github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9549] 502 Bad Gateway with stream=True using ollama and httpx on Windows (v0.5.7) #6228

New Issue
Closed
opened 2026-04-12 17:38:28 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 17:38:28 -05:00
Owner
Copy Link

Originally created by @lujixiang on GitHub (Mar 6, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9549

What is the issue?

描述：
服务版本：0.5.7
库版本：ollama 0.4.7, httpx
环境：Windows
问题：流式请求报 502，非流式正常；requests 流式正常。
重现步骤：提供脚本和输出。

Relevant log output

import requests
import json

def test_non_stream():
    print("测试非流式请求...")
    url = "http://localhost:11434/api/generate"
    data = {
        "model": "qwen2.5:7b",
        "prompt": "Hello",
        "stream": False
    }
    response = requests.post(url, json=data)
    print(f"状态码: {response.status_code}")
    print(f"响应内容: {response.json()}")

def test_stream():
    print("\n测试流式请求...")
    url = "http://localhost:11434/api/generate"
    data = {
        "model": "qwen2.5:7b",
        "prompt": "Hello",
        "stream": True
    }
    response = requests.post(url, json=data, stream=True)
    print(f"状态码: {response.status_code}")
    full_response = ""
    for line in response.iter_lines():
        if line:
            chunk = json.loads(line.decode('utf-8'))
            print(f"原始响应: {line.decode('utf-8')}")
            if "response" in chunk:
                full_response += chunk["response"]
    print(f"完整回答: {full_response}")

if __name__ == "__main__":
    test_non_stream()
    test_stream()
python test_ollama.py
200 {'model': 'qwen2.5:7b', 'created_at': '2025-03-06T13:56:52.9099449Z', 'response': 'Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything specific.', 'done': True, 'done_reason': 'stop', 'context': [151644, 8948, 198, 2610, 525, 1207, 16948, 11, 3465, 553, 54364, 14817, 13, 1446, 525, 264, 10950, 17847, 13, 151645, 198, 151644, 872, 198, 9707, 151645, 198, 151644, 77091, 198, 9707, 0, 2585, 646, 358, 7789, 498, 3351, 30, 31733, 1910, 311, 2548, 752, 894, 4755, 476, 1077, 752, 1414, 421, 498, 1184, 1492, 448, 4113, 3151, 13], 'total_duration': 20852036000, 'load_duration': 2624889800, 'prompt_eval_count': 30, 'prompt_eval_duration': 5905000000, 'eval_count': 29, 'eval_duration': 12319000000}
PS G:\ChineseMedicine\RAGQnASystem> python test_ollama.py
测试非流式请求...
状态码: 200
响应内容: {'model': 'qwen2.5:7b', 'created_at': '2025-03-06T14:00:50.5144825Z', 'response': 'Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything.', 'done': True, 'done_reason': 'stop', 'context': [151644, 8948, 198, 2610, 525, 1207, 16948, 11, 3465, 553, 54364, 14817, 13, 1446, 525, 264, 10950, 17847, 13, 151645, 198, 151644, 872, 198, 9707, 151645, 198, 151644, 77091, 198, 9707, 0, 2585, 646, 358, 7789, 498, 3351, 30, 31733, 1910, 311, 2548, 752, 894, 4755, 476, 1077, 752, 1414, 421, 498, 1184, 1492, 448, 4113, 13], 'total_duration': 9805980600, 'load_duration': 19316100, 'prompt_eval_count': 30, 'prompt_eval_duration': 545000000, 'eval_count': 28, 'eval_duration': 9240000000}

