Hardware-Friendly Token-Group Activation Quantization for Low-Bit Mamba Super-Resolution

Jinwoo Chung; Jangho Kim

cs.AIOct 12, 2026

Hardware-Friendly Token-Group Activation Quantization for Low-Bit Mamba Super-Resolution

Authors: Jinwoo Chung, Jangho Kim

Organizations: Kookmin University Seoul, Republic of Korea

Abstract

Mamba-based super-resolution models are efficient alternatives to Transformer-based restoration backbones, but their low-bit post-training quantization remains challenging because selective scan, Mamba's input-dependent state-space operator, induces different activation ranges across spatial tokens. Thus, a single layer-wise clipping range poorly matches all tokens. We propose Token-Group Quantization, a hardware-friendly activation PTQ framework that statically approximates a hardware-unfriendly post-scan grouping oracle, where tokens with similar preferred clipping ranges are assigned to the same group. To avoid runtime post-scan grouping in static low-bit inference, a lightweight predictor infers the groups from pre-scan features available before the quantized scan path. During calibration, we form pseudo-labels by clustering post-scan activation descriptors that capture similar rounding--clipping behavior. At inference, the frozen predictor maps pre-scan features to groups and selects entries from offline-calibrated group-wise clipping-bound tables, without runtime statistics collection, dynamic clipping updates, or post-scan group assignment. To address low-bit restoration degradation, we further introduce a Frequency-Preserving Refinement (FPR) objective for clipping-bound calibration. FPR aligns Fourier amplitude and phase between full-precision teacher and quantized student outputs to preserve high-frequency structures. Across PTQ protocols, bit-widths, backbones, and tasks, our method consistently outperforms prior Mamba and SR PTQ baselines, transfers to Transformer SR and Vision Mamba classification, composes with rotation-based quantization, and achieves 2.23x speedup on Jetson Orin Nano over FP16. Code is available here.

Figures & tables

Panel of Figure 1 — Figure 1: Selective scan induces token-wise activation heterogeneity. (a) Post-scan activations exhibit a broader distribution than pre-scan activations in a representative ASSM layer (layer 18). (b) Per-token range CoV increases after selective scan in most ASSM layers.

Figure 2: Clipping-scale regimes. Violin plots show token-level preferred clipping scales $c_{i}^{\star}$ under random, post-scan descriptor K-means (oracle), and predicted groups.

Figure 3: Overview of Token-Group Quantization. A frozen pre-scan predictor predicts token groups, and grouped INT sites use offline-calibrated clipping-bound tables without runtime statistics.

Scale	Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Scale	Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$\times$ 4	MambaIRv2-Light	32 / 32	32.51	0.8992	28.84	0.7878	27.75	0.7426	26.82	0.8079	31.24	0.9182
	PTQ4VM [ 8 ] (WACV’25)	4 / 4	30.82	0.8670	27.69	0.7546	26.95	0.7115	24.76	0.7321	28.19	0.8660
	Quamba [ 6 ] (ICLR’25)	4 / 4	31.01	0.8715	27.77	0.7585	26.99	0.7149	25.01	0.7470	28.57	0.8752
	MambaQuant [ 44 ] (ICLR’25)	4 / 4	30.74	0.8650	27.17	0.7413	26.37	0.6920	23.28	0.6694	26.73	0.8186
	OAQ-SR [ 38 ] (ICCV’25)	4 / 4	30.40	0.8577	27.41	0.7470	26.75	0.7045	24.31	0.7136	27.39	0.8485

Table 1: Quantitative comparison under the Weight-freeze PTQ protocol on MambaIRv2-Light.

Module	FLOPs	Ratio	Quant.
	(G)	(%)
Linear + Conv	8.49	85.9	Yes
Scan operands	0.49	4.9	Yes
Other Conv	0.90	9.1	No
Total	9.88	100.0	–

Table 2: FLOPs and W/A quantization scope.

