代码收藏家技术教程 2025-05-18

PyQt5与llama-cpp-python集成时底层冲突的解析与解决过程

我最近在写一个python应用的时候遇到了一个报错：

2025-05-12 11:36:59 – ERROR – [__main__:minimal_pyqt_llama_test.py:32] – Llama model load FAILED: exception: access violation reading 0x0000000000000000
Traceback (most recent call last):
File "D:\PythonProjects\GameAssistant\minimal_pyqt_llama_test.py", line 28, in attempt_load_llama
llm = Llama(model_path=model_path, n_ctx=n_ctx, verbose=verbose, n_gpu_layers=n_gpu_layers)
File "D:\PythonProjects\GameAssistant\game_assistant_env\lib\site-packages\llama_cpp\llama.py", line 204, in __init__
llama_cpp.llama_backend_init()
OSError: exception: access violation reading 0x0000000000000000

一开始，我认为是我的文件地址写错了，于是我反复对比D:/models/Phi-3-mini-4k-instruct-q4.gguf却发现确实没有问题，那就很费解了，为什么会出现这种情况，于是我反复测试反复测试，在经过长达2天的测试后，我决定先看看底层之间是否有冲突，于是我编写了两个测试文件分别是：

# test_llm_load.py

from llama_cpp import Llama

import logging

logging.basicConfig(level=logging.DEBUG)

logger = logging.getLogger(__name__)

model_path = "D:/models/Phi-3-mini-4k-instruct-q4.gguf"

try:

logger.info(f"Attempting to load model from: {model_path}")

llm = Llama(

model_path=model_path,

n_ctx=512,

n_gpu_layers=0,

verbose=True

)

logger.info("Model loaded successfully!")

except Exception as e:

logger.error(f"Error loading or using Llama model: {e}", exc_info=True)

and

# minimal_pyqt_llama_test.py

import sys

import logging

from PyQt5.QtWidgets import QApplication, QWidget, QVBoxLayout, QPushButton, QTextEdit

# — 日志配置 —

logging.basicConfig(

level=logging.DEBUG,

format='%(asctime)s – %(levelname)s – [%(name)s:%(filename)s:%(lineno)d] – %(message)s',

datefmt='%Y-%m-%d %H:%M:%S'

)

logger = logging.getLogger(__name__)

# — LLM 加载函数 —

def attempt_load_llama():

logger.info("Attempting to load Llama model…")

try:

from llama_cpp import Llama # 确保导入

# — 配置与之前测试一致 —

model_path = "D:/models/Phi-3-mini-4k-instruct-q4.gguf" # <— 再次确认你的路径

n_ctx = 512

verbose = True

n_gpu_layers = 0

logger.info(f"Params – Path: '{model_path}', Ctx: {n_ctx}, Verbose: {verbose}, GPU Layers: {n_gpu_layers}")

logger.info("About to call Llama(…) constructor.")

llm = Llama(model_path=model_path, n_ctx=n_ctx, verbose=verbose, n_gpu_layers=n_gpu_layers)

logger.info("Llama model loaded successfully in this minimal test!")

return llm # 返回实例以便后续可能的操作

except Exception as e:

logger.error(f"Llama model load FAILED: {e}", exc_info=True)

return None

# — PyQt 应用 —

if __name__ == '__main__':

logger.info("Minimal PyQt Llama Test Application starting…")

# 1. 尝试在 QApplication 初始化前加载 (通常不推荐，但为了测试)

# logger.info("— Test 1: Loading Llama BEFORE QApplication —")

# pre_app_llm = attempt_load_llama()

# if pre_app_llm:

# logger.info("Test 1: Load BEFORE QApplication SUCCEEDED.")

# pre_app_llm = None # 释放

# else:

# logger.error("Test 1: Load BEFORE QApplication FAILED.")

# logger.info("— End of Test 1 —")

# 2. 初始化 QApplication

app = QApplication(sys.argv)

logger.info("QApplication initialized.")

