parent_paper_title
stringclasses
63 values
parent_paper_arxiv_id
stringclasses
63 values
citation_shorthand
stringlengths
2
56
raw_citation_text
stringlengths
9
63
cited_paper_title
stringlengths
5
161
cited_paper_arxiv_link
stringlengths
32
37
cited_paper_abstract
stringlengths
406
1.92k
has_metadata
bool
1 class
is_arxiv_paper
bool
2 classes
bib_paper_authors
stringlengths
2
2.44k
bib_paper_year
float64
1.97k
2.03k
bib_paper_month
stringclasses
16 values
bib_paper_url
stringlengths
20
116
bib_paper_doi
stringclasses
269 values
bib_paper_journal
stringlengths
3
148
original_title
stringlengths
5
161
search_res_title
stringlengths
4
122
search_res_url
stringlengths
22
267
search_res_content
stringlengths
19
1.92k
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
cnn_pruning
\cite{cnn_pruning}
Pruning convolutional neural networks for resource efficient inference
null
null
true
false
Molchanov, Pavlo and Tyree, Stephen and Karras, Tero and Aila, Timo and Kautz, Jan
2,016
null
null
null
arXiv preprint arXiv:1611.06440
Pruning convolutional neural networks for resource efficient inference
Pruning Convolutional Neural Networks for Resource Efficient Inference
http://arxiv.org/pdf/1611.06440v2
We propose a new formulation for pruning convolutional kernels in neural networks to enable efficient inference. We interleave greedy criteria-based pruning with fine-tuning by backpropagation - a computationally efficient procedure that maintains good generalization in the pruned network. We propose a new criterion ba...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
Pruner-Zero
\cite{Pruner-Zero}
Pruner-Zero: Evolving Symbolic Pruning Metric From Scratch for Large Language Models
null
null
true
false
Peijie Dong and Lujun Li and Zhenheng Tang and Xiang Liu and Xinglin Pan and Qiang Wang and Xiaowen Chu
2,024
null
https://openreview.net/forum?id=1tRLxQzdep
null
null
Pruner-Zero: Evolving Symbolic Pruning Metric From Scratch for Large Language Models
Pruner-Zero: Evolving Symbolic Pruning Metric from scratch for LLMs
https://github.com/pprp/Pruner-Zero
GitHub - pprp/Pruner-Zero: [ICML24] Pruner-Zero: Evolving Symbolic Pruning Metric from scratch for LLMs [ICML24] Pruner-Zero: Evolving Symbolic Pruning Metric from scratch for LLMs | main.py | main.py | Update rest code at once | Jun 6, 2024 | | main_opt.py | main_opt.py | Update rest code at once | Jun 6, 2024 | **Pru...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
SparseGPT
\cite{SparseGPT}
SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot
http://arxiv.org/abs/2301.00774v3
We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy. This is achieved via a new pruning method called SparseGPT, specifically designed to work efficiently and accurately ...
true
true
Elias Frantar and Dan Alistarh
2,023
null
https://proceedings.mlr.press/v202/frantar23a.html
null
null
SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot
SparseGPT: Massive Language Models Can Be Accurately ...
https://arxiv.org/abs/2301.00774
by E Frantar · 2023 · Cited by 887 — We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot.See more
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
DistiLLM
\cite{DistiLLM}
DistiLLM: Towards Streamlined Distillation for Large Language Models
http://arxiv.org/abs/2402.03898v2
Knowledge distillation (KD) is widely used for compressing a teacher model to a smaller student model, reducing its inference cost and memory footprint while preserving model capabilities. However, current KD methods for auto-regressive sequence models (e.g., large language models) suffer from missing a standardized ob...
true
true
Jongwoo Ko and Sungnyun Kim and Tianyi Chen and Se{-}Young Yun
2,024
null
https://openreview.net/forum?id=lsHZNNoC7r
null
null
DistiLLM: Towards Streamlined Distillation for Large Language Models
DistiLLM: Towards Streamlined Distillation for Large Language Models
http://arxiv.org/pdf/2402.03898v2
Knowledge distillation (KD) is widely used for compressing a teacher model to a smaller student model, reducing its inference cost and memory footprint while preserving model capabilities. However, current KD methods for auto-regressive sequence models (e.g., large language models) suffer from missing a standardized ob...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
MiniLLM
\cite{MiniLLM}
MiniLLM: Knowledge Distillation of Large Language Models
http://arxiv.org/abs/2306.08543v4
Knowledge Distillation (KD) is a promising technique for reducing the high computational demand of large language models (LLMs). However, previous KD methods are primarily applied to white-box classification models or training small models to imitate black-box model APIs like ChatGPT. How to effectively distill the kno...
true
true
Yuxian Gu and Li Dong and Furu Wei and Minlie Huang
2,024
null
https://openreview.net/forum?id=5h0qf7IBZZ
null
null
MiniLLM: Knowledge Distillation of Large Language Models
MiniLLM: Knowledge Distillation of Large Language Models
http://arxiv.org/pdf/2306.08543v4
Knowledge Distillation (KD) is a promising technique for reducing the high computational demand of large language models (LLMs). However, previous KD methods are primarily applied to white-box classification models or training small models to imitate black-box model APIs like ChatGPT. How to effectively distill the kno...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
pytorch-1
\cite{pytorch-1}
PyTorch: An Open Source Machine Learning Framework
null
null
true
false
{PyTorch Contributors}
2,024
null
https://pytorch.org/
null
null
PyTorch: An Open Source Machine Learning Framework
PyTorch
https://en.wikipedia.org/wiki/PyTorch
PyTorch is a machine learninglibrary based on the Torch library,[4][5][6] used for applications such as computer vision and natural language processing,[7] originally developed by Meta AI and now part of the Linux Foundation umbrella.[8][9][10][11] It is one of the most popular deep learning frameworks, alongside other...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
tensorflow
\cite{tensorflow}
TensorFlow: An Open Source Machine Learning Framework for Everyone
null
null
true
false
{TensorFlow Contributors}
2,024
null
https://www.tensorflow.org/
null
null
TensorFlow: An Open Source Machine Learning Framework for Everyone
tensorflow/tensorflow: An Open Source Machine Learning ...
https://github.com/tensorflow/tensorflow
TensorFlow is an end-to-end open source platform for machine learning. It has a comprehensive, flexible ecosystem of tools, libraries, and community resources.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
deepspeed
\cite{deepspeed}
DeepSpeed: Advancing the Science of AI Through Efficient Training of Large Models
null
null
true
false
{Microsoft DeepSpeed Team}
2,024
null
https://www.deepspeed.ai/
null
null
DeepSpeed: Advancing the Science of AI Through Efficient Training of Large Models
DeepSpeed: Latest News
https://www.deepspeed.ai/
DeepSpeed is a deep learning optimization library that makes distributed training easy, efficient, and effective.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
huggingface_transformers
\cite{huggingface_transformers}
Transformers Documentation
null
null
true
false
{HuggingFace Team}
2,024
null
https://huggingface.co/docs/transformers/index
null
null
Transformers Documentation
Transformers — transformers 3.0.2 documentation - Hugging Face
https://huggingface.co/transformers/v3.0.2/index.html
Transformers is a state-of-the-art NLP library for Pytorch and TensorFlow 2.0, providing architectures for NLU and NLG with high performance and low barrier to
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
flexgen
\cite{flexgen}
FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU
http://arxiv.org/abs/2303.06865v2
The high computational and memory requirements of large language model (LLM) inference make it feasible only with multiple high-end accelerators. Motivated by the emerging demand for latency-insensitive tasks with batched processing, this paper initiates the study of high-throughput LLM inference using limited resource...
true
true
Sheng, Ying and Zheng, Lianmin and Yuan, Binhang and Li, Zhuohan and Ryabinin, Max and Chen, Beidi and Liang, Percy and R{\'e}, Christopher and Stoica, Ion and Zhang, Ce
2,023
null
null
null
null
FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU
[PDF] FlexGen: High-Throughput Generative Inference of Large Language ...
https://openreview.net/pdf?id=RRntzKrBTp
FlexGen is a high-throughput engine for running LLMs with limited GPU memory, using GPU, CPU, and disk, and achieving 1 token/s throughput.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
Orca
\cite{Orca}
Orca: {A} Distributed Serving System for Transformer-Based Generative Models
null
null
true
false
Gyeong{-}In Yu and Joo Seong Jeong and Geon{-}Woo Kim and Soojeong Kim and Byung{-}Gon Chun
2,022
null
https://www.usenix.org/conference/osdi22/presentation/yu
null
null
Orca: {A} Distributed Serving System for Transformer-Based Generative Models
Orca: A Distributed Serving System for Transformer-Based ...
https://www.usenix.org/conference/osdi22/presentation/yu
+ Poster Session # Orca: A Distributed Serving System for Transformer-Based Generative Models In this paper, we propose iteration-level scheduling, a new scheduling mechanism that schedules execution at the granularity of iteration (instead of request) where the scheduler invokes the execution engine to run only a sing...