测试流式请求...
状态码: 200
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.133863Z","response":"Hello","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.5364512Z","response":"!","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.9203841Z","response":" Nice","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.3023129Z","response":" to","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.4141542Z","response":" meet","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.5314663Z","response":" you","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.6494147Z","response":".","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.774694Z","response":" How","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.8839354Z","response":" can","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:55.3037153Z","response":" I","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:55.7045539Z","response":" assist","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.1183736Z","response":" you","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.5583607Z","response":" today","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.8107199Z","response":"?","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.2806849Z","response":" Whether","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.4403473Z","response":" you","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.8858919Z","response":" have","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.0407432Z","response":" questions","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.4702173Z","response":",","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.941108Z","response":" need","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.1471744Z","response":" information","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.3500194Z","response":",","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.818258Z","response":" or","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.3007422Z","response":" just","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.7110117Z","response":" want","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.8368074Z","response":" to","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.9453674Z","response":" chat","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:01.3350672Z","response":",","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:01.7314499Z","response":" feel","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:02.1333104Z","response":" free","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:02.6028868Z","response":" to","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:03.0893268Z","response":" let","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:03.6624977Z","response":" me","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:04.1405238Z","response":" know","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:04.5443316Z","response":".","done":false}
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:05.017529Z","response":"","done":true,"done_reason":"stop","context":[151644,8948,198,2610,525,1207,16948,11,3465,553,54364,14817,13,1446,525,264,10950,17847,13,151645,198,151644,872,198,9707,151645,198,151644,77091,198,9707,0,28859,311,3367,498,13,2585,646,358,7789,498,3351,30,13139,498,614,4755,11,1184,1995,11,476,1101,1366,311,6236,11,2666,1910,311,1077,752,1414,13],"total_duration":12472752600,"load_duration":16525600,"prompt_eval_count":30,"prompt_eval_duration":569000000,"eval_count":36,"eval_duration":11886000000}
完整回答: Hello! Nice to meet you. How can I assist you today? Whether you have questions, need information, or just want to chat, feel free to let me know.
使用之前的脚本（包含 ollama 和 httpx
import ollama
import httpx
import json

def test_ollama_stream():
    print("测试 ollama 库流式请求...")
    client = ollama.Client(host="http://localhost:11434")
    print("请求服务: http://localhost:11434/api/generate")
    print("请求参数: model='qwen2.5:7b', prompt='Hello', stream=True")
    
    try:
        full_response = ""
        for chunk in client.generate(model="qwen2.5:7b", prompt="Hello", stream=True):
            if "response" in chunk:
                full_response += chunk["response"]
                print(f"部分回答: {chunk['response']}")
            elif "error" in chunk:
                print(f"服务返回错误: {chunk['error']}")
                break
            if chunk.get("done", False):
                break
        print(f"完整回答: {full_response}")
    except ollama.ResponseError as e:
        print(f"服务错误: {e.status_code} - {e.error}")
        print(f"异常详情: {str(e)}")

def test_httpx_stream():
    print("\n测试 httpx 流式请求...")
    url = "http://localhost:11434/api/generate"
    data = {"model": "qwen2.5:7b", "prompt": "Hello", "stream": True}
    headers = {"Content-Type": "application/json", "User-Agent": "curl/8.1.2"}
    
    try:
        with httpx.stream("POST", url, json=data, headers=headers, timeout=30) as response:
            print(f"响应状态码: {response.status_code}")
            full_response = ""
            for line in response.iter_lines():
                if line:
                    chunk = json.loads(line)
                    print(f"原始响应: {line}")
                    if "response" in chunk:
                        full_response += chunk["response"]
                        print(f"部分回答: {chunk['response']}")
                    if chunk.get("done", False):
                        break
            print(f"完整回答: {full_response}")
    except httpx.HTTPStatusError as e:
        print(f"服务错误: {e.response.status_code} - {e.response.text}")
    except Exception as e:
        print(f"未知错误: {str(e)}")

if __name__ == "__main__":
    test_ollama_stream()
    test_httpx_stream()
python test_ollama.py
测试 ollama 库流式请求...
请求服务: http://localhost:11434/api/generate
请求参数: model='qwen2.5:7b', prompt='Hello', stream=True
服务错误: 502 - 
异常详情:  (status code: 502)

测试 httpx 流式请求...
响应状态码: 502
完整回答:


