Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Panacea: Novel DNN Accelerator using Accuracy-Preserving Asymmetric Quantization and Energy-Saving Bit-Slice Sparsity

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims a bit-slice GEMM that skips frequent nonzero high-order slices of asymmetric quantized activations, preserving exactness via a compensation term, and an accelerator, Panacea, that cuts MACs by 61% and beats prior…

desk verdict AQS-GEMM's compensation trick is sound and new; the main risk is that the headline LLM gains rest on unshown activation sparsity. read the letter →

arxiv 2412.10059 v1 pith:LEG4SBY2 submitted 2024-12-13 cs.AR cs.AI

Dongyun Kam , Myeongji Yun , Sunwoo Yoo , Seungwoo Hong , Zhengya Zhang , Youngjoo Lee This is my paper · ORCID
classification cs.ARcs.AI
keywords asymmetricquantizationbit-sliceGEMMDNNacceleratorslicesparsityzero-pointmanipulationdistribution-basedbit-slicingrun-lengthencodingtransformerinference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to prove that bit-slice GEMM no longer needs to wait for zero slices to save work. It proposes AQS-GEMM, which treats the most frequent nonzero high-order slice value produced by asymmetric activation quantization as a compressible pattern, skips the multiply-accumulate operations of the compressed slice vectors, and restores the exact result with a compensation term. On top of that, zero-point manipulation and distribution-based bit-slicing push more activation values into the skippable range during post-training calibration. The authors build these ideas into a DNN accelerator, Panacea, and report 61% fewer MAC operations than dense GEMM, with better energy efficiency and throughput than prior bit-slice and dense accelerators on transformer models. If correct, this closes the gap between accuracy-friendly asymmetric quantization and hardware-friendly slice sparsity.

What carries the argument

The carrying object is the AQS-GEMM reformulation in Eq. (6), which rewrites $(W_{HO}+W_{LO})x_{HO}$ as $(W_{HO}+W_{LO})x^U_{HO} - r(W_{HO}+W_{LO})J_U + b'$, so that skipped high-order slice vectors, all equal to the frequent value $r$, are compensated with weight slices already loaded for the uncompressed vectors. Two calibration-time helpers make the compression effective: zero-point manipulation shifts the activation zero point so the distribution centers in the skip range, and distribution-based bit-slicing chooses 4-, 5-, or 6-bit low-order slices per layer to widen the skip range for wide distributions. The hardware counterpart is the Panacea processing element, which splits work between dynamic operators for sparse slice GEMMs and static operators for dense low-order GEMMs and uses run-length encoding and double-tile processing to keep utilization high.

What would settle it

Measure the high-order slice-vector sparsity produced by ZPM and DBS on a transformer layer outside the paper's benchmark set, using the same 1x4 grouping: if the frequent-slice fraction falls below the break-even point where Panacea's throughput crosses SIMD in Fig. 13, the reported energy and throughput advantages for that model would not materialize.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that the obstacle to exploiting slice sparsity under asymmetric activation quantization, namely the near absence of all-zero high-order slices, is removable. Writing the high-order activation slice matrix as $x_{HO} = x^U_{HO} + x^C_{HO}$, where the compressed part equals $r$ times an indicator matrix and $r$ is the zero-point high-order slice value, the paper rearranges the GEMM to compute only the uncompressed part and then compensates exactly using weight slices already loaded for the uncompressed vectors, plus a precomputed bias term. With the sparsity boosters ZPM and DBS, the scheme attains high vector-level sparsity on real transformer layers and cuts MACs by 61% relative to dense GEMM. The paper reports that Panacea delivers 1.97x higher energy efficiency and 1.88x higher throughput than Sibia on OPT-2.7B while keeping the accuracy advantage of asymmetric quantization.

Load-bearing premise

The efficiency gains depend on activation high-order slice vectors being dominated by a single frequent value $r$ in real layers, so that 1x4 vectors can be compressed; if that distribution is wider or multimodal, compression falls and Panacea can be slower than SIMD, as the paper's own Fig. 13 shows at low slice sparsity.

Editorial extensions

If this is right

  • Bit-slice accelerators can use asymmetric activation quantization without sacrificing exactness, because skipping nonzero slices no longer changes the answer.
  • Activation slice sparsity becomes tunable at calibration time through ZPM and DBS, rather than being fixed by the quantization format.
  • Because compensation reuses weight slices already loaded for the uncompressed computations, the memory-traffic penalty normally associated with nonzero slices is avoided.
  • On the evaluated transformer benchmarks, Panacea beats both the bit-slice accelerator Sibia and dense SIMD and systolic designs in energy efficiency and throughput at equal multiplier counts.
  • The scheme scales to 4-bit weights with mixed precision, widening the low-bit operating range for large language models while preserving perplexity.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial: the same compress-the-frequent-nonzero-value idea could apply to other linear layers and to quantized embeddings or token-level outliers, but the paper only validates it on transformer and ResNet layers, so that extension is untested.
  • Editorial: because the DWO/SWO operator allocation strongly affects throughput, a per-layer choice of how many dynamic operators to use, predicted from each layer's measured sparsity, could improve on the fixed 4-DWO/8-SWO configuration.
  • Editorial: the accuracy claim rests on calibration-set distributions; a robustness check across longer sequences, multilingual data, or different calibration subsets would be the natural stress test for the sparsity boosters.
  • Editorial: the 61% MAC reduction is an average over evaluated layers; reporting the distribution across layers would let users predict which models benefit most.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Panacea, an accelerator that reformulates bit-slice GEMM under asymmetric activation quantization so that frequent nonzero high-order (HO) slices can be compressed and skipped without changing the exact GEMM result. The key algorithmic contribution is AQS-GEMM (Section III-B), which rewrites the skipped contribution of r-valued HO activation vectors as a compensation term that reuses weight slices already loaded for the uncompressed computation, derived in Eq. (6). To raise the resulting vector-level sparsity, the paper introduces zero-point manipulation (ZPM) and distribution-based bit-slicing (DBS) during PTQ calibration (Section III-C). The hardware design combines dynamic and static workload operators, a specialized dataflow, run-length encoding, and double-tile processing (Section III-D). The evaluation compares Panacea against Sibia, systolic arrays, and a SIMD accelerator on DeiT-base, BERT-base, GPT-2, OPT variants, Llama-3.2, and ResNet-18, reporting up to 1.97x energy-efficiency and 1.88x throughput gains over Sibia on OPT-2.7B while maintaining perplexity close to FP16.