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao4
\cite{zhao4}
A Reconfigurable 0.69-1.02 nJ/Classification Biomedical AI Processor for Intelligent Health Monitoring Devices
null
null
true
false
Zhao, Yuanzhe and Wang, Yuheng and Wang, Zijian and Zhu, Yan and Martins, RP and Chan, Chi-Hang and Zhang, Minglei
2,025
null
null
null
null
A Reconfigurable 0.69-1.02 nJ/Classification Biomedical AI Processor for Intelligent Health Monitoring Devices
4.5 BioAIP: A Reconfigurable Biomedical AI Processor with Adaptive ...
https://www.researchgate.net/publication/350171989_45_BioAIP_A_Reconfigurable_Biomedical_AI_Processor_with_Adaptive_Learning_for_Versatile_Intelligent_Health_Monitoring
A Reconfigurable 0.69-1.02nJ/Classification Biomedical AI Processor for Intelligent Health Monitoring Devices. Conference Paper. Apr 2025. Yuanzhe Zhao · Yuheng
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao5
\cite{zhao5}
A 28nm Value-Wise Hybrid-Domain Compute-in-Memory Macro with Heterogeneous Memory Fabric and Asynchronous Sparsity Manager
null
null
true
false
Zhao, Yuanzhe and Wang, Yang and Wang, Yuheng and Xie, Heng and Zhu, Yan and Martins, RP and Chan, Chi-Hang and Yin, Shouyi and Zhang, Minglei
2,025
null
null
null
null
A 28nm Value-Wise Hybrid-Domain Compute-in-Memory Macro with Heterogeneous Memory Fabric and Asynchronous Sparsity Manager
A 28nm Value-Wise Hybrid-Domain Compute-in-Memory ...
https://unpaywall.org/10.1109%2FCICC63670.2025.10982876
Abstract: Edge AI devices need to be energy-efficient and compact while maintaining sufficient accuracy. Compute-in-memory (CIM) is a promising approach to
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao6
\cite{zhao6}
A One-Shot Floating-Point Compute-in-Memory Macro Featuring PVT Robustness and Mismatch Tolerance for Edge LLMs
null
null
true
false
Zhao, Yuanzhe and Xie, Heng and Wang, Zijian and Tian, Chunlin and Li, Li and Zhu, Yan and Martins, RP and Chan, Chi-Hang and Zhang, Minglei
2,025
null
null
null
null
A One-Shot Floating-Point Compute-in-Memory Macro Featuring PVT Robustness and Mismatch Tolerance for Edge LLMs
A One-Shot Floating-Point Compute-in-Memory Macro ...
https://www.researchgate.net/publication/391898058_A_One-Shot_Floating-Point_Compute-in-Memory_Macro_Featuring_PVT_Robustness_and_Mismatch_Tolerance_for_Edge_LLMs
Robustness. Conference Paper. A One-Shot Floating-Point Compute-in-Memory Macro Featuring PVT Robustness and Mismatch Tolerance for Edge LLMs.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao7
\cite{zhao7}
A 28-nm 3.32-nJ/Frame Compute-in-Memory CNN Processor With Layer Fusion for Always-on Applications
null
null
true
false
Zhao, Yuanzhe and He, Pengyu and Zhu, Yan and Martins, Rui P and Chan, Chi-Hang and Zhang, Minglei
2,025
null
null
null
IEEE Transactions on Circuits and Systems I: Regular Papers
A 28-nm 3.32-nJ/Frame Compute-in-Memory CNN Processor With Layer Fusion for Always-on Applications
A 28-nm 3.32-nJ/Frame Compute-in-Memory CNN Processor With ...
https://ieeexplore.ieee.org/document/10902457/
This work presents an always-on CNN processor featuring compute-in-memory (CIM) and layer-fusion (LF) techniques. It demonstrates an end-to-end neural network
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao8
\cite{zhao8}
A Reconfigurable Floating-Point Compute-In-Memory With Analog Exponent Pre-Processes
null
null
true
false
He, Pengyu and Zhao, Yuanzhe and Xie, Heng and Wang, Yang and Yin, Shouyi and Li, Li and Zhu, Yan and Martins, Rui P and Chan, Chi-Hang and Zhang, Minglei
2,024
null
null
null
IEEE Solid-State Circuits Letters
A Reconfigurable Floating-Point Compute-In-Memory With Analog Exponent Pre-Processes
A Reconfigurable Floating-Point Compute-in-Memory With ...
http://ieeexplore.ieee.org/document/10683795/
This letter presents a reconfigurable floating-point compute-in-memory (FP-CIM) macro that preprocesses the exponent in the analog domain.See more
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
tambe2021edgebert
\cite{tambe2021edgebert}
EdgeBERT: Sentence-Level Energy Optimizations for Latency-Aware Multi-Task NLP Inference
http://arxiv.org/abs/2011.14203v5
Transformer-based language models such as BERT provide significant accuracy improvement for a multitude of natural language processing (NLP) tasks. However, their hefty computational and memory demands make them challenging to deploy to resource-constrained edge platforms with strict latency requirements. We present Ed...
true
true
Tambe, Thierry and Hooper, Coleman and Pentecost, Lillian and Jia, Tianyu and Yang, En-Yu and Donato, Marco and Sanh, Victor and Whatmough, Paul and Rush, Alexander M and Brooks, David and others
2,021
null
null
null
null
EdgeBERT: Sentence-Level Energy Optimizations for Latency-Aware Multi-Task NLP Inference
Sentence-Level Energy Optimizations for Latency-Aware ...
https://dl.acm.org/doi/10.1145/3466752.3480095
We present EdgeBERT, an in-depth algorithm-hardware co-design for latency-aware energy optimizations for multi-task NLP.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao2023approxcaliper
\cite{zhao2023approxcaliper}
Approxcaliper: A programmable framework for application-aware neural network optimization
null
null
true
false
Zhao, Yifan and Sharif, Hashim and Pao-Huang, Peter and Shah, Vatsin and Sivakumar, Arun Narenthiran and Valverde Gasparino, Mateus and Mahmoud, Abdulrahman and Zhao, Nathan and Adve, Sarita and Chowdhary, Girish and others
2,023
null
null
null
Proceedings of Machine Learning and Systems
Approxcaliper: A programmable framework for application-aware neural network optimization
ApproxCaliper: A Programmable Framework for ...
https://ma3mool.github.io/publication/mlsys23.html
"ApproxCaliper: A Programmable Framework for Application-aware Neural Network Optimization," Sixth Conference on Machine Learning and Systems (MLSys), Miami,
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
liberis2023differentiable
\cite{liberis2023differentiable}
Differentiable neural network pruning to enable smart applications on microcontrollers
null
null
true
false
Liberis, Edgar and Lane, Nicholas D
2,023
null
null
null
Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
Differentiable neural network pruning to enable smart applications on microcontrollers
Differentiable Neural Network Pruning to Enable Smart Applications ...
https://dl.acm.org/doi/abs/10.1145/3569468
We present a differentiable structured pruning method for convolutional neural networks, which integrates a model's MCU-specific resource usage and parameter
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
zhao2024felix
\cite{zhao2024felix}
Felix: Optimizing Tensor Programs with Gradient Descent
null
null
true
false
Zhao, Yifan and Sharif, Hashim and Adve, Vikram and Misailovic, Sasa
2,024
null
null
null
null
Felix: Optimizing Tensor Programs with Gradient Descent
Felix: Optimizing Tensor Programs with Gradient Descent
https://misailo.cs.illinois.edu/papers/felix-asplos24.pdf
Felix creates a differentiable space of tensor programs that is amenable to search by gradient descent. Felix applies continuous re- laxation on the space of
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
tam2024fedhybrid
\cite{tam2024fedhybrid}
FedHybrid: Breaking the Memory Wall of Federated Learning via Hybrid Tensor Management
null
null
true
false
Tam, Kahou and Tian, Chunlin and Li, Li and Zhao, Haikai and Xu, ChengZhong
2,024
null
null
null
null
FedHybrid: Breaking the Memory Wall of Federated Learning via Hybrid Tensor Management
Breaking the Memory Wall of Federated Learning via Hybrid Tensor ...
https://www.researchgate.net/publication/385538059_FedHybrid_Breaking_the_Memory_Wall_of_Federated_Learning_via_Hybrid_Tensor_Management
FedHybrid: Breaking the Memory Wall of Federated Learning via Hybrid Tensor Management ... Wall for Heterogeneous Federated Learning via Progressive Training.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
dvfsasplos
\cite{dvfsasplos}
Expanding Datacenter Capacity with {DVFS} Boosting: {A} safe and scalable deployment experience
null
null
true
false
Leonardo Piga and Iyswarya Narayanan and Aditya Sundarrajan and Matt Skach and Qingyuan Deng and Biswadip Maity and Manoj Chakkaravarthy and Alison Huang and Abhishek Dhanotia ...
2,024
null
https://doi.org/10.1145/3617232.3624853
10.1145/3617232.3624853
null
Expanding Datacenter Capacity with {DVFS} Boosting: {A} safe and scalable deployment experience
Expanding Datacenter Capacity with DVFS Boosting
https://www.researchgate.net/publication/379917331_Expanding_Datacenter_Capacity_with_DVFS_Boosting_A_safe_and_scalable_deployment_experience
[35] DVFS Boosting was explored as a scalable and secure approach to enhance data center capacity, tackling power consumption, hardware heterogeneity, and ...See more
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
dfvs-4
\cite{dfvs-4}
Improving {DVFS} in NoCs with Coherence Prediction
null
null
true
false
Robert Hesse and Natalie D. Enright Jerger
2,015
null
https://doi.org/10.1145/2786572.2786595
10.1145/2786572.2786595
null
Improving {DVFS} in NoCs with Coherence Prediction
Improving DVFS in NoCs with Coherence Prediction
https://dl.acm.org/doi/10.1145/2786572.2786595
In this work, we propose to utilize highly predictable properties of cache-coherence communication to derive more specific and reliable NoC traffic predictions.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
dvfs-2
\cite{dvfs-2}
Variation-aware dynamic voltage/frequency scaling
null
null
true
false
Sebastian Herbert and Diana Marculescu
2,009
null
https://doi.org/10.1109/HPCA.2009.4798265
10.1109/HPCA.2009.4798265
null
Variation-aware dynamic voltage/frequency scaling
Variation-aware dynamic voltage/frequency scaling
https://ieeexplore.ieee.org/document/4798265/
by S Herbert · 2009 · Cited by 174 — Fine-grained dynamic voltage/frequency scaling (DVFS) is an important tool in managing the balance between power and performance in chip-multiprocessors.