从你提供的运行结果来看，执行 python test_ollama.py 时，使用 ollama 库和 httpx 的流式请求（stream=True）都返回了 502 Bad Gateway，并且响应内容为空。这与之前使用 requests 库成功的情况形成对比。以下是详细分析和最终解决方案。关键点
ollama 库请求：
调用 ollama.generate（stream=True）返回 502。
e.error 为空（服务错误: 502 - ）。
异常详情仅显示状态码，无额外信息。
httpx 流式请求：
状态码 502，无响应数据（完整回答: 为空）。
未显示响应头（���能脚本未完全打印）。
对比之前：
使用 requests（stream=True 和 stream=False）都返回 200，成功生成回答。
ollama 和 httpx 的流式请求持续失败。
已知信息
服务运行：curl http://localhost:11434/ 返回 "Ollama is running"。
模型可用：ollama run qwen2.5:7b 生成回答。
API 测试：curl -X POST 返回 JSON。
requests 测试：流式和非流式请求成功。
版本：服务 0.5.7，库 0.4.7。
问题分析
1. 当前失败的原因
流式请求问题：
ollama 库和 httpx 的流式请求（stream=True）触发了服务端的 502。
服务端可能无法正确处理这些客户端的流式请求格式或连接方式。
客户端差异：
requests 的流式实现（stream=True + iter_lines）与服务端兼容。
httpx 的流式实现（httpx.stream）和 ollama 库的内部实现可能不兼容。
2. 成功与失败的对比
成功：
requests（流式和非流式）工作正常。
curl（流式和非流式）工作正常。
失败：
ollama 和 httpx 的流式请求报 502。
结论：
问题出在 ollama 和 httpx 的流式请求实现，与服务端 0.5.7 在 Windows 环境下的兼容性有关。
3. 可能的根源
服务端 bug：
0.5.7 在 Windows 上对某些流式请求处理异常。
客户端配置：
httpx 和 ollama 的默认超时、头部或连接管理可能触发服务端问题。
模型加载：
虽然 requests 成功，但未预加载模型时可能影响其他客户端。
解决步骤
1. 最终解决方案
使用 requests 代替 ollama 和 httpx：
当前环境下，requests 是唯一稳定支持流式和非流式请求的库。