Scale	Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Scale	Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$\times$ 4	MambaIRv2-Light	32 / 32	32.51	0.8992	28.84	0.7878	27.75	0.7426	26.82	0.8079	31.24	0.9182
	AIQViT [ 21 ] (AAAI’25)	4 / 4	31.51	0.8856	28.10	0.7707	27.26	0.7268	25.18	0.7546	28.76	0.8854
	SPR 2 Q [ 43 ] (ICLR’26)	4 / 4	32.04	0.8929	28.59	0.7812	27.52	0.7359	26.07	0.7864	30.12	0.9048
	Ours + LoRA	4 / 4	32.25	0.8953	28.67	0.7830	27.61	0.7377	26.24	0.7906	30.49	0.9095
	AIQViT [ 21 ] (AAAI’25)	2 / 2	30.52	0.8667	27.44	0.7541	26.79	0.7102	24.35	0.7210	27.27	0.8515

Table 3: Quantitative comparison under the Weight-training PTQ with LoRA protocol on MambaIRv2-Light.

Figure 4: Visual comparison for image SR ( $\times 2$ and $\times 4$ ) on Urban100 under 4-bit PTQ.

Table 8

Table 9

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Scale	Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Scale	Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$\times$ 4	MambaIRv1-Light	32 / 32	32.42	0.8977	28.74	0.7847	27.68	0.7400	26.52	0.7983	30.94	0.9135
	Layer-wise PTQ	4 / 4	30.39	0.8257	27.45	0.7249	26.71	0.6811	24.99	0.7185	28.62	0.8372
	OAQ-SR [ 38 ] (ICCV’25)	4 / 4	30.85	0.8507	27.80	0.7446	26.93	0.7019	25.03	0.7392	28.59	0.8611
	Ours	4 / 4	31.10	0.8627	27.91	0.7534	27.09	0.7097	25.23	0.7444	28.90	0.8682
	Layer-wise PTQ	2 / 2	23.10	0.4199	22.29	0.3725	22.25	0.3453	20.92	0.3736	21.99	0.4017

Appendix

Table 9: Generalization results on MambaIRv1-Light under the Weight-freeze PTQ protocol.

Method	W / A	CBSD68		Kodak24		McMaster		Urban100
Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
MambaIRv2-Base	32 / 32	34.15	0.9388	35.15	0.9363	35.07	0.9421	34.72	0.9484
Layer-wise PTQ	4 / 4	32.84	0.9161	33.49	0.9098	33.06	0.9169	32.62	0.9229
OAQ-SR [ 38 ] (ICCV’25)	4 / 4	33.52	0.9290	34.33	0.9243	34.04	0.9305	33.59	0.9371
Ours	4 / 4	33.78	0.9335	34.69	0.9299	34.51	0.9356	34.04	0.9419
Layer-wise PTQ	2 / 2	27.20	0.7871	27.00	0.7667	24.31	0.7564	25.44	0.7860

Appendix

Table 10: Generalization results on color image denoising with noise level

\sigma=15

under the Weight-freeze PTQ protocol.

Quality	Method	W / A	Classic5		LIVE1
Quality	Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$q=10$	MambaIRv2-Base	32 / 32	30.37	0.8269	29.91	0.8301
	Layer-wise PTQ	4 / 4	28.95	0.7834	28.74	0.7928
	OAQ-SR [ 38 ] (ICCV’25)	4 / 4	29.59	0.8064	29.26	0.8125
	Ours	4 / 4	29.93	0.8183	29.57	0.8230
	Layer-wise PTQ	2 / 2	27.57	0.7324	27.48	0.7474

Appendix

Table 11: Generalization results on JPEG artifact removal under the Weight-freeze PTQ protocol.

Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
SwinIR-light	32 / 32	38.14	0.9611	33.86	0.9206	32.31	0.9012	32.76	0.9340	39.12	0.9783
2DQuant (NeurIPS’24)	4 / 4	37.87	0.9594	33.41	0.9161	32.02	0.8971	31.84	0.9251	38.31	0.9761
FIMA-Q (CVPR’25)	4 / 4	37.71	0.9598	33.38	0.9160	32.01	0.8964	31.95	0.9283	38.44	0.9764
APHQ-ViT (CVPR’25)	4 / 4	37.76	0.9591	33.41	0.9165	31.94	0.8965	31.98	0.9280	38.34	0.9759
Ours	4 / 4	37.91	0.9600	33.47	0.9171	32.09	0.8978	32.06	0.9287	38.51	0.9773

Appendix

Table 12:

\times 2

SR on SwinIR-light under the Weight-freeze PTQ protocol.

Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
SwinIR-light	32 / 32	38.14	0.9611	33.86	0.9206	32.31	0.9012	32.76	0.9340	39.12	0.9783
SPR 2 Q (ICLR’26)	4 / 4	37.95	0.9602	33.55	0.9178	32.14	0.8989	32.04	0.9275	38.59	0.9769
Ours + LoRA	4 / 4	38.04	0.9608	33.65	0.9193	32.22	0.9003	32.28	0.9302	38.75	0.9776
SPR 2 Q (ICLR’26)	2 / 2	37.28	0.9572	32.83	0.9113	31.63	0.8921	30.20	0.9077	37.08	0.9726
Ours + LoRA	2 / 2	37.69	0.9596	33.22	0.9157	31.96	0.8971	31.16	0.9193	37.74	0.9753

Appendix

Table 13:

\times 2

SR on SwinIR-light under the Weight-training PTQ with LoRA protocol.

Method	W / A	Vim-Tiny	Vim-Small
FP	32 / 32	76.07	80.48
Layer-wise PTQ	6 / 6	71.92	76.66
PTQ4VM (WACV’25)	6 / 6	74.42	79.62
Ours	6 / 6	75.88	79.95
Layer-wise PTQ	4 / 4	53.61	67.06
PTQ4VM (WACV’25)	4 / 4	56.62	70.41

Appendix

Table 14: ImageNet Top-1 accuracy in percent on Vision Mamba under PTQ. Calibration and quantization scope follow PTQ4VM.

Backbone	Method	W / A	Set14		Urban100
Backbone	Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
MambaIRv2-Base	FP	32 / 32	34.89	0.9275	34.49	0.9468
	Layer-wise PTQ	4 / 4	32.39	0.9060	29.25	0.8930
	OAQ-SR (ICCV’25)	4 / 4	33.04	0.9132	31.25	0.9195
	Ours	4 / 4	33.28	0.9157	31.60	0.9236
	Layer-wise PTQ	2 / 2	30.54	0.8556	27.53	0.8398

Appendix

Table 15:

\times 2

SR on MambaIRv1-Base and MambaIRv2-Base under the Weight-freeze PTQ protocol.

Method	W / A	Set5		Set14		BSD100		Urban100		Manga109
Method	W / A	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
MambaIRv2-Light	32 / 32	38.26	0.9615	34.09	0.9221	32.36	0.9019	33.26	0.9378	39.35	0.9785
Affine	4 / 4	35.80	0.9452	32.81	0.9091	30.89	0.8789	30.57	0.9163	34.67	0.9550
Hadamard	4 / 4	36.33	0.9492	32.61	0.9085	31.04	0.8823	30.36	0.9123	35.73	0.9655
SpinQuant	4 / 4	36.42	0.9500	32.50	0.9063	31.08	0.8828	30.74	0.9188	35.54	0.9622
Ours	4 / 4	37.56	0.9578	33.24	0.9149	31.86	0.8950	31.34	0.9204	37.85	0.9742

Appendix

Table 16:

\times 2

SR on MambaIRv2-Light with channel-axis transforms ported onto the identical Weight-freeze PTQ protocol. Affine covers AffineQuant and FlatQuant, and Hadamard covers QuaRot.

Perturbation rate $\rho$	Set14		Urban100
Perturbation rate $\rho$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$\rho=0\%$ (ours, no perturbation)	33.24	0.9149	31.34	0.9204
$\rho=5\%$	33.21	0.9145	31.34	0.9204
$\rho=10\%$	33.18	0.9143	31.33	0.9202
$\rho=20\%$	33.09	0.9137	31.25	0.9193
$\rho=30\%$	33.01	0.9127	31.13	0.9178

Appendix

Table 17: Robustness to inference-time predictor perturbation on MambaIRv2-Light (

\times 2

, W4A4) under the Weight-freeze PTQ protocol. We use the calibrated bound tables and the trained predictor from our default setting and, only at evaluation time, randomly reassign a fraction

\rho

of predicted group ids to a uniformly random group in

\{1,\dots,K\}

. Calibration, bound refinement, and the predictor are unchanged across rows. The

\rho{=}0\%

row matches our default deployment.