Significance. If the reported results hold, the paper makes a valuable contribution: it is, to my knowledge, the first bit-slice GEMM formulation that skips frequent nonzero HO slices arising from asymmetric quantization while preserving exactness. The algebraic derivation in Section III-B is parameter-free and verifiable—Eq. (6) correctly reconstructs the skipped term by accumulating weight slices already fetched for the uncompressed vectors, so the central algorithmic claim is sound without fitted parameters. The reported MAC reduction, the ZPM/DBS sparsity improvements, and the hardware design with compensators and double-tile processing are all concrete, falsifiable claims. The main weakness is empirical: the headline efficiency numbers for the largest LLMs rest on vector-sparsity measurements that are only partially reported, and the evaluation pipeline is not yet independently reproducible.

major comments (4)
  1. [Section IV, Figs. 13–14 and 16–17] The efficiency claims for OPT-2.7B and Llama-3.2 (Figs. 16–17, e.g., 1.97x and 1.88x over Sibia) depend entirely on the HO slice-vector sparsity rho_x of activations after ZPM and DBS, because AQS-GEMM only skips 1x4 vectors whose HO slices are all equal to r. However, Fig. 14(b) reports vector sparsity only for DeiT-base, BERT-base, and GPT-2; the models used for the headline numbers are absent. Fig. 13 is the paper's own crossover check, showing that Panacea is slower than SIMD at low rho_x. Without measured rho_x for OPT-2.7B and Llama-3.2 layers, or a sensitivity analysis showing the reported gains survive plausible sparsity variations, the central efficiency conclusion is not yet established for the largest models.
  2. [Section III-C, Fig. 9] The DBS categorization step that determines the slice bit-width is not specified to a reproducible degree: the paper refers to 'three distinct ranges' of std x z but never states the z-score thresholds or the target sparsity, and the calibration data used for the histogram analysis is not described. Because these thresholds directly control how many activation vectors become skippable, their absence makes the sparsity numbers in Fig. 14 and the energy numbers in Figs. 15–17 impossible to reproduce or to check for unintended overfitting to the reported benchmarks.
  3. [Section III-C and Fig. 10] DBS deliberately discards the LSBs of long LO slices (Fig. 10), which is an approximate step. The resulting accuracy loss is quantified only for DeiT-base (acceptably 0.6%p); for the OPT and Llama-3.2 models that provide the headline performance numbers, no accuracy or perplexity results are reported for DBS itself. Since the 'accuracy-preserving' part of the title claim depends on this approximation being negligible on the evaluated LLMs, the paper should report PPL or downstream-task accuracy with and without DBS for those models.
  4. [Section IV, first paragraph] The energy and throughput evaluation relies on a custom simulation flow—counting cycles and activated modules, then estimating energy with CACTI 7.0 and post-layout building-block results—but no validation of this flow against the implemented 28nm FD-SOI layout is shown, and the open-source link (Section V) is given without available code. Since the headline gains (e.g., 1.97x energy efficiency) are simulator-based, the paper should either release the simulator/code or provide a validation subsection comparing simulator output against the measured or post-layout numbers of a testbed design.
minor comments (5)
  1. [Section III-B, Table I] The workload formulas in Table I assume 'sufficient memory space' and a particular tiling, but the definitions of rho_x and rho_w are not stated precisely; please add a sentence defining them over the K-dimension of the tiled GEMM, consistent with Section III-C.
  2. [Section III-C, Eq. (7)] The notation zp'' and r'' is introduced after Eq. (7) without explicit formulas; please state that they are obtained by applying Eq. (7) with the DBS-adjusted LO bit-width l.
  3. [Section III-D] The phrase 'sign-unsigned multipliers' is ambiguous; it likely means signed-by-unsigned multipliers, which should be stated explicitly.
  4. [Section IV, Fig. 14(a)] The y-axis label reads 'HO vector sparsity of activations in the DeiT-base model’s layers'—the apostrophe is misplaced, and the caption should specify whether the graphed values are before or after ZPM/DBS.
  5. [Section IV, Fig. 16] The paper claims all designs use identical hardware resources, but Panacea includes two compensators and additional DTP buffers; please clarify how the area comparisons in Fig. 15(c) account for these extra components.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the AQS-GEMM compensation identity is an exact algebraic reformulation and the reported speedups are benchmark-measured rather than fitted outputs.

