MINT-Quantization/spike_related.py at main · Intelligent-Computing-Lab-Panda/MINT-Quantization · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from training_utils import *
import tracemalloc
import gc

class ZIF(torch.autograd.Function):
    @staticmethod
    def forward(ctx, input):
        out = (input > 0).float()
        # L = torch.tensor([gamma])
        ctx.save_for_backward(input)

        return out

    @staticmethod
    def backward(ctx, grad_output):
        input, = ctx.saved_tensors
        grad_input = grad_output.clone()

        grad_input = grad_input * ((1.0 - torch.abs(input)).clamp(min=0))

        # tmp = torch.ones_like(input)
        # tmp = torch.where(input.abs() < 0.5, 1., 0.)
        # grad_input = grad_input*torch.where(torch.abs(input-th)<1., 1, 0)
        # grad_input = grad_input*torch.where(input.abs() < 1., 1., 0.)
        return grad_input


class LIFSpike(nn.Module):
    def __init__(self, thresh=0.5, leak=0.5, gamma=1.0, soft_reset=True, quant_u=False, num_bits_u=4):
        """
        Implementing the LIF neurons.
        @param thresh: firing threshold;
        @param tau: membrane potential decay factor;
        @param gamma: hyper-parameter for controlling the sharpness in surrogate gradient;
        @param soft_reset: whether using soft-reset or hard-reset.
        """
        super(LIFSpike, self).__init__()

        # self.act = ZIF.apply
        self.quant_u = quant_u
        self.num_bits_u = num_bits_u

        self.thresh = thresh
        self.leak = leak
        self.gamma = gamma
        self.soft_reset = soft_reset

        self.membrane_potential = 0
        # print(self.thresh)

    def reset_mem(self):
        self.membrane_potential = 0

    def forward(self, s, share, beta, bias):
        # act = ZIF.apply
        # if self.training:
        # beta*
        # x = gamma_*x + bias
        H = s + self.membrane_potential

        # s = act(H-self.thresh)
        grad = ((1.0 - torch.abs(H-self.thresh)).clamp(min=0))
        s = (((H-self.thresh) > 0).float() - H*grad).detach() + H*grad.detach()
        # s = (H -(self.thresh/beta)>0).float()
        if self.soft_reset:
            U = (H - s*self.thresh)*self.leak
        else:
            U = H*self.leak*(1-s)

        if self.quant_u:
            if share:
                self.membrane_potential = u_q(U,self.num_bits_u,beta)
            else:
                self.membrane_potential= b_q(U,self.num_bits_u)
        else:
            self.membrane_potential = U
        # else:
        #     # print(torch.unique(s).shape)

        #     H = s + self.membrane_potential
        #     # print(torch.unique(H).shape)
        #     if share:
        #         # s = torch.zeros_like(H).cuda()
        #         # s[H >(self.thresh/beta-bias)] = 1.0
        #         s = (H -(self.thresh/beta-bias)>0).float()
        #     else:
        #         s = ((H-self.thresh) > 0).float()
        #     # print(torch.unique(H).shape)
        #     if self.soft_reset:
        #         U = (H - s*self.thresh)*self.leak
        #     else:
        #         U = H*self.leak*(1-s)

        #     # if self.quant_u:
        #     #     if share:
        #     #         self.membrane_potential = (U).round().clamp(min=2**(-(self.num_bits_u-1)),max=2**(self.num_bits_u-1)-1)
        #     # # if self.quant_u:
        #     # #     if share:
        #     # #         self.membrane_potential,_ = b_q_inference(U,self.num_bits_u)
        #     # #     else:
        #     # #         self.membrane_potential = b_q_inference(U,self.num_bits_u)
        #     # else:
        #     self.membrane_potential = U

        return s

    def direct_forward(self, s, share, beta):
        # act = ZIF.apply
        # if self.training:
        # beta*
        # x = gamma_*x + bias
        H = s + self.membrane_potential

        # s = act(H-self.thresh)
        grad = ((1.0 - torch.abs(H-self.thresh)).clamp(min=0))
        s = (((H-self.thresh) > 0).float() - H*grad).detach() + H*grad.detach()
        if self.soft_reset:
            U = (H - s*self.thresh)*self.leak
        else:
            U = H*self.leak*(1-s)

        if self.quant_u:
            if share:
                self.membrane_potential = u_q(U,self.num_bits_u,beta)
            else:
                self.membrane_potential= b_q(U,self.num_bits_u)
        else:
            self.membrane_potential = U

        return s