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
dvfs-3
\cite{dvfs-3}
System level analysis of fast, per-core {DVFS} using on-chip switching regulators
null
null
true
false
Wonyoung Kim and Meeta Sharma Gupta and Gu{-}Yeon Wei and David M. Brooks
2,008
null
https://doi.org/10.1109/HPCA.2008.4658633
10.1109/HPCA.2008.4658633
null
System level analysis of fast, per-core {DVFS} using on-chip switching regulators
System Level Analysis of Fast, Per-Core DVFS Using On- ...
https://www.slideserve.com/malise/system-level-analysis-of-fast-per-core-dvfs-using-on-chip-switching-regulators
System Level Analysis of Fast, Per-Core DVFS Using On-Chip Switching Regulators. Wonyoung Kim, Meeta Gupta Prof. Gu-Yeon Wei, Prof. David Brooks
CLONE: Customizing LLMs for Efficient Latency-Aware Inference at the Edge
2506.02847v1
bateni2020neuos
\cite{bateni2020neuos}
$\{$NeuOS$\}$: A $\{$Latency-Predictable$\}$$\{$Multi-Dimensional$\}$ Optimization Framework for $\{$DNN-driven$\}$ Autonomous Systems
null
null
true
false
Bateni, Soroush and Liu, Cong
2,020
null
null
null
null
$\{$NeuOS$\}$: A $\{$Latency-Predictable$\}$$\{$Multi-Dimensional$\}$ Optimization Framework for $\{$DNN-driven$\}$ Autonomous Systems
Soroush Bateni - DBLP
https://dblp.org/pid/224/5652
NeuOS: A Latency-Predictable Multi-Dimensional Optimization Framework for DNN-driven Autonomous Systems. USENIX ATC 2020: 371-385. [i2]. view.
Refining Datapath for Microscaling ViTs
2505.22194v1
lin2017accurate
\cite{lin2017accurate}
Towards Accurate Binary Convolutional Neural Network
http://arxiv.org/abs/1711.11294v1
We introduce a novel scheme to train binary convolutional neural networks (CNNs) -- CNNs with weights and activations constrained to {-1,+1} at run-time. It has been known that using binary weights and activations drastically reduce memory size and accesses, and can replace arithmetic operations with more efficient bit...
true
true
Xiaofan Lin and Cong Zhao and Wei Pan
2,017
null
null
null
null
Towards Accurate Binary Convolutional Neural Network
Towards Accurate Binary Convolutional Neural Network
http://arxiv.org/pdf/1711.11294v1
We introduce a novel scheme to train binary convolutional neural networks (CNNs) -- CNNs with weights and activations constrained to {-1,+1} at run-time. It has been known that using binary weights and activations drastically reduce memory size and accesses, and can replace arithmetic operations with more efficient bit...
Refining Datapath for Microscaling ViTs
2505.22194v1
zhang2018lqnets
\cite{zhang2018lqnets}
LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
null
null
true
false
Dongqing Zhang and Jiaolong Yang and Dongqiangzi Ye and Gang Hua
2,018
null
null
null
null
LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
http://arxiv.org/pdf/1807.10029v1
Although weight and activation quantization is an effective approach for Deep Neural Network (DNN) compression and has a lot of potentials to increase inference speed leveraging bit-operations, there is still a noticeable gap in terms of prediction accuracy between the quantized model and the full-precision model. To a...
Refining Datapath for Microscaling ViTs
2505.22194v1
wu2018training
\cite{wu2018training}
Training and Inference with Integers in Deep Neural Networks
http://arxiv.org/abs/1802.04680v1
Researches on deep neural networks with discrete parameters and their deployment in embedded systems have been active and promising topics. Although previous works have successfully reduced precision in inference, transferring both training and inference processes to low-bitwidth integers has not been demonstrated simu...
true
true
Wu, Shuang and Li, Guoqi and Chen, Feng and Shi, Luping
2,018
null
null
null
arXiv preprint arXiv:1802.04680
Training and Inference with Integers in Deep Neural Networks
Training and Inference with Integers in Deep Neural Networks
http://arxiv.org/pdf/1802.04680v1
Researches on deep neural networks with discrete parameters and their deployment in embedded systems have been active and promising topics. Although previous works have successfully reduced precision in inference, transferring both training and inference processes to low-bitwidth integers has not been demonstrated simu...
Refining Datapath for Microscaling ViTs
2505.22194v1
krishnamoorthi2018quantizing
\cite{krishnamoorthi2018quantizing}
Quantizing deep convolutional networks for efficient inference: A whitepaper
null
null
true
false
Krishnamoorthi, Raghuraman
2,018
null
null
null
arXiv preprint arXiv:1806.08342
Quantizing deep convolutional networks for efficient inference: A whitepaper
Quantizing deep convolutional networks for efficient inference: A whitepaper
http://arxiv.org/pdf/1806.08342v1
We present an overview of techniques for quantizing convolutional neural networks for inference with integer weights and activations. Per-channel quantization of weights and per-layer quantization of activations to 8-bits of precision post-training produces classification accuracies within 2% of floating point networks...
Refining Datapath for Microscaling ViTs
2505.22194v1
dai2021vs
\cite{dai2021vs}
Vs-quant: Per-vector scaled quantization for accurate low-precision neural network inference
null
null
true
false
Dai, Steve and Venkatesan, Rangha and Ren, Mark and Zimmer, Brian and Dally, William and Khailany, Brucek
2,021
null
null
null
Proceedings of Machine Learning and Systems
Vs-quant: Per-vector scaled quantization for accurate low-precision neural network inference
[PDF] VS-Quant: Per-vector Scaled Quantization for Accurate Low ... - arXiv
https://arxiv.org/pdf/2102.04503
We find that per-vector scaling consistently achieves better inference accuracy at low precision compared to conventional scaling techniques for
Refining Datapath for Microscaling ViTs
2505.22194v1
harma2022accuracy
\cite{harma2022accuracy}
Accuracy Boosters: Epoch-Driven Mixed-Mantissa Block Floating-Point for DNN Training
null
null
true
false
Harma, Simla Burcu and S{\"o}nmez, Canberk and Falsafi, Babak and Jaggi, Martin and Oh, Yunho
2,022
null
null
null
arXiv preprint arXiv:2211.10737
Accuracy Boosters: Epoch-Driven Mixed-Mantissa Block Floating-Point for DNN Training
[PDF] Epoch-Driven Mixed-Mantissa Block Floating Point for DNN Training
https://openreview.net/pdf?id=nfmfqzQ4Mwl
Using analytic models, we show Accuracy Boosters enable increasing arithmetic density for an HBFP training accelerator by up to 21.3× compared to FP32 and up to
Refining Datapath for Microscaling ViTs
2505.22194v1
darvish2020pushing
\cite{darvish2020pushing}
Pushing the limits of narrow precision inferencing at cloud scale with microsoft floating point
null
null
true
false
Darvish Rouhani, Bita and Lo, Daniel and Zhao, Ritchie and Liu, Ming and Fowers, Jeremy and Ovtcharov, Kalin and Vinogradsky, Anna and Massengill, Sarah and Yang, Lita and Bittner, Ray and others
2,020
null
null
null
Advances in neural information processing systems
Pushing the limits of narrow precision inferencing at cloud scale with microsoft floating point
[PDF] Pushing the Limits of Narrow Precision Inferencing at Cloud Scale ...
https://proceedings.neurips.cc/paper/2020/file/747e32ab0fea7fbd2ad9ec03daa3f840-Paper.pdf
In this paper, we explore the limits of Microsoft Floating Point (MSFP), a new class of datatypes developed for production cloud-scale inferencing on custom
Refining Datapath for Microscaling ViTs
2505.22194v1
darvish2023shared
\cite{darvish2023shared}
With Shared Microexponents, A Little Shifting Goes a Long Way
http://arxiv.org/abs/2302.08007v2
This paper introduces Block Data Representations (BDR), a framework for exploring and evaluating a wide spectrum of narrow-precision formats for deep learning. It enables comparison of popular quantization standards, and through BDR, new formats based on shared microexponents (MX) are identified, which outperform other...
true
true
Darvish Rouhani, Bita and Zhao, Ritchie and Elango, Venmugil and Shafipour, Rasoul and Hall, Mathew and Mesmakhosroshahi, Maral and More, Ankit and Melnick, Levi and Golub, Maximilian and Varatkar, Girish and others
2,023
null
null
null
null
With Shared Microexponents, A Little Shifting Goes a Long Way
With Shared Microexponents, A Little Shifting Goes a Long Way
http://arxiv.org/pdf/2302.08007v2
This paper introduces Block Data Representations (BDR), a framework for exploring and evaluating a wide spectrum of narrow-precision formats for deep learning. It enables comparison of popular quantization standards, and through BDR, new formats based on shared microexponents (MX) are identified, which outperform other...
Refining Datapath for Microscaling ViTs
2505.22194v1
andri2022going
\cite{andri2022going}
Going Further With Winograd Convolutions: Tap-Wise Quantization for Efficient Inference on 4x4 Tile
http://arxiv.org/abs/2209.12982v1
Most of today's computer vision pipelines are built around deep neural networks, where convolution operations require most of the generally high compute effort. The Winograd convolution algorithm computes convolutions with fewer MACs compared to the standard algorithm, reducing the operation count by a factor of 2.25x ...
true
true
Andri, Renzo and Bussolino, Beatrice and Cipolletta, Antonio and Cavigelli, Lukas and Wang, Zhe
2,022
null
null
null
null
Going Further With Winograd Convolutions: Tap-Wise Quantization for Efficient Inference on 4x4 Tile
Tap-Wise Quantization for Efficient Inference on 4x4 Tile - arXiv
https://arxiv.org/abs/2209.12982
The Winograd convolution algorithm computes convolutions with fewer MACs compared to the standard algorithm, reducing the operation count by a
Refining Datapath for Microscaling ViTs
2505.22194v1
song2020drq
\cite{song2020drq}
Drq: dynamic region-based quantization for deep neural network acceleration
null
null
true
false
Song, Zhuoran and Fu, Bangqi and Wu, Feiyang and Jiang, Zhaoming and Jiang, Li and Jing, Naifeng and Liang, Xiaoyao
2,020
null
null
null
null
Drq: dynamic region-based quantization for deep neural network acceleration
DRQ Dynamic Region-based Quantization for Deep Neural Network ...
https://github.com/BirenResearch/AIChip_Paper_List/blob/master/notes/ISCA/DRQ%20Dynamic%20Region-based%20Quantization%20for%20Deep%20Neural%20Network%20Acceleration.md
Paper title: DRQ: Dynamic Region-based Quantization for Deep Neural Network Acceleration · Publication: ISCA'20 · Problem to solve: Quantification is an effective
Refining Datapath for Microscaling ViTs
2505.22194v1
zadeh2022mokey
\cite{zadeh2022mokey}
Mokey: Enabling Narrow Fixed-Point Inference for Out-of-the-Box Floating-Point Transformer Models
http://arxiv.org/abs/2203.12758v1
Increasingly larger and better Transformer models keep advancing state-of-the-art accuracy and capability for Natural Language Processing applications. These models demand more computational power, storage, and energy. Mokey reduces the footprint of state-of-the-art 32-bit or 16-bit floating-point transformer models by...
true
true
Zadeh, Ali Hadi and Mahmoud, Mostafa and Abdelhadi, Ameer and Moshovos, Andreas
2,022
null
null
null
null
Mokey: Enabling Narrow Fixed-Point Inference for Out-of-the-Box Floating-Point Transformer Models
Mokey: Enabling Narrow Fixed-Point Inference for Out-of-the-Box ...
https://arxiv.org/abs/2203.12758
Mokey reduces the footprint of state-of-the-art 32-bit or 16-bit floating-point transformer models by quantizing all values to 4-bit indexes into dictionaries.