OS

Windows

GPU

Nvidia

CPU

AMD

Ollama version

版本：服务 0.5.7，库 0.4.7。

Originally created by @lujixiang on GitHub (Mar 6, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9549 ### What is the issue? 描述： 服务版本：0.5.7 库版本：ollama 0.4.7, httpx 环境：Windows 问题：流式请求报 502，非流式正常；requests 流式正常。 重现步骤：提供脚本和输出。 ### Relevant log output ```shell import requests import json def test_non_stream(): print("测试非流式请求...") url = "http://localhost:11434/api/generate" data = { "model": "qwen2.5:7b", "prompt": "Hello", "stream": False } response = requests.post(url, json=data) print(f"状态码: {response.status_code}") print(f"响应内容: {response.json()}") def test_stream(): print("\n测试流式请求...") url = "http://localhost:11434/api/generate" data = { "model": "qwen2.5:7b", "prompt": "Hello", "stream": True } response = requests.post(url, json=data, stream=True) print(f"状态码: {response.status_code}") full_response = "" for line in response.iter_lines(): if line: chunk = json.loads(line.decode('utf-8')) print(f"原始响应: {line.decode('utf-8')}") if "response" in chunk: full_response += chunk["response"] print(f"完整回答: {full_response}") if __name__ == "__main__": test_non_stream() test_stream() python test_ollama.py 200 {'model': 'qwen2.5:7b', 'created_at': '2025-03-06T13:56:52.9099449Z', 'response': 'Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything specific.', 'done': True, 'done_reason': 'stop', 'context': [151644, 8948, 198, 2610, 525, 1207, 16948, 11, 3465, 553, 54364, 14817, 13, 1446, 525, 264, 10950, 17847, 13, 151645, 198, 151644, 872, 198, 9707, 151645, 198, 151644, 77091, 198, 9707, 0, 2585, 646, 358, 7789, 498, 3351, 30, 31733, 1910, 311, 2548, 752, 894, 4755, 476, 1077, 752, 1414, 421, 498, 1184, 1492, 448, 4113, 3151, 13], 'total_duration': 20852036000, 'load_duration': 2624889800, 'prompt_eval_count': 30, 'prompt_eval_duration': 5905000000, 'eval_count': 29, 'eval_duration': 12319000000} PS G:\ChineseMedicine\RAGQnASystem> python test_ollama.py 测试非流式请求... 状态码: 200 响应内容: {'model': 'qwen2.5:7b', 'created_at': '2025-03-06T14:00:50.5144825Z', 'response': 'Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything.', 'done': True, 'done_reason': 'stop', 'context': [151644, 8948, 198, 2610, 525, 1207, 16948, 11, 3465, 553, 54364, 14817, 13, 1446, 525, 264, 10950, 17847, 13, 151645, 198, 151644, 872, 198, 9707, 151645, 198, 151644, 77091, 198, 9707, 0, 2585, 646, 358, 7789, 498, 3351, 30, 31733, 1910, 311, 2548, 752, 894, 4755, 476, 1077, 752, 1414, 421, 498, 1184, 1492, 448, 4113, 13], 'total_duration': 9805980600, 'load_duration': 19316100, 'prompt_eval_count': 30, 'prompt_eval_duration': 545000000, 'eval_count': 28, 'eval_duration': 9240000000} 测试流式请求... 状态码: 200 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.133863Z","response":"Hello","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.5364512Z","response":"!","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:53.9203841Z","response":" Nice","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.3023129Z","response":" to","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.4141542Z","response":" meet","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.5314663Z","response":" you","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.6494147Z","response":".","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.774694Z","response":" How","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:54.8839354Z","response":" can","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:55.3037153Z","response":" I","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:55.7045539Z","response":" assist","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.1183736Z","response":" you","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.5583607Z","response":" today","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:56.8107199Z","response":"?","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.2806849Z","response":" Whether","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.4403473Z","response":" you","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:57.8858919Z","response":" have","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.0407432Z","response":" questions","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.4702173Z","response":",","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:58.941108Z","response":" need","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.1471744Z","response":" information","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.3500194Z","response":",","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:00:59.818258Z","response":" or","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.3007422Z","response":" just","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.7110117Z","response