ollama 0.9 发布,支持流式响应和推理模型

简介

Ollama 连发两个大版本,分别是 v0.8.0v0.9.0,0.8 版本后直接发 0.9 大版本,中间没有小功能迭代。

在 0.8 版本中,ollama 支持工具调用流式响应(stream response with tool calls)。

在 0.9 版本中,ollama 增强了对推理模型的支持,支持开启和关闭模型的推理模式。

0.8.0

支持工具调用的开源模型比较多,有:

我们以官方仓库 ollama-python/examples 中的例子修改得到获取天气的样例代码 weather.py:

from ollama import ChatResponse, chat

def get_current_weather(location: str, format: str) -> str:
  return "33"

# Tools can still be manually defined and passed into chat
get_current_weather_tool = {
  'type': 'function',
  'function': {
    'name': 'get_current_weather',
    'description': 'Get weather of location',
    'parameters': {
      'type': 'object',
      'required': ['location', 'format'],
      'properties': {
        'location': {'location': 'string', 'description': 'The location'},
        'format': {'format': 'string', 'description': 'The result format'},
      },
    },
  },
}

messages = [{'role': 'user', 'content': 'What is the weather today in Shanghai?'}]
print('Prompt:', messages[0]['content'])

available_functions = {
  'get_current_weather': get_current_weather,
}

response: ChatResponse = chat(
  'qwen3:latest',
  messages=messages,
  tools=[get_current_weather, get_current_weather_tool],
  stream=True,
)

for resp in response:
  # print('resp {}'.format(resp))
  if not resp.message.tool_calls:
    print(resp['message']['content'], end='', flush=True)
  else:
    # There may be multiple tool calls in the response
    for tool in resp.message.tool_calls:
      # Ensure the function is available, and then call it
      if function_to_call := available_functions.get(tool.function.name):
        print('Calling function:', tool.function.name)
        print('Arguments:', tool.function.arguments)
        output = function_to_call(**tool.function.arguments)
        print('Function output:', output)
      else:
        print('Function', tool.function.name, 'not found')

    # Add the function response to messages for the model to use
    messages.append(resp.message)
    messages.append({'role': 'tool', 'content': str(output), 'name': tool.function.name})

    # Get final response from model with function outputs
    for fr in chat('qwen3:latest', messages=messages, stream=True):
      print(fr['message']['content'], end='', flush=True)

上述代码是获取指定位置的温度,得到结果后将温度传给模型,生成最终的结果:

$ python weather.py
Prompt: What is the weather today in Shanghai?
<think>
Okay, the user is asking for the weather in Shanghai today. Let me check the tools provided. There's a function called get_current_weather. The parameters required are location and format. The description mentions getting the weather for a location. Even though the parameters' descriptions are a bit unclear, location is a string, so I should provide "Shanghai" as the location. For format, maybe the user wants a simple response, so I'll use "json" as a common format. I'll call the function with these arguments.
</think>

Calling function: get_current_weather
Arguments: {'format': 'json', 'location': 'Shanghai'}
Function output: 33
<think>
Okay, the user asked about the weather in Shanghai today. I called the get_current_weather function with the location set to Shanghai and format as JSON. The response came back as "33". Hmm, that's not a typical weather description. Maybe the API returned a temperature in Celsius? 33°C is quite hot. Let me check if there's more data. Wait, the response was just "33", so maybe the API is broken or there was an error. Alternatively, maybe it's a placeholder. I should inform the user that the information is incomplete. I'll mention the temperature if I assume it's Celsius and note that other details aren't available. Let me make sure to ask them to check again later or try another source.
</think>

The current weather in Shanghai is not fully available. However, based on the data received, the temperature is approximately **33°C**. For more detailed information (e.g., humidity, wind speed, or conditions), please check again later or use a dedicated weather service. Let me know if you'd like further assistance! 🌤️

0.9.0

当前,只有 DeepSeek R1 Qwen3 支持推理能力,后续会有更多模型支持。

开启和关闭推理能力,可以在调用 Ollama API 时指定 “think”: true

curl http://localhost:11434/api/chat -d '{
  "model": "deepseek-r1",
  "messages": [
    {
      "role": "user",
      "content": "Why is the sky blue?"
    },
  ],
  "think": true
}'

也可以在 ollama 终端用/set think/set nothink 来开启和关闭推理模式。

总结

工具调用的流式响应,可以带来更好的用户体验。推理模式开启和关闭,ollama 支持的有点晚,印象中最先支持的是 qwen3。

这两个功能都很酷,目前看一些开源 对话应用、支持库都还没支持 开启、关闭推理模式,希望开源社区尽快支持。