Refining Datapath for Microscaling ViTs
2505.22194v1
zhao2021cambricon
\cite{zhao2021cambricon}
Cambricon-Q: A hybrid architecture for efficient training
null
null
true
false
Zhao, Yongwei and Liu, Chang and Du, Zidong and Guo, Qi and Hu, Xing and Zhuang, Yimin and Zhang, Zhenxing and Song, Xinkai and Li, Wei and Zhang, Xishan and others
2,021
null
null
null
null
Cambricon-Q: A hybrid architecture for efficient training
Cambricon-Q: A Hybrid Architecture for Efficient Training
https://www.computer.org/csdl/proceedings-article/isca/2021/333300a706/1vNjDWVoisw
Cambricon-Q features a hybrid architecture consisting of an ASIC acceleration core and a near-data-processing (NDP) engine. The acceleration core mainly targets
Refining Datapath for Microscaling ViTs
2505.22194v1
wang2019haq
\cite{wang2019haq}
HAQ: Hardware-Aware Automated Quantization with Mixed Precision
http://arxiv.org/abs/1811.08886v3
Model quantization is a widely used technique to compress and accelerate deep neural network (DNN) inference. Emergent DNN hardware accelerators begin to support mixed precision (1-8 bits) to further improve the computation efficiency, which raises a great challenge to find the optimal bitwidth for each layer: it requi...
true
true
Wang, Kuan and Liu, Zhijian and Lin, Yujun and Lin, Ji and Han, Song
2,019
null
null
null
null
HAQ: Hardware-Aware Automated Quantization with Mixed Precision
HAQ: Hardware-Aware Automated Quantization with Mixed Precision
https://arxiv.org/abs/1811.08886
In this paper, we introduce the Hardware-Aware Automated Quantization (HAQ) framework which leverages the reinforcement learning to automatically determine the
Refining Datapath for Microscaling ViTs
2505.22194v1
dettmers2022llm
\cite{dettmers2022llm}
Llm. int8 (): 8-bit matrix multiplication for transformers at scale
null
null
true
false
Dettmers, Tim and Lewis, Mike and Belkada, Younes and Zettlemoyer, Luke
2,022
null
null
null
arXiv preprint arXiv:2208.07339
Llm. int8 (): 8-bit matrix multiplication for transformers at scale
LLM.int8(): 8-bit matrix multiplication for transformers at scale
https://dl.acm.org/doi/10.5555/3600270.3602468
We develop a procedure for Int8 matrix multiplication for feed-forward and attention projection layers in transformers, which cut the memory needed for
Refining Datapath for Microscaling ViTs
2505.22194v1
frantar2022gptq
\cite{frantar2022gptq}
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
http://arxiv.org/abs/2210.17323v2
Generative Pre-trained Transformer models, known as GPT or OPT, set themselves apart through breakthrough performance across complex language modelling tasks, but also by their extremely high computational and storage costs. Specifically, due to their massive size, even inference for large, highly-accurate GPT models m...
true
true
Frantar, Elias and Ashkboos, Saleh and Hoefler, Torsten and Alistarh, Dan
2,022
null
null
null
arXiv preprint arXiv:2210.17323
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
[PDF] gptq: accurate post-training quantization - arXiv
https://arxiv.org/pdf/2210.17323
Generative Pre-trained Transformer models, known as GPT or OPT, set them- selves apart through breakthrough performance across complex
Refining Datapath for Microscaling ViTs
2505.22194v1
dong2019hawq
\cite{dong2019hawq}
HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precision
http://arxiv.org/abs/1905.03696v1
Model size and inference speed/power have become a major challenge in the deployment of Neural Networks for many applications. A promising approach to address these problems is quantization. However, uniformly quantizing a model to ultra low precision leads to significant accuracy degradation. A novel solution for this...
true
true
Dong, Zhen and Yao, Zhewei and Gholami, Amir and Mahoney, Michael W and Keutzer, Kurt
2,019
null
null
null
null
HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precision
[PDF] Hessian AWare Quantization of Neural Networks With Mixed-Precision
https://www.stat.berkeley.edu/~mmahoney/pubs/HAWQ_ICCV_2019_paper.pdf
HAWQ allows for the automatic se- lection of the relative quantization precision of each layer, based on the layer's Hessian spectrum. Moreover, HAWQ provides a
Refining Datapath for Microscaling ViTs
2505.22194v1
xiao2022smoothquant
\cite{xiao2022smoothquant}
SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
http://arxiv.org/abs/2211.10438v7
Large language models (LLMs) show excellent performance but are compute- and memory-intensive. Quantization can reduce memory and accelerate inference. However, existing methods cannot maintain accuracy and hardware efficiency at the same time. We propose SmoothQuant, a training-free, accuracy-preserving, and general-p...
true
true
Xiao, Guangxuan and Lin, Ji and Seznec, Mickael and Demouth, Julien and Han, Song
2,022
null
null
null
arXiv preprint arXiv:2211.10438
SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
SmoothQuant: Accurate and Efficient Post-Training Quantization for ...
https://arxiv.org/abs/2211.10438
Image 2: arxiv logo>cs> arXiv:2211.10438 **arXiv:2211.10438** (cs) View a PDF of the paper titled SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models, by Guangxuan Xiao and 5 other authors View a PDF of the paper titled SmoothQuant: Accurate and Efficient Post-Training Quantizatio...
Refining Datapath for Microscaling ViTs
2505.22194v1
yao2022zeroquant
\cite{yao2022zeroquant}
ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers
http://arxiv.org/abs/2206.01861v1
How to efficiently serve ever-larger trained natural language models in practice has become exceptionally challenging even for powerful cloud servers due to their prohibitive memory/computation requirements. In this work, we present an efficient and affordable post-training quantization approach to compress large Trans...
true
true
Yao, Zhewei and Yazdani Aminabadi, Reza and Zhang, Minjia and Wu, Xiaoxia and Li, Conglong and He, Yuxiong
2,022
null
null
null
Advances in Neural Information Processing Systems
ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers
ZeroQuant: Efficient and Affordable Post-Training Quantization for ...
https://arxiv.org/abs/2206.01861
In this work, we present an efficient and affordable post-training quantization approach to compress large Transformer-based models, termed as ZeroQuant.
Refining Datapath for Microscaling ViTs
2505.22194v1
liu2023psq
\cite{liu2023psq}
PSQ: An Automatic Search Framework for Data-Free Quantization on PIM-based Architecture
null
null
true
false
Liu, Fangxin and Yang, Ning and Jiang, Li
2,023
null
null
null
null
PSQ: An Automatic Search Framework for Data-Free Quantization on PIM-based Architecture
PSQ: An Automatic Search Framework for Data-Free Quantization ...
https://ieeexplore.ieee.org/servlet/Login?logout=/document/10361000
The scheme tightly combines the search principle of quantization and the PIM architecture to provide smooth hardware-friendly quantization. We leverage the
Refining Datapath for Microscaling ViTs
2505.22194v1
liu2024spark
\cite{liu2024spark}
SPARK: Scalable and Precision-Aware Acceleration of Neural Networks via Efficient Encoding
null
null
true
false
Liu, Fangxin and Yang, Ning and Li, Haomin and Wang, Zongwu and Song, Zhuoran and Pei, Songwen and Jiang, Li
2,024
null
null
null
null
SPARK: Scalable and Precision-Aware Acceleration of Neural Networks via Efficient Encoding
Scalable and Precision-Aware Acceleration of Neural ...
https://ieeexplore.ieee.org/document/10476472/
by F Liu · 2024 · Cited by 31 — SPARK: Scalable and Precision-Aware Acceleration of Neural Networks via Efficient Encoding ; Article #: ; Date of Conference: 02-06 March 2024 ; Date Added to IEEE
Refining Datapath for Microscaling ViTs
2505.22194v1
chang2021mix
\cite{chang2021mix}
Mix and Match: A Novel FPGA-Centric Deep Neural Network Quantization Framework
http://arxiv.org/abs/2012.04240v2
Deep Neural Networks (DNNs) have achieved extraordinary performance in various application domains. To support diverse DNN models, efficient implementations of DNN inference on edge-computing platforms, e.g., ASICs, FPGAs, and embedded systems, are extensively investigated. Due to the huge model size and computation am...
true
true
Chang, Sung-En and Li, Yanyu and Sun, Mengshu and Shi, Runbin and So, Hayden K-H and Qian, Xuehai and Wang, Yanzhi and Lin, Xue
2,021
null
null
null
null
Mix and Match: A Novel FPGA-Centric Deep Neural Network Quantization Framework
[PDF] A Novel FPGA-Centric Deep Neural Network Quantization Framework
https://par.nsf.gov/servlets/purl/10232486
This paper proposes a DNN quantization framework that applies different quantization schemes for different weight matrix rows, using a hardware-friendly SP2
Refining Datapath for Microscaling ViTs
2505.22194v1
wu2023msd
\cite{wu2023msd}
MSD: Mixing Signed Digit Representations for Hardware-efficient DNN Acceleration on FPGA with Heterogeneous Resources
null
null
true
false
Wu, Jiajun and Zhou, Jiajun and Gao, Yizhao and Ding, Yuhao and Wong, Ngai and So, Hayden Kwok-Hay
2,023
null
null
null
null
MSD: Mixing Signed Digit Representations for Hardware-efficient DNN Acceleration on FPGA with Heterogeneous Resources
MSD: Mixing Signed Digit Representations for Hardware-efficient ...