# 3. 尝试在 QApplication 初始化后，创建任何窗口前加载

logger.info("— Test 2: Loading Llama AFTER QApplication, BEFORE window —")

post_app_llm = attempt_load_llama()

if post_app_llm:

logger.info("Test 2: Load AFTER QApplication SUCCEEDED.")

# post_app_llm = None # 释放

else:

logger.error("Test 2: Load AFTER QApplication FAILED.")

logger.info("— End of Test 2 —")

# 4. 创建一个简单的窗口并显示

main_window = QWidget()

main_window.setWindowTitle("Minimal PyQt Llama Test")

layout = QVBoxLayout()

log_display = QTextEdit()

log_display.setReadOnly(True)

layout.addWidget(log_display)

# 将日志也输出到 QTextEdit (可选，但方便查看)

class QTextEditLogger(logging.Handler):

def __init__(self, text_edit_widget):

super().__init__()

self.widget = text_edit_widget

def emit(self, record):

msg = self.format(record)

self.widget.append(msg)

text_edit_handler = QTextEditLogger(log_display)

text_edit_handler.setFormatter(logging.Formatter('%(asctime)s – %(levelname)s – %(message)s'))

logging.getLogger().addHandler(text_edit_handler) # 添加到根 logger

if post_app_llm: # 如果 Test 2 成功，可以尝试用它做点事

test_button = QPushButton("Test LLM Inference (if loaded)")

def on_test_button_click():

if post_app_llm:

try:

prompt_test = "<|system|>\nHello<|user|>\nWhat is 1+1?<|assistant|>"

logger.info(f"Sending prompt to LLM: {prompt_test[:30]}…")

output = post_app_llm(prompt_test, max_tokens=10)

response = output["choices"][0]["text"].strip()

logger.info(f"LLM Response: {response}")

log_display.append(f"LLM Response: {response}")

except Exception as e_infer:

logger.error(f"LLM inference error: {e_infer}", exc_info=True)

else:

logger.warning("LLM not loaded, cannot run inference.")

test_button.clicked.connect(on_test_button_click)

layout.addWidget(test_button)

main_window.setLayout(layout)

main_window.resize(600, 400)

main_window.show()

logger.info("Minimal PyQt App: Entering event loop…")

sys.exit(app.exec_())

这两个一个是大模型的单独测试，一个是在pyQt上进行的大模型的测试，测试结果是test_llm_load.py测试成功而minimal_pyqt_llama_test.py测试失败，这已经明显的指示出来时pyQt和llama-cpp-python底层库之间的冲突问题了。

报错日志：

2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:37] – Minimal PyQt Llama Test Application starting…
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:52] – QApplication initialized.
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:55] – — Test 2: Loading Llama AFTER QApplication, BEFORE window —
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:16] – Attempting to load Llama model…
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:25] – Params – Path: 'D:/models/Phi-3-mini-4k-instruct-q4.gguf', Ctx: 512, Verbose: True, GPU Layers: 0
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:26] – About to call Llama(…) constructor.
2025-05-12 12:05:12 – ERROR – [__main__:minimal_pyqt_llama_test.py:32] – Llama model load FAILED: exception: access violation reading 0x0000000000000000
Traceback (most recent call last):
File "D:\PythonProjects\GameAssistant\minimal_pyqt_llama_test.py", line 28, in attempt_load_llama
llm = Llama(model_path=model_path, n_ctx=n_ctx, verbose=verbose, n_gpu_layers=n_gpu_layers)
File "D:\PythonProjects\GameAssistant\game_assistant_env\lib\site-packages\llama_cpp\llama.py", line 204, in __init__
llama_cpp.llama_backend_init()
OSError: exception: access violation reading 0x0000000000000000
2025-05-12 12:05:12 – ERROR – [__main__:minimal_pyqt_llama_test.py:61] – Test 2: Load AFTER QApplication FAILED.
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:62] – — End of Test 2 —
2025-05-12 12:05:12 – INFO – [__main__:minimal_pyqt_llama_test.py:110] – Minimal PyQt App: Entering event loop…