mmdetection3d: local variable ‘beta1‘ referenced before assignment

报错就不截图了，gogole后发现在pytorch的github上发现是AdamW优化器的一个bug，git #52944
原问题地址
[optim] bugfix when all parameters have no grad
已经有大佬解决了，不太会用git，所以只能手动去改源文件
先找到bug报错的adamW优化器源代码，需要改的地方就三个文件，每个文件分别把对应的那行代码挪一下：

1. …/torch/optim/adadelta.py

52        for group in self.param_groups:
53            params_with_grad = []
54            grads = []
55            square_avgs = []
56            acc_deltas = []
57            lr, rho, eps, weight_decay = group['lr'], group['rho'], group['eps'], group['weight_decay']  #changed bug 52944 ++加这一行
...
78					#lr, rho, eps, weight_decay = group['lr'], group['rho'], group['eps'], group['weight_decay'] #changed bug 52944 --注释掉这一行

2. …/torch/optim/adam.py

68        for group in self.param_groups:
69            params_with_grad = []
70            grads = []
71            exp_avgs = []
72            exp_avg_sqs = []
73            state_sums = []
74            max_exp_avg_sqs = []
75            state_steps = []
76            beta1, beta2 = group['betas']  #changed bug 52944 ++加这一行
...
108            	#beta1, beta2 = group['betas']  #changed bug 52944 --注释掉这一行

3. …/torch/optim/adamw.py

67        for group in self.param_groups:
68            params_with_grad = []
69            grads = []
70            exp_avgs = []
71            exp_avg_sqs = []
72            state_sums = []
73            max_exp_avg_sqs = []
74            state_steps = []
75            amsgrad = group['amsgrad']
76            beta1, beta2 = group['betas'] # changed bug52944 ++加这一行
...
105           	#beta1, beta2 = group['betas'] #changed bug52944 -- 注释掉这一行