https://www.researchgate.net/publication/372264814_MSD_Mixing_Signed_Digit_Representations_for_Hardware-efficient_DNN_Acceleration_on_FPGA_with_Heterogeneous_Resources
MSD: Mixing Signed Digit Representations for Hardware-efficient DNN Acceleration on FPGA with Heterogeneous Resources ... effectively improve training efficiency
Refining Datapath for Microscaling ViTs
2505.22194v1
sharma2018bit
\cite{sharma2018bit}
Bit Fusion: Bit-Level Dynamically Composable Architecture for Accelerating Deep Neural Networks
http://arxiv.org/abs/1712.01507v2
Fully realizing the potential of acceleration for Deep Neural Networks (DNNs) requires understanding and leveraging algorithmic properties. This paper builds upon the algorithmic insight that bitwidth of operations in DNNs can be reduced without compromising their classification accuracy. However, to prevent accuracy l...
true
true
Sharma, Hardik and Park, Jongse and Suda, Naveen and Lai, Liangzhen and Chau, Benson and Kim, Joon Kyung and Chandra, Vikas and Esmaeilzadeh, Hadi
2,018
null
null
null
null
Bit Fusion: Bit-Level Dynamically Composable Architecture for Accelerating Deep Neural Networks
[PDF] Bit Fusion: Bit-Level Dynamically Composable Architecture for ...
https://bpb-us-w2.wpmucdn.com/sites.coecis.cornell.edu/dist/7/587/files/2023/06/Sharma_2018_Bit.pdf
RETROSPECTIVE: Bit Fusion: Bit-Level. Dynamically Composable Architecture for. Accelerating Deep Neural Networks. Hardik Sharma1. Jongse Park2. Naveen Suda3.
Refining Datapath for Microscaling ViTs
2505.22194v1
fan2022adaptable
\cite{fan2022adaptable}
Adaptable Butterfly Accelerator for Attention-based NNs via Hardware and Algorithm Co-design
http://arxiv.org/abs/2209.09570v1
Attention-based neural networks have become pervasive in many AI tasks. Despite their excellent algorithmic performance, the use of the attention mechanism and feed-forward network (FFN) demands excessive computational and memory resources, which often compromises their hardware performance. Although various sparse var...
true
true
Fan, Hongxiang and Chau, Thomas and Venieris, Stylianos I and Lee, Royson and Kouris, Alexandros and Luk, Wayne and Lane, Nicholas D and Abdelfattah, Mohamed S
2,022
null
null
null
null
Adaptable Butterfly Accelerator for Attention-based NNs via Hardware and Algorithm Co-design
Adaptable Butterfly Accelerator for Attention-based NNs via ...
https://ieeexplore.ieee.org/iel7/9923754/9923780/09923888.pdf
by H Fan · 2022 · Cited by 74 — In this work, we address this challenge by adopting an algorithm and hardware co-design approach. On the algorithmic level, a hardware-friendly model called.
Refining Datapath for Microscaling ViTs
2505.22194v1
ham20203
\cite{ham20203}
A$^3$: Accelerating Attention Mechanisms in Neural Networks with Approximation
http://arxiv.org/abs/2002.10941v1
With the increasing computational demands of neural networks, many hardware accelerators for the neural networks have been proposed. Such existing neural network accelerators often focus on popular neural network types such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs); however, not much ...
true
true
Ham, Tae Jun and Jung, Sung Jun and Kim, Seonghak and Oh, Young H and Park, Yeonhong and Song, Yoonho and Park, Jung-Hun and Lee, Sanghee and Park, Kyoung and Lee, Jae W and others
2,020
null
null
null
null
A$^3$: Accelerating Attention Mechanisms in Neural Networks with Approximation
A$^3$: Accelerating Attention Mechanisms in Neural Networks with ...
https://arxiv.org/abs/2002.10941
Based on this observation, we design and architect A3, which accelerates attention mechanisms in neural networks with algorithmic approximation
Refining Datapath for Microscaling ViTs
2505.22194v1
ham2021elsa
\cite{ham2021elsa}
ELSA: Hardware-software co-design for efficient, lightweight self-attention mechanism in neural networks
null
null
true
false
Ham, Tae Jun and Lee, Yejin and Seo, Seong Hoon and Kim, Soosung and Choi, Hyunji and Jung, Sung Jun and Lee, Jae W
2,021
null
null
null
null
ELSA: Hardware-software co-design for efficient, lightweight self-attention mechanism in neural networks
ELSA: Hardware-software Co-design for efficient, lightweight ...
https://s-space.snu.ac.kr/handle/10371/183738
ELSA: Hardware-software Co-design for efficient, lightweight self-attention mechanism in neural networks. Cited 111 time in Web of Science Cited 133 time in
Refining Datapath for Microscaling ViTs
2505.22194v1
hong2022dfx
\cite{hong2022dfx}
DFX: A Low-latency Multi-FPGA Appliance for Accelerating Transformer-based Text Generation
http://arxiv.org/abs/2209.10797v1
Transformer is a deep learning language model widely used for natural language processing (NLP) services in datacenters. Among transformer models, Generative Pre-trained Transformer (GPT) has achieved remarkable performance in text generation, or natural language generation (NLG), which needs the processing of a large ...
true
true
Hong, Seongmin and Moon, Seungjae and Kim, Junsoo and Lee, Sungjae and Kim, Minsub and Lee, Dongsoo and Kim, Joo-Young
2,022
null
null
null
null
DFX: A Low-latency Multi-FPGA Appliance for Accelerating Transformer-based Text Generation
DFX: A Low-latency Multi-FPGA Appliance for Accelerating ...
https://ieeexplore.ieee.org/document/9895626
by S Hong · 2022 · Cited by 104 — DFX is a multi-FPGA appliance that accelerates transformer-based text generation. DFX adopts model parallelism to efficiently process the large-scale language
Refining Datapath for Microscaling ViTs
2505.22194v1
kao2023flat
\cite{kao2023flat}
FLAT: An Optimized Dataflow for Mitigating Attention Bottlenecks
http://arxiv.org/abs/2107.06419v7
Attention mechanisms, primarily designed to capture pairwise correlations between words, have become the backbone of machine learning, expanding beyond natural language processing into other domains. This growth in adaptation comes at the cost of prohibitively large memory requirements and computational complexity, esp...
true
true
Kao, Sheng-Chun and Subramanian, Suvinay and Agrawal, Gaurav and Yazdanbakhsh, Amir and Krishna, Tushar
2,023
null
null
null
null
FLAT: An Optimized Dataflow for Mitigating Attention Bottlenecks
FLAT: An Optimized Dataflow for Mitigating Attention Bottlenecks
http://arxiv.org/pdf/2107.06419v7
Attention mechanisms, primarily designed to capture pairwise correlations between words, have become the backbone of machine learning, expanding beyond natural language processing into other domains. This growth in adaptation comes at the cost of prohibitively large memory requirements and computational complexity, esp...
Refining Datapath for Microscaling ViTs
2505.22194v1
li2020ftrans
\cite{li2020ftrans}
FTRANS: Energy-Efficient Acceleration of Transformers using FPGA
http://arxiv.org/abs/2007.08563v1
In natural language processing (NLP), the "Transformer" architecture was proposed as the first transduction model replying entirely on self-attention mechanisms without using sequence-aligned recurrent neural networks (RNNs) or convolution, and it achieved significant improvements for sequence to sequence tasks. The in...
true
true
Li, Bingbing and Pandey, Santosh and Fang, Haowen and Lyv, Yanjun and Li, Ji and Chen, Jieyang and Xie, Mimi and Wan, Lipeng and Liu, Hang and Ding, Caiwen
2,020
null
null
null
null
FTRANS: Energy-Efficient Acceleration of Transformers using FPGA
[PDF] FTRANS: Energy-Efficient Acceleration of Transformers using FPGA
https://scispace.com/pdf/ftrans-energy-efficient-acceleration-of-transformers-using-4ipjn26xe9.pdf
In this paper, we propose an energy-efficient acceleration frame- work, Ftrans, for transformer-based large scale language repre- sentations using FPGA. Ftrans
Refining Datapath for Microscaling ViTs
2505.22194v1
lu2021sanger
\cite{lu2021sanger}
Sanger: A co-design framework for enabling sparse attention using reconfigurable architecture
null
null
true
false
Lu, Liqiang and Jin, Yicheng and Bi, Hangrui and Luo, Zizhang and Li, Peng and Wang, Tao and Liang, Yun
2,021
null
null
null
null
Sanger: A co-design framework for enabling sparse attention using reconfigurable architecture
hatsu3/Sanger - GitHub
https://github.com/hatsu3/Sanger
This repository implements the proposed framework in the paper Sanger: A Co-Design Framework for Enabling Sparse Attention using Reconfigurable Architecture (
Refining Datapath for Microscaling ViTs
2505.22194v1
zadeh2020gobo
\cite{zadeh2020gobo}
GOBO: Quantizing Attention-Based NLP Models for Low Latency and Energy Efficient Inference
http://arxiv.org/abs/2005.03842v2
Attention-based models have demonstrated remarkable success in various natural language understanding tasks. However, efficient execution remains a challenge for these models which are memory-bound due to their massive number of parameters. We present GOBO, a model quantization technique that compresses the vast majori...
true
true
Zadeh, Ali Hadi and Edo, Isak and Awad, Omar Mohamed and Moshovos, Andreas
2,020
null
null
null
null
GOBO: Quantizing Attention-Based NLP Models for Low Latency and Energy Efficient Inference
[PDF] GOBO: Quantizing Attention-Based NLP Models for Low Latency ...
https://microarch.org/micro53/papers/738300a811.pdf
GOBO is a model quantization technique that compresses 99.9% of NLP model parameters to 3 bits, maintaining accuracy without fine-tuning.