LR crop	Tokens	Ratio to calib.	FP		Ours		Gap
LR crop	Tokens	Ratio to calib.	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
$32\times 32$	1,024	0.25 $\times$	33.77	0.9127	31.88	0.8929	1.89	0.0198
$64\times 64$	4,096	1 $\times$	32.60	0.9084	30.48	0.8825	2.12	0.0259
$128\times 128$	16,384	4 $\times$	31.89	0.9105	29.78	0.8873	2.11	0.0232
$256\times 256$	65,536	16 $\times$	32.13	0.9217	30.07	0.9003	2.06	0.0214

Appendix

Table 18: Resolution robustness on Urban100 at

\times 2

and W4A4 over 20 images. Predictor and bound tables are frozen and only the LR input crop size varies. Gap is FP minus Ours.

$K$	Set14		Urban100
$K$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
1	32.55	0.9099	30.98	0.9084
4	33.20	0.9141	31.37	0.9193
8, default	33.28	0.9157	31.60	0.9236
12	33.34	0.9163	31.68	0.9244
16	33.37	0.9168	31.71	0.9249

Appendix

Table 19: Effect of the number of token groups

K

on MambaIRv2-Base at

\times 2

and W4A4.

Frequency term	Amplitude error			Phase error
Frequency term	Low	Mid	High	Low	Mid	High
none	0.0208	0.2853	0.536	0.0213	0.3211	0.673
+ amplitude only	0.0206	0.2784	0.512	0.0212	0.3191	0.668
+ phase only	0.0213	0.2830	0.531	0.0216	0.3181	0.635
+ $\mathcal{L}_{\mathrm{FPR}}$ , Ours	0.0214	0.2850	0.518	0.0219	0.3213	0.643

Appendix

Table 20: Band-wise Fourier errors of the quantized output against the FP output on Urban100 at

\times 2

and W4A4. Lower is better. All variants use token groups, and none denotes

\mathcal{L}_{\mathrm{pix}}{+}\mathcal{L}_{\mathrm{feat}}

without any frequency term.

$\lambda_{\mathrm{FPR}}$	Set14		Urban100
$\lambda_{\mathrm{FPR}}$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$
$0.005$	33.11	0.9124	31.19	0.9184
$0.01$ (ours)	33.24	0.9149	31.34	0.9204
$0.05$	33.22	0.9140	31.31	0.9197
$0.1$	33.09	0.9116	31.04	0.9175

Appendix

Table 21: Sensitivity to FPR objective weight

\lambda_{\mathrm{FPR}}

on MambaIRv2-Light (

\times 2

, W4A4) under the Weight-freeze PTQ protocol. All other settings are kept at their default values.

\lambda_{\mathrm{FPR}}{=}0.01

corresponds to our default setting.

Figure 5: Clipping-scale regimes in early/middle layers. Token-level preferred clipping scales

c_{i}^{\star}

for layers 1 and 6 under random, post-scan descriptor K-means (oracle), and predicted groups.

Figure 6: Clipping-scale regimes in later layers. Token-level preferred clipping scales

c_{i}^{\star}

for layers 11 and 16 under random, post-scan descriptor K-means (oracle), and predicted groups.

Activation Quantization

Activation quantization reduces the precision of neural-network activations to lower inference and training memory use and accelerate computation.

See all 117 papers

May 21, 2026

QuantSR+: Pushing the Limit of Quantized Image Super-Resolution Networks

Low-bit quantization is widely used to compress super-resolution (SR) models and reduce storage and computation costs for deployment on resource-limited devices. However, when SR models are pushed to ultra-low precision (2-4 bits), performance can drop sharply due to diminished representational capacity and the detail-sensitive nature of SR. To address these issues, we propose QuantSR+, a unified framework that improves quantization operators, network design, and training optimization, achieving better trade-offs between accuracy and efficiency than prior low-bit SR methods. QuantSR+ mainly relies on three technical contributions: (1) Redistribution-driven Bit Determination (RBD), which reshapes quantization distributions in both forward and backward passes to preserve representation fidelity; (2) Quantized Slimmable Architecture (QSA), which begins with an over-parameterized model and progressively prunes less critical blocks to meet efficiency budgets while pushing the accuracy performance; and (3) Slimming-guided Function-localized Distillation (SFD), which enforces block-aware feature alignment via a direct loss and a progressive, function-local training schedule to capture quantization effects better and speed up convergence. Extensive experiments show that QuantSR+ achieves state-of-the-art performance against both specialized quantized SR methods and generic quantization approaches. For SwinIR-S on Urban100 (x4), it improves PSNR by 0.29 dB over the 2-bit SOTA baseline. Meanwhile, it delivers strong efficiency gains at 2-bit, reducing operations by up to 87.9% and storage by 89.4%. QuantSR+ is effective for both convolutional and transformer-based SR models, indicating broad applicability.