full rationale

The paper's central derivation, AQS-GEMM, is not circular. Equation (5)-(6) rewrite (WHO+WLO)xHO as (WHO+WLO)xU_HO - r(WHO+WLO)JU + b', using only xHO = xU_HO + r*JC and JC = 1 - JU, with b' precomputed offline as r(WHO+WLO)1. This is an exact algebraic identity for any activation distribution; it does not assume the sparsity it later reports. The ZPM and DBS heuristics are co-design choices that influence measured slice sparsity, but the sparsity values in Figs. 14-16 are obtained by simulating the benchmark models after those choices, not by renaming a fitted parameter. The two self-citations, [38] and [39], appear only in a general related-work enumeration of pruning accelerators and are never used to justify the AQS-GEMM identity, the compensation term, or the energy/throughput results; they are therefore not load-bearing. The main limitation, that high speedups depend on empirically high activation HO slice-vector sparsity and that Fig. 13 shows lower throughput than SIMD at low sparsity, is a correctness/robustness concern rather than a circularity concern. Missing OPT-2.7B/Llama-3.2 sparsity measurements and the unquantified DBS accuracy loss on LLMs are evidence gaps, not definitional reductions.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a small number of hand-chosen algorithm and architecture parameters, on the empirical assumption that activation distributions are concentrated, and on an unreleased simulator. No new physical entities are introduced. The mathematical identity at the core of AQS-GEMM is clean, but the efficiency numbers depend on the free parameters and domain assumptions listed above.

free parameters (5)
  • DBS type thresholds (std x z ranges) = not disclosed
    The paper categorizes activation distributions into three types using 'three distinct ranges' of std times z but never gives the cutoff values; these are chosen during calibration and directly determine slice sparsity gains.
  • DBS LO slice bit-widths (l) = l=4 for type-1, l=5 for type-2, l=6 for type-3
    Hand-picked bit allocations that expand the skip range; they trade precision (0.6 percentage points on DeiT-base) for sparsity, so the reported efficiency depends on these choices.
  • ZPM block bit-width (l) = l=4
    Zero-point manipulation shifts zp to the center of a 2^l block in Eq. (7); the choice l=4 matches the 4-bit HO/LO split and is a design decision that affects sparsity and accuracy.
  • DWO/SWO operator ratio = 4 DWOs and 8 SWOs per PEA
    Chosen from the observed vector sparsity of DeiT-base, BERT-base, and GPT-2; Fig. 13 shows a different ratio changes throughput, so the headline gains depend on this configuration.
  • DTP enable condition = WMEM can hold 2TM x K weight tile and WBUF can hold two weight sub-tiles
    The double-tile processing flow is enabled only when buffers are large enough; this heuristic improves utilization at high sparsity and is a hardware decision, not derived.
assumptions (5)
  • standard math The exact decomposition xHO = xU_HO + r x JC and the identity in Eq. (6) hold under standard integer arithmetic.
    The compensation term reuses loaded weights because JU marks uncompressed vectors; this is an algebraic identity, not an empirical claim.
  • domain assumption Real DNN activation distributions are concentrated enough around the zero point that after ZPM and DBS, most 1x4 HO slice vectors are all-r.
    Section III-C uses this to justify compression; Section IV validates it only on the listed benchmarks, so it is an empirical assumption about future or unlisted layers.
  • domain assumption Shifting the zero point via ZPM does not materially change model accuracy.
    Section III-C states 'does not cause a considerable change in accuracy' without a full accuracy table per model; the claim is load-bearing for using ZPM unconditionally.
  • domain assumption DBS's lossy LSB truncation and short-HO zero-filling cause acceptable accuracy loss across all target models.
    Section III-C quantifies 0.6 percentage points loss only for DeiT-base; for OPT and Llama, PPL is reported, but the interaction of DBS with all layers is not separately isolated.
  • domain assumption The cycle/energy simulator correctly accounts for RLE decoding, index matching, scheduling, compensation, and DTP overheads.
    Section IV counts cycles based on HuggingFace models and post-layout blocks, but no simulator or event-counting code is released, so this cannot be independently checked.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Panacea: Novel DNN Accelerator using Accuracy-Preserving Asymmetric Quantization and Energy-Saving Bit-Slice Sparsity." pith.science (2026). https://pith.science/paper/LEG4SBY2

@misc{pith2026241210059,
  author       = {Pith},
  title        = {Pith review of: Panacea: Novel DNN Accelerator using Accuracy-Preserving Asymmetric Quantization and Energy-Saving Bit-Slice Sparsity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LEG4SBY2}},
  note         = {Machine review of arXiv:2412.10059}
}
read the original abstract

Low bit-precisions and their bit-slice sparsity have recently been studied to accelerate general matrix-multiplications (GEMM) during large-scale deep neural network (DNN) inferences. While the conventional symmetric quantization facilitates low-resolution processing with bit-slice sparsity for both weight and activation, its accuracy loss caused by the activation's asymmetric distributions cannot be acceptable, especially for large-scale DNNs. In efforts to mitigate this accuracy loss, recent studies have actively utilized asymmetric quantization for activations without requiring additional operations. However, the cutting-edge asymmetric quantization produces numerous nonzero slices that cannot be compressed and skipped by recent bit-slice GEMM accelerators, naturally consuming more processing energy to handle the quantized DNN models. To simultaneously achieve high accuracy and hardware efficiency for large-scale DNN inferences, this paper proposes an Asymmetrically-Quantized bit-Slice GEMM (AQS-GEMM) for the first time. In contrast to the previous bit-slice computing, which only skips operations of zero slices, the AQS-GEMM compresses frequent nonzero slices, generated by asymmetric quantization, and skips their operations. To increase the slice-level sparsity of activations, we also introduce two algorithm-hardware co-optimization methods: a zero-point manipulation and a distribution-based bit-slicing. To support the proposed AQS-GEMM and optimizations at the hardware-level, we newly introduce a DNN accelerator, Panacea, which efficiently handles sparse/dense workloads of the tiled AQS-GEMM to increase data reuse and utilization. Panacea supports a specialized dataflow and run-length encoding to maximize data reuse and minimize external memory accesses, significantly improving its hardware efficiency. Our benchmark evaluations show Panacea outperforms existing DNN accelerators.