Refining Datapath for Microscaling ViTs
2505.22194v1
tambe2021edgebert
\cite{tambe2021edgebert}
EdgeBERT: Sentence-Level Energy Optimizations for Latency-Aware Multi-Task NLP Inference
http://arxiv.org/abs/2011.14203v5
Transformer-based language models such as BERT provide significant accuracy improvement for a multitude of natural language processing (NLP) tasks. However, their hefty computational and memory demands make them challenging to deploy to resource-constrained edge platforms with strict latency requirements. We present Ed...
true
true
Tambe, Thierry and Hooper, Coleman and Pentecost, Lillian and Jia, Tianyu and Yang, En-Yu and Donato, Marco and Sanh, Victor and Whatmough, Paul and Rush, Alexander M. and Brooks, David and Wei, Gu-Yeon
2,021
null
https://doi.org/10.1145/3466752.3480095
10.1145/3466752.3480095
null
EdgeBERT: Sentence-Level Energy Optimizations for Latency-Aware Multi-Task NLP Inference
Sentence-Level Energy Optimizations for Latency-Aware ...
https://dl.acm.org/doi/10.1145/3466752.3480095
We present EdgeBERT, an in-depth algorithm-hardware co-design for latency-aware energy optimizations for multi-task NLP.
Refining Datapath for Microscaling ViTs
2505.22194v1
qin2023fact
\cite{qin2023fact}
FACT: FFN-attention Co-optimized transformer architecture with eager correlation prediction
null
null
true
false
Qin, Yubin and Wang, Yang and Deng, Dazheng and Zhao, Zhiren and Yang, Xiaolong and Liu, Leibo and Wei, Shaojun and Hu, Yang and Yin, Shouyi
2,023
null
null
null
null
FACT: FFN-attention Co-optimized transformer architecture with eager correlation prediction
FACT: FFN-Attention Co-optimized Transformer Architecture with ...
https://dl.acm.org/doi/pdf/10.1145/3579371.3589057
We first propose an eager prediction algorithm which predicts the attention matrix before QKV generation. It fur- ther detects the unnecessary
Refining Datapath for Microscaling ViTs
2505.22194v1
zeng2024flightllm
\cite{zeng2024flightllm}
FlightLLM: Efficient Large Language Model Inference with a Complete Mapping Flow on FPGAs
http://arxiv.org/abs/2401.03868v2
Transformer-based Large Language Models (LLMs) have made a significant impact on various domains. However, LLMs' efficiency suffers from both heavy computation and memory overheads. Compression techniques like sparsification and quantization are commonly used to mitigate the gap between LLM's computation/memory overhea...
true
true
Zeng, Shulin and Liu, Jun and Dai, Guohao and Yang, Xinhao and Fu, Tianyu and Wang, Hongyi and Ma, Wenheng and Sun, Hanbo and Li, Shiyao and Huang, Zixiao and others
2,024
null
null
null
arXiv preprint arXiv:2401.03868
FlightLLM: Efficient Large Language Model Inference with a Complete Mapping Flow on FPGAs
FlightLLM: Efficient Large Language Model Inference with a ...
https://dl.acm.org/doi/10.1145/3626202.3637562
This paper proposes FlightLLM, enabling efficient LLMs inference with a complete mapping flow on FPGAs.
Refining Datapath for Microscaling ViTs
2505.22194v1
li2022auto
\cite{li2022auto}
Auto-vit-acc: An fpga-aware automatic acceleration framework for vision transformer with mixed-scheme quantization
null
null
true
false
Li, Zhengang and Sun, Mengshu and Lu, Alec and Ma, Haoyu and Yuan, Geng and Xie, Yanyue and Tang, Hao and Li, Yanyu and Leeser, Miriam and Wang, Zhangyang and others
2,022
null
null
null
null
Auto-vit-acc: An fpga-aware automatic acceleration framework for vision transformer with mixed-scheme quantization
[PDF] Auto-ViT-Acc: An FPGA-Aware Automatic Acceleration Framework ...
https://www.sfu.ca/~zhenman/files/C25-FPL2022-Auto-ViT-Acc.pdf
Missing: 04/08/2025
Refining Datapath for Microscaling ViTs
2505.22194v1
dong2023heatvit
\cite{dong2023heatvit}
HeatViT: Hardware-Efficient Adaptive Token Pruning for Vision Transformers
http://arxiv.org/abs/2211.08110v2
While vision transformers (ViTs) have continuously achieved new milestones in the field of computer vision, their sophisticated network architectures with high computation and memory costs have impeded their deployment on resource-limited edge devices. In this paper, we propose a hardware-efficient image-adaptive token...
true
true
Dong, Peiyan and Sun, Mengshu and Lu, Alec and Xie, Yanyue and Liu, Kenneth and Kong, Zhenglun and Meng, Xin and Li, Zhengang and Lin, Xue and Fang, Zhenman and others
2,023
null
null
null
null
HeatViT: Hardware-Efficient Adaptive Token Pruning for Vision Transformers
Hardware-Efficient Adaptive Token Pruning for Vision Transformers
https://ieeexplore.ieee.org/iel7/10070856/10070923/10071047.pdf
HeatViT is a hardware-efficient token pruning framework for ViTs, using a token selector to reduce token number and package non-informative tokens.
Refining Datapath for Microscaling ViTs
2505.22194v1
huang2023integer
\cite{huang2023integer}
An Integer-Only and Group-Vector Systolic Accelerator for Efficiently Mapping Vision Transformer on Edge
null
null
true
false
Huang, Mingqiang and Luo, Junyi and Ding, Chenchen and Wei, Zikun and Huang, Sixiao and Yu, Hao
2,023
null
null
null
IEEE Transactions on Circuits and Systems I: Regular Papers
An Integer-Only and Group-Vector Systolic Accelerator for Efficiently Mapping Vision Transformer on Edge
An Integer-Only and Group-Vector Systolic Accelerator for ...
https://colab.ws/articles/10.1109%2Ftcsi.2023.3312775
Therefore, in this work, we propose the ViA, a novel vision transformer (ViT) accelerator architecture based on FPGA, to execute the transformer
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
DBLP:conf/rtss/BrandenburgG16
\cite{DBLP:conf/rtss/BrandenburgG16}
Global Scheduling Not Required: Simple, Near-Optimal Multiprocessor Real-Time Scheduling with Semi-Partitioned Reservations
null
null
true
false
Bj{\"{o}}rn B. Brandenburg and Mahircan Gul
2,016
null
null
null
null
Global Scheduling Not Required: Simple, Near-Optimal Multiprocessor Real-Time Scheduling with Semi-Partitioned Reservations
[PDF] Simple, Near-Optimal Multiprocessor Real-Time Scheduling with ...
https://people.mpi-sws.org/~bbb/papers/pdf/rtss16b.pdf
Missing: 04/08/2025
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
ekberg2021partitioned
\cite{ekberg2021partitioned}
Partitioned Scheduling of Recurrent Real-Time Tasks
null
null
true
false
Ekberg, Pontus and Baruah, Sanjoy
2,021
null
null
null
null
Partitioned Scheduling of Recurrent Real-Time Tasks
Partitioned Scheduling of Recurrent Real-Time Tasks
https://user.it.uu.se/~ponek616/files/RTSS21/RTSS21.pdf
by P Ekberg · Cited by 11 — Under the partitioned paradigm of multiprocessor scheduling for recurrent tasks, each task is pre-assigned to an individual processor and all jobs generated by
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Burchard:1995
\cite{Burchard:1995}
New strategies for assigning real-time tasks to multiprocessor systems
null
null
true
false
Burchard, Almut and Liebeherr, Jörg and Oh, Yingfeng and Son, Sang H.
1,995
null
null
null
IEEE Transactions on Computers
New strategies for assigning real-time tasks to multiprocessor systems
New Strategies for Assigning Real-Time Tasks to ...
https://www.computer.org/csdl/journal/tc/1995/12/t1429/13rRUwd9CF8
by J Liebeherr · 1995 · Cited by 378 — There are two strategies for scheduling real-time tasks on a multiprocessor system. In a global scheme each occurrence of a real-time task may be executed on a
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Dhall:1978
\cite{Dhall:1978}
On a real-time scheduling problem
null
null
true
false
Dhall, Sudarshan K and Liu, Chung Laung
1,978
null
null
null
Operations research
On a real-time scheduling problem
On a Real-Time Scheduling Problem | Operations Research
https://pubsonline.informs.org/doi/10.1287/opre.26.1.127
The scheduling problem is to specify an order in which the requests of a set of tasks are to be executed and the processor to be used, with the goal of meeting
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Baruah:2005
\cite{Baruah:2005}
The partitioned multiprocessor scheduling of sporadic task systems
null
null
true
false
Baruah, Sanjoy and Fisher, Nathan
2,005
null
null
null
null
The partitioned multiprocessor scheduling of sporadic task systems
The partitioned multiprocessor scheduling of sporadic task ...
https://ieeexplore.ieee.org/document/1563119/
por S Baruah · 2005 · Mencionado por 204 — A polynomial-time algorithm is presented for partitioning a collection of sporadic tasks among the processors of an identical multiprocessor platform.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Lopez:2000
\cite{Lopez:2000}
Worst-case utilization bound for EDF scheduling on real-time multiprocessor systems
null
null
true
false
José María López and Garcia, Manuel and Díaz, Jose and Garcia, Frk Daniel
2,000
null
null
null
null
Worst-case utilization bound for EDF scheduling on real-time multiprocessor systems
EDF Scheduling on Heterogeneous Multiprocessors
https://research.engineering.wustl.edu/~baruah/DISSERTATIONS/01funk.pdf
by SH Funk · 2004 · Cited by 57 — Worst- case utilization bound for EDF scheduling on real-time multiprocessor systems. In Proceedings of the EuroMicro Conference on Real-Time Systems, pages
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Fisher:2006
\cite{Fisher:2006}
The partitioned multiprocessor scheduling of non-preemptive sporadic task systems
null
null
true
false
Fisher, Nathan and Baruah, Sanjoy
2,006
null
null
null
null
The partitioned multiprocessor scheduling of non-preemptive sporadic task systems
The Partitioned Multiprocessor Scheduling of Sporadic Task Systems
https://fishern.eng.wayne.edu/papers/2005-baruahFisher-RTSS.pdf
On multiprocessor systems, two alternative paradigms for scheduling collections of sporadic tasks have been considered: partitioned and global scheduling. In the partitioned approach, the tasks are statically partitioned among the processors, i.e., each task is assigned to a processor and is always executed on it .