":" want","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.8368074Z","response":" to","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:00.9453674Z","response":" chat","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:01.3350672Z","response":",","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:01.7314499Z","response":" feel","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:02.1333104Z","response":" free","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:02.6028868Z","response":" to","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:03.0893268Z","response":" let","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:03.6624977Z","response":" me","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:04.1405238Z","response":" know","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:04.5443316Z","response":".","done":false} 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T14:01:05.017529Z","response":"","done":true,"done_reason":"stop","context":[151644,8948,198,2610,525,1207,16948,11,3465,553,54364,14817,13,1446,525,264,10950,17847,13,151645,198,151644,872,198,9707,151645,198,151644,77091,198,9707,0,28859,311,3367,498,13,2585,646,358,7789,498,3351,30,13139,498,614,4755,11,1184,1995,11,476,1101,1366,311,6236,11,2666,1910,311,1077,752,1414,13],"total_duration":12472752600,"load_duration":16525600,"prompt_eval_count":30,"prompt_eval_duration":569000000,"eval_count":36,"eval_duration":11886000000} 完整回答: Hello! Nice to meet you. How can I assist you today? Whether you have questions, need information, or just want to chat, feel free to let me know. 使用之前的脚本（包含 ollama 和 httpx import ollama import httpx import json def test_ollama_stream(): print("测试 ollama 库流式请求...") client = ollama.Client(host="http://localhost:11434") print("请求服务: http://localhost:11434/api/generate") print("请求参数: model='qwen2.5:7b', prompt='Hello', stream=True") try: full_response = "" for chunk in client.generate(model="qwen2.5:7b", prompt="Hello", stream=True): if "response" in chunk: full_response += chunk["response"] print(f"部分回答: {chunk['response']}") elif "error" in chunk: print(f"服务返回错误: {chunk['error']}") break if chunk.get("done", False): break print(f"完整回答: {full_response}") except ollama.ResponseError as e: print(f"服务错误: {e.status_code} - {e.error}") print(f"异常详情: {str(e)}") def test_httpx_stream(): print("\n测试 httpx 流式请求...") url = "http://localhost:11434/api/generate" data = {"model": "qwen2.5:7b", "prompt": "Hello", "stream": True} headers = {"Content-Type": "application/json", "User-Agent": "curl/8.1.2"} try: with httpx.stream("POST", url, json=data, headers=headers, timeout=30) as response: print(f"响应状态码: {response.status_code}") full_response = "" for line in response.iter_lines(): if line: chunk = json.loads(line) print(f"原始响应: {line}") if "response" in chunk: full_response += chunk["response"] print(f"部分回答: {chunk['response']}") if chunk.get("done", False): break print(f"完整回答: {full_response}") except httpx.HTTPStatusError as e: print(f"服务错误: {e.response.status_code} - {e.response.text}") except Exception as e: print(f"未知错误: {str(e)}") if __name__ == "__main__": test_ollama_stream() test_httpx_stream() python test_ollama.py 测试 ollama 库流式请求... 请求服务: http://localhost:11434/api/generate 请求参数: model='qwen2.5:7b', prompt='Hello', stream=True 服务错误: 502 - 异常详情: (status code: 502) 测试 httpx 流式请求... 响应状态码: 502 完整回答: 从你提供的运行结果来看，执行 python test_ollama.py 时，使用 ollama 库和 httpx 的流式请求（stream=True）都返回了 502 Bad Gateway，并且响应内容为空。这与之前使用 requests 库成功的情况形成对比。以下是详细分析和最终解决方案。关键点 ollama 库请求： 调用 ollama.generate（stream=True）返回 502。 e.error 为空（服务错误: 502 - ）。 异常详情仅显示状态码，无额外信息。 httpx 流式请求： 状态码 502，无响应数据（完整回答: 为空）。 未显示响应头（���能脚本未完全打印）。 对比之前： 使用 requests（stream=True 和 stream=False）都返回 200，成功生成回答。 ollama 和 httpx 的流式请求持续失败。 已知信息 服务运行：curl http://localhost:11434/ 返回 "Ollama is running"。 模型可用：ollama run qwen2.5:7b 生成回答。 API 测试：curl -X POST 返回 JSON。 requests 测试：流式和非流式请求成功。 版本：服务 0.5.7，库 0.4.7。 问题分析 1. 当前失败的原因 流式请求问题： ollama 库和 httpx 的流式请求（stream=True）触发了服务端的 502。 服务端可能无法正确处理这些客户端的流式请求格式或连接方式。 客户端差异： requests 的流式实现（stream=True + iter_lines）与服务端兼容。 httpx 的流式实现（httpx.stream）和 ollama 库的内部实现可能不兼容。 2. 成功与失败的对比 成功： requests（流式和非流式）工作正常。 curl（流式和非流式）工作正常。 失败： ollama 和 httpx 的流式请求报 502。 结论： 问题出在 ollama 和 httpx 的流式请求实现，与服务端 0.5.7 在 Windows 环境下的兼容性有关。 3. 可能的根源 服务端 bug： 0.5.7 在 Windows 上对某些流式请求处理异常。 客户端配置： httpx 和 ollama 的默认超时、头部或连接管理可能触发服务端问题。 模型加载： 虽然 requests 成功，但未预加载模型时可能影响其他客户端。 解决步骤 1. 最终解决方案 使用 requests 代替 ollama 和 httpx： 当前环境下，requests 是唯一稳定支持流式和非流式请求的库。 ``` ### OS Windows ### GPU Nvidia ### CPU AMD ### Ollama version 版本：服务 0.5.7，库 0.4.7。
GiteaMirror added the bug label 2026-04-12 17:38:29 -05:00
GiteaMirror commented 2026-04-12 17:38:31 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 6, 2025):