Figures

Figures reproduced from arXiv: 2412.10059 by the authors.

Figure 1
Figure 1. Accuracy comparison on recent works utilizing symmetric [18]–[22] [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Examples of uniform quantization methods: 8-bit (a) symmetric and [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. An example of the bit-slice GEMM using 7-bit symmetric quantization [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (13 more)
Figure 5
Figure 5. Figure 5: (a) Distributions of asymmetrically quantized activations. (b) Accuracy [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 7
Figure 7. Figure 7: (a) Compressing vectors only comprising frequent slices in [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Distributions of an asymmetrically quantized activation (a) without [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: Categorizing distributions of quantized activations into three types [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Dynamically applying bit-slicing rules to different types during the [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: Overall architecture of Panacea incorporating the AQS-GEMM core with 16 PEAs, each of which consists of 4 DWOs and 8 SWOs [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Panacea’s output stationary dataflow computing the tiled AQS￾GEMM to maximize data reuse. 4-bit width of slices by filling short HO slices with redundant zero bits and discarding LSBs in long LO slices. This increases the average slice sparsity by 20% (more than 50% f…
Figure 13
Figure 13. Figure 13: Throughput evaluation of Panacea when using different design options, changing the HO slice sparsities of weight and activation, and changing the size of weight and activation; (a) 4 DWOs and 8 SWOs per PEA, and (b) 8 DWOs and 4 SWOs per PEA. ternal memory. This dataf…
Figure 14
Figure 14. Figure 14: (a) HO vector sparsity of activations in the DeiT-base model’s layers (b) The vector sparsity of weight/activation HO slices observed by [PITH_FULL_IMAGE:figures/full_fig_p010_14.png]
Figure 15
Figure 15. Figure 15: (a) Energy breakdown and (b) throughput of the DNN accelerators for different benchmarks. (c) Relative area cost for applying the proposed methods. [PITH_FULL_IMAGE:figures/full_fig_p011_15.png]
Figure 16
Figure 16. Figure 16: Energy efficiency, throughput, and accuracy loss evaluations of [PITH_FULL_IMAGE:figures/full_fig_p011_16.png]
Figure 19
Figure 19. Figure 19: Energy breakdown, latency, and perplexity of [PITH_FULL_IMAGE:figures/full_fig_p012_19.png]
Figure 18
Figure 18. Figure 18: Relative energy efficiency and perplexity for OPT-2.7B [60] when [PITH_FULL_IMAGE:figures/full_fig_p012_18.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MCBP: A Memory-Compute Efficient LLM Inference Accelerator Leveraging Bit-Slice-enabled Sparsity and Repetitiveness

    cs.AR 2025-09 conditional novelty 7.0 of 10

    A bit-slice-based accelerator (MCBP) jointly reduces GEMM computation, weight traffic, and KV cache traffic for LLM inference, claiming 9.43x speedup and 31.1x energy efficiency over A100.

Reference graph

Works this paper leans on

80 extracted references · 75 canonical work pages · cited by 1 Pith paper

  1. [75]

    GLUE: A multi-task benchmark and analysis platform for natural language understanding

    Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In arXiv.org, 2018

    work page 2018

  2. [80]

    Pointer sentinel mixture models

    Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models. In arXiv.org, 2016. 15

    work page 2016

  3. [54]

    Non-blocking simultaneous multithread- ing: Embracing the resiliency of deep neural networks

    Gil Shomron and Uri Weiser. Non-blocking simultaneous multithread- ing: Embracing the resiliency of deep neural networks. In Proc. International Symposium on Microarchitecture (MICRO) , pages 256– 269, 2020

    work page 2020

  4. [1]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 770–778, 2016

    2016

  5. [2]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weis- senborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In arXiv.org, 2020

    work page 2020

  6. [3]

    Training data-efficient image transformers & distillation through attention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through attention. In Proc. International conference on machine learning (ICML) , pages 10347–10357, 2021

    work page 2021

  7. [4]

    Learning natural language inference with LSTM

    Shuohang Wang and Jing Jiang. Learning natural language inference with LSTM. In arXiv.org, 2015

    work page 2015

  8. [5]

    Bert: Pre-training of deep bidirectional transformers for language un- derstanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language un- derstanding. In arXiv.org, 2018

    work page 2018

Show all 80 references
  1. [6]

    Improving language understanding by generative pre-training, 2018

    Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training, 2018

    2018

  2. [7]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019

    2019

  3. [8]

    GPT-3: Its nature, scope, limits, and consequences

    Luciano Floridi and Massimo Chiriatti. GPT-3: Its nature, scope, limits, and consequences. Minds Mach., 30:681–694, 2020

    2020

  4. [9]

    Llama 2: Open foundation and fine-tuned chat models

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Alma- hairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Can- ton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fer- nandes, Jeremy...