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Senoussaoui:2020
\cite{Senoussaoui:2020}
Allocation of Real-Time Tasks onto Identical Core Platforms under Deferred fixed Preemption-Point Model
null
null
true
false
Senoussaoui, Ikram and Zahaf, Houssam-Eddine and Benhaoua, Mohammed Kamel and Lipari, Giuseppe and Olejnik, Richard
2,020
null
null
10.1145/3394810.3394821
null
Allocation of Real-Time Tasks onto Identical Core Platforms under Deferred fixed Preemption-Point Model
[PDF] Allocation of Real-Time Tasks onto Identical Core Platforms ... - HAL
https://hal.science/hal-02886816/document
Missing: 04/08/2025
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
fonseca2016response
\cite{fonseca2016response}
Response time analysis of sporadic DAG tasks under partitioned scheduling
null
null
true
false
Fonseca, Jos{\'e} and Nelissen, Geoffrey and Nelis, Vincent and Pinho, Lu{\'\i}s Miguel
2,016
null
null
null
null
Response time analysis of sporadic DAG tasks under partitioned scheduling
Response time analysis of sporadic DAG tasks under partitioned ...
https://ieeexplore.ieee.org/document/7509443
More precisely, we present a response time analysis for sporadic DAG tasks atop multiprocessors under partitioned fixed-priority scheduling. We assume the
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
casini2018partitioned
\cite{casini2018partitioned}
Partitioned fixed-priority scheduling of parallel tasks without preemptions
null
null
true
false
Casini, Daniel and Biondi, Alessandro and Nelissen, Geoffrey and Buttazzo, Giorgio
2,018
null
null
null
null
Partitioned fixed-priority scheduling of parallel tasks without preemptions
Partitioned Fixed-Priority Scheduling of Parallel Tasks ...
http://ieeexplore.ieee.org/document/8603232/
Abstract: The study of parallel task models executed with predictable scheduling approaches is a fundamental problem for real-time multiprocessor systems.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Zahaf:2020
\cite{Zahaf:2020}
Preemption-Aware Allocation, Deadline Assignment for Conditional DAGs on Partitioned EDF
null
null
true
false
Zahaf, Houssam-Eddine and Lipari, Giuseppe and Niar, Smail and Hassan Benyamina, Abou El
2,020
null
null
10.1109/RTCSA50079.2020.9203643
null
Preemption-Aware Allocation, Deadline Assignment for Conditional DAGs on Partitioned EDF
Preemption-Aware Allocation, Deadline Assignment for Conditional ...
https://www.computer.org/csdl/proceedings-article/rtcsa/2020/09203643/1nkD7ZL5ycE
Preemption-Aware Allocation, Deadline Assignment for Conditional DAGs on Partitioned EDF. 2020, pp. 1-10,. DOI Bookmark: 10.1109/RTCSA50079.2020.9203643.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Ueter:2021
\cite{Ueter:2021}
{Hard Real-Time Stationary GANG-Scheduling}
null
null
true
false
Ueter, Niklas and G\"{u}nzel, Mario and von der Br\"{u}ggen, Georg and Chen, Jian-Jia
2,021
null
null
10.4230/LIPIcs.ECRTS.2021.10
null
{Hard Real-Time Stationary GANG-Scheduling}
[PDF] Hard Real-Time Stationary GANG-Scheduling - DROPS
https://drops.dagstuhl.de/storage/00lipics/lipics-vol196-ecrts2021/LIPIcs.ECRTS.2021.10/LIPIcs.ECRTS.2021.10.pdf
Contributions: In this paper we explore stationary gang scheduling for a set of sporadic real- time tasks with constrained deadlines (i.e., the relative
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
sun2024strict
\cite{sun2024strict}
Strict Partitioning for Sporadic Rigid Gang Tasks
http://arxiv.org/abs/2403.10726v2
The rigid gang task model is based on the idea of executing multiple threads simultaneously on a fixed number of processors to increase efficiency and performance. Although there is extensive literature on global rigid gang scheduling, partitioned approaches have several practical advantages (e.g., task isolation and r...
true
true
Sun, Binqi and Kloda, Tomasz and Caccamo, Marco
2,024
null
null
null
null
Strict Partitioning for Sporadic Rigid Gang Tasks
Strict Partitioning for Sporadic Rigid Gang Tasks
http://arxiv.org/pdf/2403.10726v2
The rigid gang task model is based on the idea of executing multiple threads simultaneously on a fixed number of processors to increase efficiency and performance. Although there is extensive literature on global rigid gang scheduling, partitioned approaches have several practical advantages (e.g., task isolation and r...
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
sun2024partitioned
\cite{sun2024partitioned}
Partitioned scheduling and parallelism assignment for real-time DNN inference tasks on multi-TPU
null
null
true
false
Sun, Binqi and Kloda, Tomasz and Wu, Chu-ge and Caccamo, Marco
2,024
null
null
null
null
Partitioned scheduling and parallelism assignment for real-time DNN inference tasks on multi-TPU
[PDF] Partitioned Scheduling and Parallelism Assignment for Real-Time ...
https://laas.hal.science/hal-04803800/document
We propose an NPG strict partitioning strategy for scheduling. DNN tasks on multi-TPU and a strict partitioning heuristic to determine the
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Zahaf:2021
\cite{Zahaf:2021}
Contention-Aware GPU Partitioning and Task-to-Partition Allocation for Real-Time Workloads
http://arxiv.org/abs/2105.10312v1
In order to satisfy timing constraints, modern real-time applications require massively parallel accelerators such as General Purpose Graphic Processing Units (GPGPUs). Generation after generation, the number of computing clusters made available in novel GPU architectures is steadily increasing, hence, investigating su...
true
true
Zahaf, Houssam-Eddine and Olmedo, Ignacio Sanudo and Singh, Jayati and Capodieci, Nicola and Faucou, Sebastien
2,021
null
null
10.1145/3453417.3453439
null
Contention-Aware GPU Partitioning and Task-to-Partition Allocation for Real-Time Workloads
Contention-Aware GPU Partitioning and Task-to- ...
http://pagesperso.ls2n.fr/~zahaf-h/research/2021/rtns_2021.pdf
by HE Zahaf · Cited by 12 — Contention-Aware GPU Partitioning and Task-to-Partition Allocation for. Real-Time Workloads. Houssam-Eddine Zahaf, Ignacio Sañudo Olmedo, Jayati Singh, Nicola.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
abeni2022partitioning
\cite{abeni2022partitioning}
Partitioning real-time workloads on multi-core virtual machines
null
null
true
false
Abeni, Luca and Biondi, Alessandro and Bini, Enrico
2,022
null
null
null
Journal of Systems Architecture
Partitioning real-time workloads on multi-core virtual machines
[PDF] Partitioning Real-Time Workloads on Multi-Core Virtual Machines
https://iris.unito.it/retrieve/fb327f6c-20e4-4a18-b202-acf4643df773/main.pdf
duling), this paper proposes and compares some approaches for partitioning the real-time workloads in multi-core VMs. Some of the proposed
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Mo:2023
\cite{Mo:2023}
Energy Optimized Task Mapping for Reliable and Real-Time Networked Systems
null
null
true
false
Mo, Lei and Zhou, Qi and Kritikakou, Angeliki and Cao, Xianghui
2,023
null
null
10.1145/3584985
ACM Trans. Sen. Netw.
Energy Optimized Task Mapping for Reliable and Real-Time Networked Systems
Energy Optimized Task Mapping for Reliable and Real-Time ...
https://dl.acm.org/doi/10.1145/3584985
Energy efficiency, real-time response, and data transmission reliability are important objectives during networked systems design.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
MDBCCP:13
\cite{MDBCCP:13}
{Real-time cache management framework for multi-core architectures}
null
null
true
false
Mancuso, Renato and Dudko, Roman and Betti, Emiliano and Cesati, Marco and Caccamo, Marco and Pellizzoni, Rodolfo
2,013
null
null
null
null
{Real-time cache management framework for multi-core architectures}
[PDF] Cache Management and Time-triggered Scheduling for Hard Real ...
https://mediatum.ub.tum.de/attfile/1200769/hd2/incoming/2014-Apr/394974.pdf
Real-time cache management framework for multi-core archi- tectures. In 2013 IEEE 19th Real-Time and Embedded Technology and. Applications Symposium (RTAS),
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Kim16:EMSOFT
\cite{Kim16:EMSOFT}
Real-time cache management for multi-core virtualization
null
null
true
false
Kim, Hyoseung and Rajkumar, Ragunathan
2,016
null
null
null
null
Real-time cache management for multi-core virtualization
Real-time cache management for multi-core virtualization
https://ieeexplore.ieee.org/document/7743233/
In this paper, we propose a real-time cache management framework for multi-core virtualization. Our framework introduces two hypervisor-level techniques, vLLC
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
KWCFAS:17
\cite{KWCFAS:17}
Attacking the One-Out-Of-m Multicore Problem by Combining Hardware Management with Mixed-Criticality Provisioning
null
null
true
false
Kim, Namhoon and Ward, Bryan C. and Chisholm, Micaiah and Fu, Cheng-Yang and Anderson, James H. and Smith, F. Donelson
2,017
null
null
null
Real-Time Systems
Attacking the One-Out-Of-m Multicore Problem by Combining Hardware Management with Mixed-Criticality Provisioning
IEEE Real-Time and Embedded Technology and Applications ...
http://www.findresearch.org/conferences/conf/rtas/2016/conference.html
Attacking the One-Out-Of-m Multicore Problem by Combining Hardware Management with Mixed-Criticality Provisioning · Details. Discussion Comments: 0.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
KSMCV:19
\cite{KSMCV:19}
Deterministic memory hierarchy and virtualization for modern multi-core embedded systems
null
null
true
false
Tomasz {Kloda} and Marco {Solieri} and Renato {Mancuso} and Nicola {Capodieci} and Paolo {Valente} and Marko {Bertogna}
2,019
null
null
null
null
Deterministic memory hierarchy and virtualization for modern multi-core embedded systems
[PDF] The Key Role of Memory in Next-Generation Embedded Systems for ...
https://scispace.com/pdf/the-key-role-of-memory-in-next-generation-embedded-systems-43iu341iwc.pdf
Deterministic Memory Hierarchy and Virtualization for. Modern Multi-Core Embedded Systems. In 25th IEEE Real-Time and Embedded. Technology and Applications
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
xilinx-xen-cache-color
\cite{xilinx-xen-cache-color}
{Xilinx Xen Support with Cache-Coloring}
null
null
true
false
Xilinx
null
null
null
null
null
{Xilinx Xen Support with Cache-Coloring}
Cache Coloring: Interference-free Real-time Virtualization
https://xenproject.org/blog/cache-coloring-interference-free-real-time-virtualization/
Stefano Stabellini from Xilinx gave a talk on Cache Coloring, a new feature for Xen that helps better support real-time workloads.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
minerva-jailhouse
\cite{minerva-jailhouse}
{Memory-aware Jailhouse hypervisor}
null
null
true
false
{Minerva Systems}
null
null
null
null
null
{Memory-aware Jailhouse hypervisor}
Understanding the Jailhouse hypervisor, part 1 - LWN.net
https://lwn.net/Articles/578295/
To enable the hypervisor, Jailhouse needs to initialize its subsystems, create a Linux cell according to the system configuration, enable VT-x on each CPU, and, finally, migrate Linux into its cell to continue running in guest mode. The entry point is defined in hypervisor/setup.c as arch_entry, which is coded in assem...
