requests模块的SSE请求

Question

老师，这里的request的请求，如果需要流式返回的话，我测试的时候，发现需要显示的加上这个stream=True的参数才能实现

Answer 1

慕村4260697 2025-12-07 09:56:18

with requests.request(
        "POST",
        "https://api.deepseek.com/chat/completions",
        headers={
            "Content-Type": "application/json",
            "Accept": "application/json",
            "Authorization": f"Bearer {os.getenv('DEEPSEEK_API_KEY')}"
        },
        # 推理模型："model": "deepseek-reasoner",
        # 聊天模型："model": "deepseek-chat",
        json={
            "model": "deepseek-reasoner",
            "messages": [
                {"role": "user", "content": "你好，介绍一下你自己"}
            ],
            "stream": True
        },
        stream=True,
) as resp:

测试了一下，确实需要增加两个

"stream": True

json中的那个stream:true是为了让模型的返回能支持流式分片返回，但是课程中的方式没有开启http库的流式返回，因此最终的结果是一次性全部打印出来了，但是内容是分片的。

需要在requests网络库里也单独再启stream:true，才能真正在控制台响应那里实现流式打印的效果。

Answer 2

泽辉呀 2025-11-12 08:35:19

能看下在外部不加stream=True的时候，resp的响应内容么？

• 我最可爱ya #1

  补充：
  在外部不加stream=True的内容是直接返回过来的是一次性是全部的，但会按照流失效果 分开的内容
  1.{"id":"710999f4-d93f-4bba-bddc-a9141de1fbe4","object":"chat.completion.chunk","created":1774319317,"model":"deepseek-reasoner","system_fingerprint":"fp_eaab8d114b_prod0820_fp8_kvcache_new_kvcache","choices":[{"index":0,"delta":{"content":null,"reasoning_content":"，"},"logprobs":null,"finish_reason":null}]}
  
  
  2.{"id":"710999f4-d93f-4bba-bddc-a9141de1fbe4","object":"chat.completion.chunk","created":1774319317,"model":"deepseek-reasoner","system_fingerprint":"fp_eaab8d114b_prod0820_fp8_kvcache_new_kvcache","choices":[{"index":0,"delta":{"content":null,"reasoning_content":"保持"},"logprobs":null,"finish_reason":null}]}
  
  3.{"id":"710999f4-d93f-4bba-bddc-a9141de1fbe4","object":"chat.completion.chunk","created":1774319317,"model":"deepseek-reasoner","system_fingerprint":"fp_eaab8d114b_prod0820_fp8_kvcache_new_kvcache","choices":[{"index":0,"delta":{"content":null,"reasoning_content":"友好"},"logprobs":null,"finish_reason":null}]}
  
  
    就是保持友好

• 泽辉呀 回复 我最可爱ya #2

  测试了的确是，我优化下，感谢反馈