Memory accumulates when calling a matlab function multiple times using matlab.engine in python.

eng = matlab.engine.start_matlab()
# 设置路径到含有MRSTR.m文件的目录
matlab_path = os.path.join(rootpath, r'Matlab\2_D_oil_Water')
eng.addpath(matlab_path, nargout=0)
##函数中的一部分
# 调用 MATLAB 函数并计算损失
    loss2, grad2 = eng.Physics_loss(m_inputs, m_outputs, m_c, nargout=2)
    # eng.workspace.clear()
    eng.eval('clear m_inputs m_outputs m_c', nargout=0)
    eng.eval('clearvars', nargout=0)
    # 清除 MATLAB 工作空间中的变量
    # eng.eval('clearvars -except Physics_loss', nargout=0)
    # 数据转化为 tensor
    physics_grad = torch.tensor(grad2).cuda()

    physics_loss = torch.tensor(loss2, requires_grad=True).cuda()

# 显式删除不再需要的变量
del outputs, inputs, outputs_m, inputs_m, C_m
del loss2, m_inputs, m_outputs, m_c, grad2
gc.collect()
torch.cuda.empty_cache()
即使不断的删除内存，只要多调用一次 eng.Physics_loss 内存就会累加
loss2, grad2 = eng.Physics_loss(m_inputs, m_outputs, m_c, nargout=2)

[Joss Knight]

What memory? System memory? GPU memory? How are you demonstrating memory growth? What does Physics_loss do?

[佳伟]

Matlab's CPU memory usage has been stable, but python's CPU memory usage has been increasing. Also, this problem does not occur if the assignment is done directly, without calling MATLAB

[佳伟]

Physics_loss is a function that calculates an error

[Piyush Kumar]

@佳伟,

Since the problem does not occur if the assignment is done directly, without calling MATLAB, I suggest you to write the code in Python and then use the pyrunfile function in MATLAB.

If this does not resolve the issue, please share the complete code so that I can try reproducing the issue at my end. I can see that there is no definition of the "Physics_loss" function in the code snippet you have provided. Also, how have you calculated the MATLAB CPU memory usage and Python's CPU memory usage?