BLaST: High Performance Inference and Pretraining using BLock Sparse Transformers

• URL: http://arxiv.org/abs/2507.03117v2
• Date: Sun, 26 Oct 2025 10:28:01 GMT
• Title: BLaST: High Performance Inference and Pretraining using BLock Sparse Transformers
• Authors: Patrik Okanovic, Sameer Deshmukh, Grzegorz Kwasniewski, Yi Zhu, Haruto Fujii, Sakina Fatima, Maciej Besta, Kentaro Katayama, Takumi Honda, Yusuke Nagasaka, Torsten Hoefler,
• Abstract summary: We introduce (Bl)ock (a)nd (S)parse (T)ransformers (BLaST))<n>BLaST achieves up to 95% sparsity in matrix multiplication with negligible accuracy loss.<n>We show a 2.2x inference speed Llama 3.2 with 16 GPUs with 4.45x reduction in inference memory footprint.
• Score: 18.53326467276254
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The energy consumption of large-scale ML models is dominated by data movement, shuffling billions of parameters across memory hierarchies and data centers. Sparsification offers a principled way to mitigate these costs by pruning redundant weights and activations, thereby reducing data movement. Effective sparsification to prune redundant parameters is still challenging: existing methods incur significant accuracy degradation, performance overhead, or both. We introduce (Bl)ock (a)nd (S)parse (T)ransformers (BLaST), a general, robust, and reliable method for sparsification, applicable to linear layers in all settings. Our method iteratively sparsifies weight matrices into a block sparsity pattern suitable for efficient sparse matrix-matrix (SpMM) multiplication. BLaST achieves up to 95% sparsity in MLP weights with negligible accuracy loss (majority <2.25%). We show a 2.2x inference speedup for Llama 3.2 with 16 GPUs, and up to 4.45x reduction in inference memory footprint resulting in a 2.9x reduction in GPU setup and operating costs.

