Spiking transformers combine attention-based representation learning with sparse spike communication, yet their weights and membrane variables still incur substantial storage and computation overhead. Post-training quantization (PTQ) offers a practical way to reduce these overheads, but preserving neuronal firing under weight and membrane quantization remains challenging. In this paper, we propose Decision-Aligned Post-Training Quantization (DAQ), a method that calibrates the quantizers to align neuronal firing decisions with those of a floating-point reference. We analyze how errors from these quantizers affect firing at decision boundaries and how they propagate over time steps in recurrent neurons. Based on this analysis, we formulate an objective that encourages quantized membrane variables to remain within the reference firing regions along the quantized model's own state trajectory. Experiments across multiple spiking transformer architectures on image classification, semantic segmentation, and event-based tracking demonstrate the effectiveness of DAQ with 4-bit weights and membrane variables. On FE108, DAQ achieves 58.8% AUC and 90.9% PR with SDTrack, while reducing energy by 66.2%.
Figures & tables
Figure 1: Accuracy versus energy on ImageNet-1K. DAQ achieves 83.4% top-1 accuracy with 4-bit weights and membrane variables (W4/U4). Bubble area represents parameter storage.
Figure 2: Overview of DAQ. The frozen floating-point model provides reference firing decisions, while the quantized model follows its own state trajectory. At shared sampled positions, a weighted decision-aligned loss encourages quantized potentials to remain within the reference decision regions. Event losses are averaged within and across groups to calibrate the weight and membrane quantizers.
Figure 3: Decision boundaries and temporal propagation of quantization errors. (a) Firing preservation depends on the reference potential and the direction of the perturbation relative to the decision boundary. (b) Matching non-firing outputs can leave different post-reset states and lead to a later firing mismatch.
Method
T×D
Precision
Energy
FE108
FELT
VisEvent
(mJ) ↓
AUC (%) ↑
PR (%) ↑
AUC (%) ↑
PR (%) ↑
AUC (%) ↑
PR (%) ↑
STARK [ 30 ]
1×1
FP
58.88
57.4
89.2
39.6
51.7
34.1
46.8
SimTrack [ 3 ]
1×1
FP
93.84
56.7
88.3
36.8
47.0
34.6
47.6
OSTrack 256 [ 35 ]
1×1
FP
98.90
54.6
87.1
35.9
45.5
32.7
46.4
ARTrack 256 [ 26 ]
1×1
FP
174.8
56.6
88.5
39.5
49.4
33.0
43.8
SeqTrack-B 256 [ 4 ]
1×1
FP
302.7
53.5
85.5
33.0
42.0
28.6
43.3
Table 1: Event-based object tracking results on FE108, FELT, and VisEvent. Ratios beside TP-Spikformer indicate average token retention. The three best distinct values per metric are highlighted as first , second , and third . Ties share the same highlight; a dash denotes an unavailable result. T denotes the number of time steps, and D denotes the maximum spike count per time step.
Method
Bits (W/U)
T×D
Param. (M)
Energy (mJ)
Top-1 (%)
Spikformer [ 42 ]
32/32
4×1
66.3
21.5
74.8
SDT-V1 [ 33 ]
32/32
4×1
66.3
6.1
77.1
Spikingformer [ 40 ]
32/32
4×1
66.3
16.3
77.6
SpikingResformer [ 22 ]
32/32
4×1
60.4
14.8
79.4
SNN-ViT [ 23 ]
32/32
4×1
53.7
35.8
80.2
SDT-V2 [ 32 ]
32/32
4×1
55.4
52.4
79.7
Table 2: ImageNet-1K classification results. Parameter storage is expressed in millions of FP32-equivalent parameters. ∗ denotes results reproduced by us with W4/U4 adaptations of the official implementations.