Haotong Qin, Xudong Ma, Xianglong Liu +4

ETH Zurich, Switzerland · Beihang University, China · Shanghai Jiao Tong University, China

Quantization-Aware Training Low-Bit Quantization Image Super-Resolution

Ask PDF

Sep 15, 2026

Channel-Wise and Token-Aware Post-Training Quantization for Visual State Space Duality

State space models (SSMs), particularly Mamba, have emerged as efficient alternatives to attention-based architectures and have been extended to vision through ViM, VMamba, and Visual State Space Duality (VSSD). Yet the low-bit post-training quantization (PTQ) behavior of VSSD remains insufficiently understood. A weight-activation split on VSSD-Tiny identifies activation quantization as the dominant low-bit bottleneck, while representative inputs to selected VSSD-backbone linear layers exhibit strong channel-wise magnitude variation and token-localized extremes. We propose the Channel-wise Token-balanced Output-Aware Clipping (CTOAC) method, which learns per-input-channel clipping bounds by minimizing a token-balanced reconstruction loss on the corresponding linear outputs. Only the selected linear layers and their input activations are quantized; other backbone operations retain their original precision. Across VSSD-Tiny, VSSD-Small, and VSSD-Base, the proposed CTOAC method retains ImageNet-1K accuracy and remains substantially more robust than the evaluated baselines at more aggressive precision settings. Applying the same quantization scope to VSSD backbones on COCO and ADE20K preserves strong object detection, instance segmentation, and semantic segmentation performance. An optimized RTX 4090 deployment configuration achieves up to 1.42x end-to-end speedup over FP32.

Jonghyeon Lim, Changhoon Yim

Intelligent Image Processing Laboratory · Department of Computer Science and Engineering, Konkuk University · Seoul, Republic of Korea

Visual SSMs Activation Quantization Post-Training Quantization Neural Network Quantization

Ask PDF

Jun 16, 2026

Ternary Mamba: Grouped Quantization-Aware Training of W1.58A16 State Space Models

State Space Models (SSMs) such as Mamba-2 offer linear-time inference but their memory footprint limits edge deployment. Prior ternary SSM work (Slender-Mamba) trains from scratch on 150B tokens; we show a pretrained checkpoint suffices, reducing the marginal token budget by 1,000x. Using grouped quantization-aware training (QAT) with knowledge distillation from a frozen FP16 teacher, we compress Mamba-2 1.3B to 3.61x (2,687 to 744 MB) and achieve 48.1% zero-shot accuracy (7-task average) in just 102M tokens (4 GPU-hours, single H100) -- approaching Bi-Mamba's 48.4% (within +/-0.9pp CI). This QAT-from-pretrained setting reveals zero-ratio collapse, a novel instability caused by learnable quantization scales that does not arise in from-scratch training. We further show that post-hoc correction strategies effective for Transformers fail for SSMs due to error accumulation through the recurrence. These results demonstrate that ternary SSMs do not require expensive from-scratch training: QAT from pretrained checkpoints with KD is a data-efficient alternative.

Ramprasath Ganesaraja, Sahil Dilip Panse, Swathika N

EdgeVerve Systems Limited

LLM Quantization Ternary Quantization LLM Compression Quantization-Aware Training Neural Network Quantization State Space Models

Ask PDF

Abstract

Activation Quantization

Explore similar work

QuantSR+: Pushing the Limit of Quantized Image Super-Resolution Networks

Channel-Wise and Token-Aware Post-Training Quantization for Visual State Space Duality

Ternary Mamba: Grouped Quantization-Aware Training of W1.58A16 State Space Models