Related papers

• Kernelized Sparse Fine-Tuning with Bi-level Parameter Competition for Vision Models [80.50996301430108]
  sparse tuning achieves remarkable performance by adjusting only the weights most relevant to downstream tasks.<n>We propose a one-stage method named SNELLA to overcome the above limitations.<n> SNELLA achieves SOTA performance with low memory usage.
  arXiv  Detail & Related papers  (2025-10-28T03:39:18Z)
• Low-rank Momentum Factorization for Memory Efficient Training [13.464518325870444]
  Momentum Factorized (MoFaSGD) maintains a dynamically updated low-rank SVD representation of the first-order momentum.<n>We demonstrate MoFaSGD's effectiveness on large language model benchmarks, achieving a competitive trade-off between memory reduction (e.g. LoRA) and performance.
  arXiv  Detail & Related papers  (2025-07-10T18:04:52Z)
• Accelerating Large Language Models through Partially Linear Feed-Forward Network [6.916037014703829]
  Large language models (LLMs) demonstrate remarkable capabilities but face deployment challenges due to their massive parameter counts.<n>We present a novel perspective inspired by constant folding in compiler optimization.<n>We propose TARDIS, which enables optimization of LLMs with non-linear activations by partially approximating them with linear functions in frequently occurring input ranges.<n>Experiments demonstrate that TARDIS achieves 80% parameter reduction in feed-forward networks, while significantly outperforming state-of-the-art pruning methods Wanda and RIA with up to 65% higher accuracy.
  arXiv  Detail & Related papers  (2025-01-17T09:20:56Z)
• Efficient Model Compression Techniques with FishLeg [30.69238973086908]
  FishLeg is a new second-order pruning method based on the Fisher-Legendre (FishLeg)<n>At the heart of FishLeg is a meta-learning approach to amortising the action of the inverse FIM.<n>We find that FishLeg achieves higher or comparable performance against two common baselines in the area.
  arXiv  Detail & Related papers  (2024-12-03T09:42:16Z)
• Sparser Training for On-Device Recommendation Systems [50.74019319100728]
  We propose SparseRec, a lightweight embedding method based on Dynamic Sparse Training (DST) It avoids dense gradients during backpropagation by sampling a subset of important vectors.
  arXiv  Detail & Related papers  (2024-11-19T03:48:48Z)
• Expanding Sparse Tuning for Low Memory Usage [103.43560327427647]
  We propose a method named SNELL (Sparse tuning with kerNELized LoRA) for sparse tuning with low memory usage. To achieve low memory usage, SNELL decomposes the tunable matrix for sparsification into two learnable low-rank matrices. A competition-based sparsification mechanism is further proposed to avoid the storage of tunable weight indexes.
  arXiv  Detail & Related papers  (2024-11-04T04:58:20Z)
• Efficient Arbitrary Precision Acceleration for Large Language Models on GPU Tensor Cores [3.6385567224218556]
  Large language models (LLMs) have been widely applied but face challenges in efficient inference. We introduce a novel bipolar-INT data format that facilitates parallel computing and supports symmetric quantization. We implement an arbitrary precision matrix multiplication scheme that decomposes and recovers at the bit level, enabling flexible precision.
  arXiv  Detail & Related papers  (2024-09-26T14:17:58Z)
• Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient [57.9629676017527]
  We propose an optimization-based structural pruning that learns the pruning masks in a probabilistic space directly by optimizing the loss of the pruned model.<n>We achieve this by learning an underlying Bernoulli distribution to sample binary pruning masks.<n>Experiments conducted on LLaMA, LLaMA-2, LLaMA-3, Vicuna, and Mistral models demonstrate the promising performance of our method in efficiency and effectiveness.
  arXiv  Detail & Related papers  (2024-06-15T09:31:03Z)
• Enabling High-Sparsity Foundational Llama Models with Efficient Pretraining and Deployment [56.44025052765861]
  Large language models (LLMs) have revolutionized Natural Language Processing (NLP), but their size creates computational bottlenecks. We introduce a novel approach to create accurate, sparse foundational versions of performant LLMs. We show a total speedup on CPUs for sparse-quantized LLaMA models of up to 8.6x.
  arXiv  Detail & Related papers  (2024-05-06T16:03:32Z)
• FinGPT-HPC: Efficient Pretraining and Finetuning Large Language Models for Financial Applications with High-Performance Computing [10.47214968497857]
  We present high-performance methods that exploit low-rank structures to pretrain and finetune large language models. Our methods achieve a speedup of 1.3X and a model compression ratio of 2.64X for pretaining without accuracy drop. For finetuning, our methods achieve an average accuracy increase of 6.3% and 24.0% in general tasks and financial tasks.
  arXiv  Detail & Related papers  (2024-02-21T05:03:17Z)
• HiRE: High Recall Approximate Top-$k$ Estimation for Efficient LLM Inference [68.59839755875252]
  HiRE comprises of two novel components: (i) a compression scheme to cheaply predict top-$k$ rows/columns with high recall, followed by full computation restricted to the predicted subset, and (ii) DA-TOP-$k$: an efficient multi-device approximate top-$k$ operator. We demonstrate that on a one billion parameter model, HiRE applied to both the softmax as well as feedforward layers, achieves almost matching pretraining and downstream accuracy, and speeds up inference latency by $1.47times$ on a single TPUv5e device.
  arXiv  Detail & Related papers  (2024-02-14T18:04:36Z)
• BiLLM: Pushing the Limit of Post-Training Quantization for LLMs [53.31402059062365]
  BiLLM is a groundbreaking 1-bit post-training quantization scheme tailored for pretrained large language models. It achieves for the first time high-accuracy inference (e.g. 8.41 perplexity on LLaMA2-70B) with only 1.08-bit weights across various LLMs families.
  arXiv  Detail & Related papers  (2024-02-06T09:26:34Z)
• From PEFT to DEFT: Parameter Efficient Finetuning for Reducing Activation Density in Transformers [52.199303258423306]
  We propose a novel density loss that encourages higher activation sparsity in pre-trained models. Our proposed method, textbfDEFT, can consistently reduce activation density by up to textbf44.94% on RoBERTa$_mathrmLarge$ and by textbf53.19% (encoder density) and textbf90.60% (decoder density) on Flan-T5$_mathrmXXL$.
  arXiv  Detail & Related papers  (2024-02-02T21:25:46Z)
• Sparse Fine-tuning for Inference Acceleration of Large Language Models [48.285897264669984]
  We consider the problem of accurate sparse fine-tuning of large language models (LLMs) We perform a detailed study of distillation-type losses, determining an L2-based distillation approach we term SquareHead. For MPT text generation, we show for the first time that sparse fine-tuning can reach 75% sparsity without accuracy drops.
  arXiv  Detail & Related papers  (2023-10-10T18:28:38Z)
• Quantized Neural Networks for Low-Precision Accumulation with Guaranteed Overflow Avoidance [68.8204255655161]
  We introduce a quantization-aware training algorithm that guarantees avoiding numerical overflow when reducing the precision of accumulators during inference. We evaluate our algorithm across multiple quantized models that we train for different tasks, showing that our approach can reduce the precision of accumulators while maintaining model accuracy with respect to a floating-point baseline.
  arXiv  Detail & Related papers  (2023-01-31T02:46:57Z)
• Parameter-Efficient Sparsity for Large Language Models Fine-Tuning [63.321205487234074]
  We propose a. sparse-efficient Sparse Training (PST) method to reduce the number of trainable parameters during sparse-aware training. Experiments with diverse networks (i.e., BERT, RoBERTa and GPT-2) demonstrate PST performs on par or better than previous sparsity methods.
  arXiv  Detail & Related papers  (2022-05-23T02:43:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.