    2023

  5. [10]

    Edge intelligence: Paving the last mile of artificial intelligence with edge computing

    Zhi Zhou, Xu Chen, En Li, Liekang Zeng, Ke Luo, and Junshan Zhang. Edge intelligence: Paving the last mile of artificial intelligence with edge computing. Proc. IEEE, 107(8):1738–1762, 2019

    2019

  6. [11]

    Deep learning-based smart task assistance in wearable augmented reality

    Kyeong-Beom Park, Minseok Kim, Sung Ho Choi, and Jae Yeol Lee. Deep learning-based smart task assistance in wearable augmented reality. Robot. Comput.-Integr. Manuf., 63:101887, 2020

    2020

  7. [12]

    Trager, Shahar Avin, Adrian Weller, Yoshua Bengio, and Diane Coyle

    Girish Sastry, Lennart Heim, Haydn Belfield, Markus Anderljung, Miles Brundage, Julian Hazell, Cullen O’Keefe, Gillian K Hadfield, Richard Ngo, Konstantin Pilz, Geroge Gor, Emma Bluemke, Sarah Shoker, Janet Egan, Robert F. Trager, Shahar Avin, Adrian Weller, Yoshua Bengio, and...

    2024

  8. [13]

    Scaling laws for neural language models

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Ben- jamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. In arXiv.org, 2020

    2020

  9. [14]

    Towards accurate and reliable energy measurement of NLP models, 2020

    Qingqing Cao, Aruna Balasubramanian, and Niranjan Balasubramanian. Towards accurate and reliable energy measurement of NLP models, 2020

    2020

  10. [15]

    Full stack optimization of transformer inference: a survey, 2023

    Sehoon Kim, Coleman Hooper, Thanakul Wattanawong, Minwoo Kang, Ruohan Yan, Hasan Genc, Grace Dinh, Qijing Huang, Kurt Keutzer, Michael W Mahoney, Yakun Sophia Shao, and Amir Gholami. Full stack optimization of transformer inference: a survey, 2023

    2023

  11. [16]

    A survey on deploying mobile deep learning applications: A systemic and technical perspective

    Yingchun Wang, Jingyi Wang, Weizhan Zhang, Yufeng Zhan, Song Guo, Qinghua Zheng, and Xuanyu Wang. A survey on deploying mobile deep learning applications: A systemic and technical perspective. Digit. Commun. Netw., 8(1):1–17, 2022

    2022

  12. [17]

    Design possibilities and challenges of DNN models: a review on the perspective of end devices

    Hanan Hussain, PS Tamizharasan, and CS Rahul. Design possibilities and challenges of DNN models: a review on the perspective of end devices. Artif. Intell. Rev., pages 1–59, 2022

    2022

  13. [18]

    Easyquant: Post-training quantization via scale optimization

    Di Wu, Qi Tang, Yongle Zhao, Ming Zhang, Ying Fu, and Debing Zhang. Easyquant: Post-training quantization via scale optimization. In arXiv.org, 2020

    2020

  14. [19]

    Post-training quantization for vision transformer

    Zhenhua Liu, Yunhe Wang, Kai Han, Wei Zhang, Siwei Ma, and Wen Gao. Post-training quantization for vision transformer. In Proc. Ad- vances in neural information processing systems (NeurIPs) , volume 34, pages 28092–28103, 2021

    2021

  15. [20]

    Brecq: Pushing the limit of post- training quantization by block reconstruction

    Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. Brecq: Pushing the limit of post- training quantization by block reconstruction. In arXiv.org, 2021

    2021

  16. [21]

    Aciq: Analytical clipping for integer quantization of neural networks

    Ron Banner, Yury Nahshan, Elad Hoffer, and Daniel Soudry. Aciq: Analytical clipping for integer quantization of neural networks. In arXiv.org, 2018

    2018

  17. [22]

    Smoothquant: Accurate and efficient post-training quan- tization for large language models

    Guangxuan Xiao, Ji Lin, Mickael Seznec, Hao Wu, Julien Demouth, and Song Han. Smoothquant: Accurate and efficient post-training quan- tization for large language models. In Proc. International conference on machine learning (ICML) , pages 38087–38099, 2023

    2023

  18. [23]

    Fq-vit: Post-training quantization for fully quantized vision transformer

    Yang Lin, Tianyu Zhang, Peiqin Sun, Zheng Li, and Shuchang Zhou. Fq-vit: Post-training quantization for fully quantized vision transformer. In arXiv.org, 2021

    2021

  19. [24]

    Pd-quant: Post-training quantization based on prediction difference metric

    Jiawei Liu, Lin Niu, Zhihang Yuan, Dawei Yang, Xinggang Wang, and Wenyu Liu. Pd-quant: Post-training quantization based on prediction difference metric. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 24427–24437, 2023

    2023

  20. [25]

    Flexround: Learnable rounding based on element-wise division for post- training quantization

    Jung Hyun Lee, Jeonghoon Kim, Se Jung Kwon, and Dongsoo Lee. Flexround: Learnable rounding based on element-wise division for post- training quantization. In Proc. International conference on machine learning (ICML), pages 18913–18939, 2023

    2023

  21. [26]

    Zeroq: A novel zero shot quantization framework

    Yaohui Cai, Zhewei Yao, Zhen Dong, Amir Gholami, Michael W Mahoney, and Kurt Keutzer. Zeroq: A novel zero shot quantization framework. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 13169–13178, 2020

    2020

  22. [27]

    Repq-vit: Scale reparameterization for post-training quantization of vision transformers

    Zhikai Li, Junrui Xiao, Lianwei Yang, and Qingyi Gu. Repq-vit: Scale reparameterization for post-training quantization of vision transformers. In Proc. International Conference on Computer Vision (ICCV) , pages 17227–17236, 2023

    2023

  23. [28]

    Outlier suppression: Pushing the limit of low-bit transformer language models

    Xiuying Wei, Yunchen Zhang, Xiangguo Zhang, Ruihao Gong, Shang- hang Zhang, Qi Zhang, Fengwei Yu, and Xianglong Liu. Outlier suppression: Pushing the limit of low-bit transformer language models. In Proc. Advances in neural information processing systems (NeurIPs) , volume 35,...