Linux:

$ ./9549.py
测试 ollama 库流式请求...
请求服务: http://localhost:11434/api/generate
请求参数: model='qwen2.5:7b', prompt='Hello', stream=True
部分回答: Hello
部分回答: !
部分回答:  Nice
部分回答:  to
部分回答:  meet
部分回答:  you
部分回答: .
部分回答:  How
部分回答:  can
部分回答:  I
部分回答:  assist
部分回答:  you
部分回答:  today
部分回答: ?
部分回答:  Whether
部分回答:  you
部分回答:  have
部分回答:  questions
部分回答: ,
部分回答:  need
部分回答:  information
部分回答: ,
部分回答:  or
部分回答:  just
部分回答:  want
部分回答:  to
部分回答:  chat
部分回答: ,
部分回答:  feel
部分回答:  free
部分回答:  to
部分回答:  let
部分回答:  me
部分回答:  know
部分回答: .
部分回答: 
完整回答: Hello! Nice to meet you. How can I assist you today? Whether you have questions, need information, or just want to chat, feel free to let me know.

测试 httpx 流式请求...
响应状态码: 200
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.354411281Z","response":"Hello","done":false}
部分回答: Hello
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.366253525Z","response":"!","done":false}
部分回答: !
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.378196334Z","response":" How","done":false}
部分回答:  How
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.391555329Z","response":" can","done":false}
部分回答:  can
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.403485553Z","response":" I","done":false}
部分回答:  I
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.415332411Z","response":" assist","done":false}
部分回答:  assist
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.427167253Z","response":" you","done":false}
部分回答:  you
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.4391184Z","response":" today","done":false}
部分回答:  today
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.451074902Z","response":"?","done":false}
部分回答: ?
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.463017349Z","response":" Feel","done":false}
部分回答:  Feel
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.474858261Z","response":" free","done":false}
部分回答:  free
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.486594985Z","response":" to","done":false}
部分回答:  to
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.49831493Z","response":" ask","done":false}
部分回答:  ask
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.510140178Z","response":" me","done":false}
部分回答:  me
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.521873395Z","response":" any","done":false}
部分回答:  any
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.533659049Z","response":" questions","done":false}
部分回答:  questions
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.54544584Z","response":" or","done":false}
部分回答:  or
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.557086334Z","response":" let","done":false}
部分回答:  let
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.568812606Z","response":" me","done":false}
部分回答:  me
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.580735361Z","response":" know","done":false}
部分回答:  know
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.592486101Z","response":" if","done":false}
部分回答:  if
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.604325581Z","response":" you","done":false}
部分回答:  you
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.616067253Z","response":" need","done":false}
部分回答:  need
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.627863018Z","response":" help","done":false}
部分回答:  help
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.639650489Z","response":" with","done":false}
部分回答:  with
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.651583483Z","response":" anything","done":false}
部分回答:  anything
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.663281511Z","response":" specific","done":false}
部分回答:  specific
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.675183873Z","response":".","done":false}
部分回答: .
原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.687032885Z","response":"","done":true,"done_reason":"stop","context":[151644,8948,198,2610,525,1207,16948,11,3465,553,54364,14817,13,1446,525,264,10950,17847,13,151645,198,151644,872,198,9707,151645,198,151644,77091,198,9707,0,2585,646,358,7789,498,3351,30,31733,1910,311,2548,752,894,4755,476,1077,752,1414,421,498,1184,1492,448,4113,3151,13],"total_duration":630280967,"load_duration":278832620,"prompt_eval_count":30,"prompt_eval_duration":6000000,"eval_count":29,"eval_duration":344000000}
部分回答: 
完整回答: Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything specific.
<!-- gh-comment-id:2704237411 --> @rick-github commented on GitHub (Mar 6, 2025): Linux: ```console $ ./9549.py 测试 ollama 库流式请求... 请求服务: http://localhost:11434/api/generate 请求参数: model='qwen2.5:7b', prompt='Hello', stream=True 部分回答: Hello 部分回答: ! 部分回答: Nice 部分回答: to 部分回答: meet 部分回答: you 部分回答: . 部分回答: How 部分回答: can 部分回答: I 部分回答: assist 部分回答: you 部分回答: today 部分回答: ? 部分回答: Whether 部分回答: you 部分回答: have 部分回答: questions 部分回答: , 部分回答: need 部分回答: information 部分回答: , 部分回答: or 部分回答: just 部分回答: want 部分回答: to 部分回答: chat 部分回答: , 部分回答: feel 部分回答: free 部分回答: to 部分回答: let 部分回答: me 部分回答: know 部分回答: . 部分回答: 完整回答: Hello! Nice to meet you. How can I assist you today? Whether you have questions, need information, or just want to chat, feel free to let me know. 测试 httpx 流式请求... 响应状态码: 200 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.354411281Z","response":"Hello","done":false} 部分回答: Hello 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.366253525Z","response":"!","done":false} 部分回答: ! 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.378196334Z","response":" How","done":false} 部分回答: How 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.391555329Z","response":" can","done":false} 部分回答: can 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.403485553Z","response":" I","done":false} 部分回答: I 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.415332411Z","response":" assist","done":false} 部分回答: assist 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.427167253Z","response":" you","done":false} 部分回答: you 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.4391184Z","response":" today","done":false} 部分回答: today 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.451074902Z","response":"?","done":false} 部分回答: ? 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.463017349Z","response":" Feel","done":false} 部分回答: Feel 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.474858261Z","response":" free","done":false} 部分回答: free 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.486594985Z","response":" to","done":false} 部分回答: to 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.49831493Z","response":" ask","done":false} 部分回答: ask 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.510140178Z","response":" me","done":false} 部分回答: me 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.521873395Z","response":" any","done":false} 部分回答: any 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.533659049Z","response":" questions","done":false} 部分回答: questions 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.54544584Z","response":" or","done":false} 部分回答: or 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.557086334Z","response":" let","done":false} 部分回答: let 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.568812606Z","response":" me","done":false} 部分回答: me 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.580735361Z","response":" know","done":false} 部分回答: know 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.592486101Z","response":" if","done":false} 部分回答: if 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.604325581Z","response":" you","done":false} 部分回答: you 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.616067253Z","response":" need","done":false} 部分回答: need 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.627863018Z","response":" help","done":false} 部分回答: help 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.639650489Z","response":" with","done":false} 部分回答: with 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.651583483Z","response":" anything","done":false} 部分回答: anything 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.663281511Z","response":" specific","done":false} 部分回答: specific 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.675183873Z","response":".","done":false} 部分回答: . 原始响应: {"model":"qwen2.5:7b","created_at":"2025-03-06T15:42:51.687032885Z","response":"","done":true,"done_reason":"stop","context":[151644,8948,198,2610,525,1207,16948,11,3465,553,54364,14817,13,1446,525,264,10950,17847,13,151645,198,151644,872,198,9707,151645,198,151644,77091,198,9707,0,2585,646,358,7789,498,3351,30,31733,1910,311,2548,752,894,4755,476,1077,752,1414,421,498,1184,1492,448,4113,3151,13],"total_duration":630280967,"load_duration":278832620,"prompt_eval_count":30,"prompt_eval_duration":6000000,"eval_count":29,"eval_duration":344000000} 部分回答: 完整回答: Hello! How can I assist you today? Feel free to ask me any questions or let me know if you need help with anything specific. ```
GiteaMirror commented 2026-04-12 17:38:32 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 7, 2025):

