in invoke
    self.generate_prompt(
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_core/language_models/chat_models.py", line 786, in generate_prompt
    return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_core/language_models/chat_models.py", line 643, in generate
    raise e
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_core/language_models/chat_models.py", line 633, in generate
    self._generate_with_cache(
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_core/language_models/chat_models.py", line 851, in _generate_with_cache
    result = self._generate(
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_ollama/chat_models.py", line 644, in _generate
    final_chunk = self._chat_stream_with_aggregation(
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_ollama/chat_models.py", line 558, in _chat_stream_with_aggregation
    tool_calls=_get_tool_calls_from_response(stream_resp),
  File "/Users/sandeephugar/VSCProjects/purplekicks/env/lib/python3.9/site-packages/langchain_ollama/chat_models.py", line 70, in _get_tool_calls_from_response
    for tc in response["message"]["tool_calls"]:
TypeError: 'NoneType' object is not iterable

测试代码 #

from langchain_ollama import ChatOllama
from langchain_core.messages import AIMessage, HumanMessage

# Initialize the LLM (Ollama)
llm = ChatOllama(model="llama3")


print("Checkpoint 1")

# Check if the model is accessible
if not llm:
    raise ValueError("LLM instance is not initialized properly.")

print("Checkpoint 2")

# Test a simple input
test_message = [HumanMessage(content="Are you ready?")]
print("checkpoint 3")
print(type(test_message))  # Should print <class 'list'>
print(type(test_message[0]))
print("Input to invoke:", test_message)

# Invoke the model and get the response
response = llm.invoke(test_message)
print(response.content)


# Check if the response is not None
if response is None:
    raise ValueError("The model did not respond. Ensure the Ollama server is running and the model is loaded.")

print("LLM is ready and responding.")
print("Checkpoint 4")

# Iterate over the response list and print each message content
for msg in response:
    print(msg.content)

print("Checkpoint 6")

已尝试方案 #

• 调试并处理错误
• 修改消息格式为元组列表形式：

  test_message = [(
      "system",
      "You are a helpful assistant here to chat.",
  ),
  ("human", "hello, nice to meet you!"),]
  

解决方案 #

1. 升级langchain-ollama依赖包 #

该错误是旧版本langchain-ollama的已知bug，当模型未返回工具调用结果时，代码会尝试遍历None值。升级到最新版本可修复此问题：

pip install --upgrade langchain-ollama

2. 修正代码中的遍历错误 #

原代码最后for msg in response:会报错，因为llm.invoke()返回的是单个AIMessage对象，而非列表。修正后的代码片段：

# Invoke the model and get the response
response = llm.invoke(test_message)
print(response.content)

print("LLM is ready and responding.")
print("Checkpoint 4")

# 直接打印响应内容，无需遍历
print("Response content:", response.content)

print("Checkpoint 6")

3. 可选：显式指定文本格式（若升级后仍有问题） #

初始化ChatOllama时显式设置输出格式为文本，避免默认的工具调用格式处理：

llm = ChatOllama(model="llama3", format="text")

内容的提问来源于stack exchange，提问作者cooldude