    2022

  24. [29]

    Outlier suppression+: Accurate 13 quantization of large language models by equivalent and optimal shifting and scaling

    Xiuying Wei, Yunchen Zhang, Yuhang Li, Xiangguo Zhang, Ruihao Gong, Jinyang Guo, and Xianglong Liu. Outlier suppression+: Accurate 13 quantization of large language models by equivalent and optimal shifting and scaling. In arXiv.org, 2023

    2023

  25. [30]

    A white paper on neural network quantization

    Markus Nagel, Marios Fournarakis, Rana Ali Amjad, Yelysei Bon- darenko, Mart Van Baalen, and Tijmen Blankevoort. A white paper on neural network quantization. In arXiv.org, 2021

    2021

  26. [31]

    Qllm: Accurate and efficient low-bitwidth quantization for large language models

    Jing Liu, Ruihao Gong, Xiuying Wei, Zhiwei Dong, Jianfei Cai, and Bohan Zhuang. Qllm: Accurate and efficient low-bitwidth quantization for large language models. In arXiv.org, 2023

    2023

  27. [32]

    Omniquant: Omnidirectionally calibrated quantization for large language models

    Wenqi Shao, Mengzhao Chen, Zhaoyang Zhang, Peng Xu, Lirui Zhao, Zhiqian Li, Kaipeng Zhang, Peng Gao, Yu Qiao, and Ping Luo. Omniquant: Omnidirectionally calibrated quantization for large language models. In arXiv.org, 2023

    2023

  28. [33]

    Bitwave: Exploiting column-based bit-level sparsity for deep learning acceleration

    Man Shi, Vikram Jain, Antony Joseph, Maurice Meijer, and Marian Verhelst. Bitwave: Exploiting column-based bit-level sparsity for deep learning acceleration. In Proc. the International Symposium on High- Performance Computer Architecture (HPCA) , pages 732–746, 2024

    2024

  29. [34]

    Cambricon-X: An accelerator for sparse neural networks

    Shijin Zhang, Zidong Du, Lei Zhang, Huiying Lan, Shaoli Liu, Ling Li, Qi Guo, Tianshi Chen, and Yunji Chen. Cambricon-X: An accelerator for sparse neural networks. In Proc. International Symposium on Microarchitecture (MICRO), pages 1–12, 2016

    2016

  30. [35]

    Sparse tensor core: Algorithm and hardware co-design for vector-wise sparse neural networks on modern gpus

    Maohua Zhu, Tao Zhang, Zhenyu Gu, and Yuan Xie. Sparse tensor core: Algorithm and hardware co-design for vector-wise sparse neural networks on modern gpus. In Proc. International Symposium on Microarchitecture (MICRO), pages 359–371, 2019

    2019

  31. [36]

    Learning both weights and connections for efficient neural network

    Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In Proc. Advances in neural information processing systems (NeurIPs) , volume 28, 2015

    2015

  32. [37]

    An algorithm–hardware co-optimized framework for accelerating n: M sparse transformers

    Chao Fang, Aojun Zhou, and Zhongfeng Wang. An algorithm–hardware co-optimized framework for accelerating n: M sparse transformers. IEEE Trans. Very Large Scale Integr. (VLSI) Syst. , 30(11):1573–1586, 2022

    2022

  33. [38]

    Energy-efficient risc-v-based vector processor for cache- aware structurally-pruned transformers

    Jung Gyu Min, Dongyun Kam, Younghoon Byun, Gunho Park, and Youngjoo Lee. Energy-efficient risc-v-based vector processor for cache- aware structurally-pruned transformers. In Proc. IEEE/ACM Interna- tional Symposium on Low Power Electronics and Design (ISLPED) , pages 1–6, 2023

    2023

  34. [39]

    Sparsity-Aware Memory Interface Architecture using Stacked XORNet Compression for Accelerating Pruned-DNN Models

    Younghoon Byun, Seungsik Moon, Baeseong Park, Se Jung Kwon, Dongsoo Lee, Gunho Park, Eunji Yoo, Jung Gyu Min, and Youngjoo Lee. Sparsity-Aware Memory Interface Architecture using Stacked XORNet Compression for Accelerating Pruned-DNN Models. Proc. Machine Learning and Systems ...

    2023

  35. [40]

    Sparsegpt: Massive language models can be accurately pruned in one-shot

    Elias Frantar and Dan Alistarh. Sparsegpt: Massive language models can be accurately pruned in one-shot. In Proc. International conference on machine learning (ICML) , pages 10323–10337, 2023

    2023

  36. [41]

    I-bert: Integer-only bert quantization

    Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. I-bert: Integer-only bert quantization. In Proc. Interna- tional conference on machine learning (ICML), pages 5506–5518, 2021

    2021

  37. [42]

    Quantization and training of neural networks for efficient integer- arithmetic-only inference

    Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard, Hartwig Adam, and Dmitry Kalenichenko. Quantization and training of neural networks for efficient integer- arithmetic-only inference. In Proc. IEEE Conference on Computer Vision and Pattern Re...

    2018

  38. [43]

    Olive: Accelerating large language models via hardware-friendly outlier-victim pair quantization

    Cong Guo, Jiaming Tang, Weiming Hu, Jingwen Leng, Chen Zhang, Fan Yang, Yunxin Liu, Minyi Guo, and Yuhao Zhu. Olive: Accelerating large language models via hardware-friendly outlier-victim pair quantization. In Proc. International Symposium on Computer Architecture (ISCA) , pa...