Method
Bits
CIFAR-10
CIFAR-100
SparseSpikformer ( ×0.63 ) [ 14 ]
32/32
94.8
76.8
AT-SNN ( ×0.28 ) [ 9 ]
32/32
95.1
78.1
STATA ( ×0.50 ) [ 43 ]
32/32
95.0
77.7
TP-Spikformer ( ×0.25 ) [ 25 ]
32/32
95.2
78.5
BRECQ ∗ [ 11 ]
4/4
94.1
73.2
PTQ4SNN † [ 29 ]
4/4
94.4
75.3
Table 3: Top-1 accuracy (%) on CIFAR-10 and CIFAR-100 at T=4 . † uses 4-bit weights and mixed-precision membrane quantization under an average 4-bit membrane budget. Numbers in method names denote token retention ratios.
Method
T×D
Bits
Param.
Energy
mIoU
SDT-V2 [ 32 ]
4×1
FP
16.5
88.1
33.6
4×1
FP
59.8
183.6
35.3
SDT-V3 [ 34 ]
1×4
FP
11.0
27.2
40.1
QSD (V2-T) [ 20 ]
4×1
W4
3.3
17.5
39.0
QSD (V2-M)
4×1
W4
9.6
37.9
40.5
TP-Spikformer ( ×0.78 ) [ 25 ]
1×4
FP
11.0
–
40.2
Table 4: Semantic segmentation results on ADE20K.
Figure 4: Qualitative comparison of FP32 and DAQ (W4/U4) for (a) semantic segmentation and (b) event-based object tracking. Ground truth is shown for reference.
Objective
L↓
Rw↓
R↓
Top-1 (%)
MSE
5.83
1.79
0.39
74.05±0.16
DAQ
5.46
1.70
0.36
75.98±0.11
Table 5: Comparison of calibration objectives on SDT-V3-10M. L is the DAQ loss, Rw is the weighted firing error, and R is the unweighted squared count error.
Model
Configuration
Top-1 (%)
SDT-8-768
Initial grids
61.87±0.27
DAQ without distance weighting
73.32±0.24
DAQ with reference states
73.74±0.04
Full DAQ
73.97±0.10
SDT-V3-10M
Initial grids
68.44±0.17
DAQ without distance weighting
75.90±0.08
Table 6: Ablation study on ImageNet-1K under W4/U4 quantization. Values are reported as mean ± sample standard deviation.
Storage (MiB)
DAQ time
Backbone
WFP/UFP
W4/ UFP
W4/U4
(GPU min)
SDT-V3-10M
40.5
7.3
5.6
5.9
SDT-V3-19M
75.5
12.4
10.1
6.3
SDT-V3-171M
664.5
92.7
85.7
8.3
SDT-8-768
314.3
93.4
47.5
19.6
Table 7: Packed parameter and membrane buffer storage at batch size one, and W4/U4 quantization time on ImageNet-1K. WFP and UFP denote the original floating-point precisions.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
ViT PTQ (W/A)
Method
Bits
DeiT-T
DeiT-S
Swin-S
ViT-B
Full-Precision
32/32
72.71
79.85
83.23
84.54
PTQ4ViT [ 36 ]
4/4
36.96
34.08
76.09
30.69
APQ-ViT [ 5 ]
4/4
47.94
43.55
77.15
41.41
RepQ-ViT [ 12 ]
4/4
57.43
69.03
79.45
68.48
AdaLog [ 27 ]
4/4
63.52
72.06
80.77
79.68
Appendix
Table 8: Four-bit PTQ results on ImageNet-1K. All entries report top-1 accuracy (%). W/A and W/U denote weight/activation and weight/membrane bit widths, respectively. ViT results follow the comparison in FIMA-Q [ 28 ] , with UQ-ViT results from [ 8 ] . All SDT-V3 models use T×D=1×4 .
State Key Laboratory of Complex & Critical Software Environment, Beihang University · School of Artificial Intelligence, Beihang University · Center for Project-Based Learning, D-ITET, ETH Zurich +2