Tried it on Win11 24H2 (26100.3194) and it worked fine.

ollama version is 0.5.7

Name: ollama
Version: 0.4.7
Summary: The official Python client for Ollama.
Home-page: https://ollama.com
Author: Ollama
Author-email: hello@ollama.com
License: MIT
Location: C:\Users\bill\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages
Requires: httpx, pydantic
Required-by:
---
Name: httpx
Version: 0.28.1
Summary: The next generation HTTP client.
Home-page: https://github.com/encode/httpx
Author:
Author-email: Tom Christie <tom@tomchristie.com>
License: BSD-3-Clause
Location: C:\Users\bill\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages
Requires: anyio, certifi, httpcore, idna
Required-by: ollama
<!-- gh-comment-id:2706957481 --> @rick-github commented on GitHub (Mar 7, 2025): Tried it on Win11 24H2 (26100.3194) and it worked fine. ``` ollama version is 0.5.7 Name: ollama Version: 0.4.7 Summary: The official Python client for Ollama. Home-page: https://ollama.com Author: Ollama Author-email: hello@ollama.com License: MIT Location: C:\Users\bill\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages Requires: httpx, pydantic Required-by: --- Name: httpx Version: 0.28.1 Summary: The next generation HTTP client. Home-page: https://github.com/encode/httpx Author: Author-email: Tom Christie <tom@tomchristie.com> License: BSD-3-Clause Location: C:\Users\bill\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages Requires: anyio, certifi, httpcore, idna Required-by: ollama ```
GiteaMirror commented 2026-04-12 17:38:34 -05:00
Author
Owner
Copy Link

@chuanxinwong commented on GitHub (Mar 9, 2025):

看看是不是你的环境变量中有 http_proxy=xxxx, 如果有的话， 先执行 set http_proxy= ， 这个环境变量有影响，我之前遇到过多次这种情况，一些sb库依赖环境变量又不说， 出现莫名其妙的问题

<!-- gh-comment-id:2708715779 --> @chuanxinwong commented on GitHub (Mar 9, 2025): 看看是不是你的环境变量中有 http_proxy=xxxx, 如果有的话， 先执行 set http_proxy= ， 这个环境变量有影响，我之前遇到过多次这种情况，一些sb库依赖环境变量又不说， 出现莫名其妙的问题
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#6228
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?