    2023

  39. [44]

    Vs-quant: Per-vector scaled quantization for accurate low-precision neural network inference

    Steve Dai, Rangha Venkatesan, Mark Ren, Brian Zimmer, William Dally, and Brucek Khailany. Vs-quant: Per-vector scaled quantization for accurate low-precision neural network inference. In Proc. Machine Learning and Systems (MLSys) , volume 3, pages 873–884, 2021

    2021

  40. [45]

    Gobo: Quantizing attention-based nlp models for low latency and energy efficient inference

    Ali Hadi Zadeh, Isak Edo, Omar Mohamed Awad, and Andreas Moshovos. Gobo: Quantizing attention-based nlp models for low latency and energy efficient inference. In Proc. International Symposium on Microarchitecture (MICRO), pages 811–824, 2020

    2020

  41. [46]

    OPTQ: Accurate quantization for generative pre-trained transformers

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. OPTQ: Accurate quantization for generative pre-trained transformers. In Proc. International Conference on Learning Representations (ICLR) , 2022

    2022

  42. [47]

    Rptq: Reorder-based post-training quantization for large language models

    Zhihang Yuan, Lin Niu, Jiawei Liu, Wenyu Liu, Xinggang Wang, Yuzhang Shang, Guangyu Sun, Qiang Wu, Jiaxiang Wu, and Bingzhe Wu. Rptq: Reorder-based post-training quantization for large language models. In arXiv.org, 2023

    2023

  43. [48]

    Ant: Exploiting adaptive numerical data type for low-bit deep neural network quantization

    Cong Guo, Chen Zhang, Jingwen Leng, Zihan Liu, Fan Yang, Yunxin Liu, Minyi Guo, and Yuhao Zhu. Ant: Exploiting adaptive numerical data type for low-bit deep neural network quantization. In Proc. International Symposium on Microarchitecture (MICRO) , pages 1414– 1433, 2022

    2022

  44. [49]

    Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients

    Shuchang Zhou, Yuxin Wu, Zekun Ni, Xinyu Zhou, He Wen, and Yuheng Zou. Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients. In arXiv.org, 2016

    2016

  45. [50]

    Noisyquant: Noisy bias-enhanced post-training activation quantization for vision transformers

    Yijiang Liu, Huanrui Yang, Zhen Dong, Kurt Keutzer, Li Du, and Shang- hang Zhang. Noisyquant: Noisy bias-enhanced post-training activation quantization for vision transformers. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 20321–20330, 2023

    2023

  46. [51]

    Ptq4vit: Post-training quantization for vision transformers with twin uniform quantization

    Zhihang Yuan, Chenhao Xue, Yiqi Chen, Qiang Wu, and Guangyu Sun. Ptq4vit: Post-training quantization for vision transformers with twin uniform quantization. In Proc. the European conference on computer vision (ECCV), pages 191–207, 2022

    2022

  47. [52]

    Qserve: W4a8kv4 quantization and system co-design for efficient llm serving

    Yujun Lin, Haotian Tang, Shang Yang, Zhekai Zhang, Guangxuan Xiao, Chuang Gan, and Song Han. Qserve: W4a8kv4 quantization and system co-design for efficient llm serving. 2024

    2024

  48. [53]

    Sibia: Signed bit-slice architecture for dense dnn acceleration with slice-level sparsity exploitation

    Dongseok Im, Gwangtae Park, Zhiyong Li, Junha Ryu, and Hoi-Jun Yoo. Sibia: Signed bit-slice architecture for dense dnn acceleration with slice-level sparsity exploitation. In Proc. the International Symposium on High-Performance Computer Architecture (HPCA), pages 69–80, 2023

    2023

  49. [55]

    Hnpu-v1: An adaptive dnn train- ing processor utilizing stochastic dynamic fixed-point and active bit- precision searching

    Donghyeon Han and Hoi-Jun Yoo. Hnpu-v1: An adaptive dnn train- ing processor utilizing stochastic dynamic fixed-point and active bit- precision searching. In On-Chip Training NPU-Algorithm, Architecture and SoC Design , pages 121–161. Springer, 2023

    2023

  50. [56]

    Lutein: Dense-sparse bit-slice architec- ture with radix-4 lut-based slice-tensor processing units

    Dongseok Im and Hoi-Jun Yoo. Lutein: Dense-sparse bit-slice architec- ture with radix-4 lut-based slice-tensor processing units. In Proc. the International Symposium on High-Performance Computer Architecture (HPCA), pages 747–759, 2024

    2024

  51. [57]

    Embedded deep neural network processing: Algorithmic and processor techniques bring deep learning to iot and edge devices

    Marian Verhelst and Bert Moons. Embedded deep neural network processing: Algorithmic and processor techniques bring deep learning to iot and edge devices. IEEE Solid-State Circuits Mag. , 9(4):55–65, 2017

    2017

  52. [58]

    MEISSA: Multi- plying matrices efficiently in a scalable systolic architecture

    Bahar Asgari, Ramyad Hadidi, and Hyesoon Kim. MEISSA: Multi- plying matrices efficiently in a scalable systolic architecture. In Proc. International Conference on Computer Design (ICCD) , pages 130–137, 2020

    2020

  53. [59]

    A 95.6-TOPS/W deep learning inference accelerator with per- vector scaled 4-bit quantization in 5 nm

    Ben Keller, Rangharajan Venkatesan, Steve Dai, Stephen G Tell, Brian Zimmer, Charbel Sakr, William J Dally, C Thomas Gray, and Brucek Khailany. A 95.6-TOPS/W deep learning inference accelerator with per- vector scaled 4-bit quantization in 5 nm. IEEE J. Solid-State Circuits , ...