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Survey-Way-Part
\cite{Survey-Way-Part}
A Survey on Way-Based Cache Partitioning
null
null
true
false
Das, Purnendu and Barbhuiya, Nurulla Mansur and Ranjan Roy, Bishwa
2,023
null
null
null
null
A Survey on Way-Based Cache Partitioning
Multi-Objective Memory Bandwidth Regulation and Cache ...
https://arxiv.org/html/2505.11554v1
A survey on way-based cache partitioning. In IEEE Silchar Subsection Conference (SILCON), pages 1–7, 2023. [18] ↑ Howard David, Chris
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
arm-dynamiciq
\cite{arm-dynamiciq}
{Arm DynamIQ Shared Unit Technical Reference Manual}
null
null
true
false
Arm
null
null
null
null
null
{Arm DynamIQ Shared Unit Technical Reference Manual}
Arm DynamIQ Shared Unit Technical Reference Manual
https://developer.arm.com/documentation/100453/latest/
This Technical Reference Manual is for the DynamIQ Shared Unit ( DSU ). It describes the overall structure of the DSU including the main interfaces.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
yun2013memguard
\cite{yun2013memguard}
{MemGuard}: Memory bandwidth reservation system for efficient performance isolation in multi-core platforms
null
null
true
false
Yun, Heechul and Yao, Gang and Pellizzoni, Rodolfo and Caccamo, Marco and Sha, Lui
2,013
null
null
null
null
{MemGuard}: Memory bandwidth reservation system for efficient performance isolation in multi-core platforms
SlideShare MemGuard: Memory Bandwidth Reservation System for Efficient Performance Isolation in Multicore Platforms | PPTX
https://www.slideshare.net/slideshow/mem-guard-rtas13web-25212473/25212473
This document describes MemGuard , an operating system mechanism for providing efficient per- core memory performance isolation on commercial off-the-shelf hardware. MemGuard uses memory bandwidth reservation to guarantee each core 's minimum memory bandwidth . It then performs predictive bandwidth
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
MemPol
\cite{MemPol}
{MemPol}: Policing Core Memory Bandwidth from Outside of the Cores
null
null
true
false
Alexander Zuepke and Andrea Bastoni and Weifan Chen and Marco Caccamo and Renato Mancuso
2,023
null
null
null
null
{MemPol}: Policing Core Memory Bandwidth from Outside of the Cores
[PDF] MemPol: Policing Core Memory Bandwidth from Outside of the Cores
https://blexi.de/papers/rtas2023.pdf
In this work, we present a novel regulation mechanism from outside the cores that monitors performance counters for the application core's
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
hassan2019reduced
\cite{hassan2019reduced}
{Reduced latency DRAM for multi-core safety-critical real-time systems}
null
null
true
false
Hassan, Mohamed
2,019
null
null
null
Real-Time Systems
{Reduced latency DRAM for multi-core safety-critical real-time systems}
Reduced latency DRAM for multi-core safety-critical real-time ...
https://www.ece.mcmaster.ca/faculty/hassan/assets/publications/hassan2019reduced.pdf
by M Hassan · 2019 · Cited by 16 — Targeting these systems, we promote an alternative off-chip memory solution that is based on the emerging Reduced Latency. DRAM (RLDRAM) protocol, and propose a
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
BRU:20
\cite{BRU:20}
{BRU: Bandwidth Regulation Unit for Real-Time Multicore Processors}
null
null
true
false
Farshchi, Farzad and Huang, Qijing and Yun, Heechul
2,020
null
null
null
null
{BRU: Bandwidth Regulation Unit for Real-Time Multicore Processors}
BRU: Bandwidth Regulation Unit for Real-Time Multicore ...
https://github.com/CSL-KU/bru-firesim
BRU: Bandwidth Regulation Unit for Real-Time Multicore Processors. This repository contains the necessary files to reproduce the experiment results in the
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
intel-rdt
\cite{intel-rdt}
{Resource Director Technology}
null
null
true
false
Intel
null
null
null
null
null
{Resource Director Technology}
Intel® Resource Director Technology (Intel® RDT)
https://www.intel.com/content/www/us/en/architecture-and-technology/resource-director-technology.html
Intel® Resource Director Technology enables monitoring and control over shared processor resources, improved consolidation density and reduced TCO.
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
XPCLLLL:19
\cite{XPCLLLL:19}
Holistic resource allocation for multicore real-time systems
null
null
true
false
Xu, Meng and Phan, Linh Thi Xuan and Choi, Hyon-Young and Lin, Yuhan and Li, Haoran and Lu, Chenyang and Lee, Insup
2,019
null
null
null
null
Holistic resource allocation for multicore real-time systems
Holistic resource allocation for multicore real-time systems ...
https://www.cis.upenn.edu/~linhphan/papers/rtas19-CaM-techreport.pdf
Rather than decoupling them, we compute the mapping of tasks and the allocation of shared resources to cores in an integrated resource allocation strategy called CaM that considers the demands on CPU, cache, and memory bandwidth concurrently to minimize resources while ensuring timing guarantees. Existing work has also...
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
SBMYK:22
\cite{SBMYK:22}
{A Closer Look at Intel Resource Director Technology (RDT)}
null
null
true
false
Sohal, Parul and Bechtel, Michael and Mancuso, Renato and Yun, Heechul and Krieger, Orran
2,022
null
null
null
null
{A Closer Look at Intel Resource Director Technology (RDT)}
A Closer Look at Intel Resource Director Technology (RDT)
https://dl.acm.org/doi/abs/10.1145/3534879.3534882
We aim at conducting a systematic investigation of the RDT mechanisms from a real-time perspective. We experimentally evaluate the functionality and
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
arm-mpam
\cite{arm-mpam}
{Arm Memory System Resource Partitioning and Monitoring (MPAM) System Component Specification}
null
null
true
false
Arm
null
null
null
null
null
{Arm Memory System Resource Partitioning and Monitoring (MPAM) System Component Specification}
Memory System Resource Partitioning and Monitoring (MPAM ...
https://developer.arm.com/documentation/107768/latest/Overview
MPAM is an optional Arm architecture addition for memory system partitioning. It's documented in two specifications, one for processor features and one for
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Altmeyer:2014
\cite{Altmeyer:2014}
OUTSTANDING PAPER: Evaluation of Cache Partitioning for Hard Real-Time Systems
null
null
true
false
Altmeyer, Sebastian and Douma, Roeland and Lunniss, Will and Davis, Robert I.
2,014
null
null
10.1109/ECRTS.2014.11
null
OUTSTANDING PAPER: Evaluation of Cache Partitioning for Hard Real-Time Systems
Will Lunniss - Google Scholar
https://scholar.google.com/citations?user=v_HmmSsAAAAJ&hl=en
Outstanding paper: Evaluation of cache partitioning for hard real-time systems. S Altmeyer, R Douma, W Lunniss, RI Davis. 2014 26th Euromicro Conference on Real
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Altmeyer:2016
\cite{Altmeyer:2016}
On the effectiveness of cache partitioning in hard real-time systems
null
null
true
false
Sebastian A. Altmeyer and Roeland Douma and Will Lunniss and Robert I. Davis
2,016
null
null
null
Real Time Systems
On the effectiveness of cache partitioning in hard real-time systems
[PDF] On the effectiveness of cache partitioning in hard real-time systems
https://eprints.whiterose.ac.uk/id/eprint/93504/1/art_3A10.1007_2Fs11241_015_9246_8.pdf
Cache partitioning is often suggested as a means of increasing the predictability of caches in pre-emptively scheduled hard real-time systems. The rationale
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Bui:2008
\cite{Bui:2008}
Impact of cache partitioning on multi-tasking real time embedded systems
null
null
true
false
Bui, Bach D. and Caccamo, Marco and Sha, Lui and Martinez, Joseph
2,008
null
null
null
null
Impact of cache partitioning on multi-tasking real time embedded systems
Impact of cache partitioning on multi-tasking real time ...
https://experts.illinois.edu/en/publications/impact-of-cache-partitioning-on-multi-tasking-real-time-embedded-
by BD Bui · 2008 · Cited by 159 — A case study and experiments show that in a typical real-time embedded system, the proposed algorithm is able to reduce the worst-case utilization by 15% (on
Multi-Objective Memory Bandwidth Regulation and Cache Partitioning for Multicore Real-Time Systems
2505.11554v1
Meroni:2023
\cite{Meroni:2023}
Mapping and Integration of Event- and Time-triggered Real-time Tasks on Partitioned Multi-core Systems
null
null
true
false
Meroni, Carlo and Craciunas, Silviu S. and Finzi, Anaïs and Pop, Paul
2,023
null
null
10.1109/ETFA54631.2023.10275547
null
Mapping and Integration of Event- and Time-triggered Real-time Tasks on Partitioned Multi-core Systems
Multi-Objective Memory Bandwidth Regulation and Cache ... - DROPS
https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ECRTS.2025.2
Mapping and integration of event- and time-triggered real-time tasks on partitioned multi-core systems. In 2023 IEEE 28th International Conference on