    2023

  54. [60]

    Opt: Open pre-trained transformer language models

    Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. O...

    2022

  55. [61]

    Llama 3.2: Revolutionizing edge ai and vision with open, customizable models

    Meta AI. Llama 3.2: Revolutionizing edge ai and vision with open, customizable models. https://ai.meta.com/blog/llama-3-2-connect-2024- vision-edge-mobile-devices, 2024

    2024

  56. [62]

    Xtc: Extreme compression for pre-trained transformers made simple and efficient

    Xiaoxia Wu, Zhewei Yao, Minjia Zhang, Conglong Li, and Yuxiong He. Xtc: Extreme compression for pre-trained transformers made simple and efficient. In Proc. Advances in neural information processing systems (NeurIPs), volume 35, pages 3217–3231, 2022

    2022

  57. [63]

    Energy-efficient neural network accelerator based on outlier-aware low-precision com- putation

    Eunhyeok Park, Dongyoung Kim, and Sungjoo Yoo. Energy-efficient neural network accelerator based on outlier-aware low-precision com- putation. In Proc. International Symposium on Computer Architecture (ISCA), pages 688–698, 2018

    2018

  58. [64]

    Drq: dynamic region-based quan- tization for deep neural network acceleration

    Zhuoran Song, Bangqi Fu, Feiyang Wu, Zhaoming Jiang, Li Jiang, Naifeng Jing, and Xiaoyao Liang. Drq: dynamic region-based quan- tization for deep neural network acceleration. In Proc. International Symposium on Computer Architecture (ISCA) , pages 1010–1021, 2020

    2020

  59. [65]

    FIGNA: Integer Unit-Based Accelerator Design for FP-INT GEMM Preserving Numerical Accuracy

    Jaeyong Jang, Yulhwa Kim, Juheun Lee, and Jae-Joon Kim. FIGNA: Integer Unit-Based Accelerator Design for FP-INT GEMM Preserving Numerical Accuracy. In Proc. the International Symposium on High- Performance Computer Architecture (HPCA) , pages 760–773, 2024

    2024

  60. [66]

    Loss aware post- training quantization

    Yury Nahshan, Brian Chmiel, Chaim Baskin, Evgenii Zheltonozhskii, Ron Banner, Alex M Bronstein, and Avi Mendelson. Loss aware post- training quantization. Mach. Learn., 110(11):3245–3262, 2021. 14

    2021

  61. [67]

    Efficient processing of deep neural networks: A tutorial and survey

    Vivienne Sze, Yu-Hsin Chen, Tien-Ju Yang, and Joel S Emer. Efficient processing of deep neural networks: A tutorial and survey. Proc. IEEE, 105(12):2295–2329, 2017

    2017

  62. [68]

    A survey of quantization methods for efficient neural network inference

    Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. A survey of quantization methods for efficient neural network inference. In Low-Power Computer Vision , pages 291–326. Chapman and Hall/CRC, 2022

    2022

  63. [69]

    Accurate post training quantization with small calibration sets

    Itay Hubara, Yury Nahshan, Yair Hanani, Ron Banner, and Daniel Soudry. Accurate post training quantization with small calibration sets. In Proc. International conference on machine learning (ICML) , pages 4466–4475, 2021

    2021

  64. [70]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proc. International Conference on Artificial Intelligence and Statistics (AISTATS) , pages 249–256, 2010

    2010

  65. [71]

    Delving deep into rectifiers: Surpassing human-level performance on imagenet classification

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proc. International Conference on Computer Vision (ICCV), pages 1026–1034, 2015

    2015

  66. [72]

    Efficient backprop

    Yann LeCun, L ´eon Bottou, Genevieve B Orr, and Klaus-Robert M ¨uller. Efficient backprop. In Neural networks: Tricks of the trade, pages 9–50. Springer, 2002

    2002

  67. [73]

    Probact: A probabilistic activation function for deep neural networks

    Kumar Shridhar, Joonho Lee, Hideaki Hayashi, Purvanshi Mehta, Brian Kenji Iwana, Seokjun Kang, Seiichi Uchida, Sheraz Ahmed, and Andreas Dengel. Probact: A probabilistic activation function for deep neural networks. In arxiv.org, 2019

    2019

  68. [74]

    Bayesian learning for neural networks , volume 118

    Radford M Neal. Bayesian learning for neural networks , volume 118. Springer Science & Business Media, 2012

    2012

  69. [76]

    Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural network

    Hardik Sharma, Jongse Park, Naveen Suda, Liangzhen Lai, Benson Chau, Joon Kyung Kim, Vikas Chandra, and Hadi Esmaeilzadeh. Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural network. In Proc. International Symposium on Computer Architecture ...

    2018

  70. [77]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei- Fei. Imagenet: A large-scale hierarchical image database. In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 248–255, 2009

    2009

  71. [78]

    Awq: Activation-aware weight quantization for on-device llm compression and acceleration

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei- Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. Awq: Activation-aware weight quantization for on-device llm compression and acceleration. Proc. Machine Learning and Systems (MLSys), 6:87–100, 2024

    2024

  72. [79]

    CACTI 7: New tools for interconnect exploration in innovative off-chip memories

    Rajeev Balasubramonian, Andrew B Kahng, Naveen Muralimanohar, Ali Shafiee, and Vaishnav Srinivas. CACTI 7: New tools for interconnect exploration in innovative off-chip memories. ACM Transactions on Architecture and Code Optimization (TACO) , 14(2):1–25, 2017

    2017

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.