Sparse Attention: A Co-Design Approach for Efficient Transformer Execution on Tensor Cores
Bibliographic record
Abstract
In recent years, the attention mechanism has demonstrated impressive performance beyond natural language processing, extending to domains such as computer vision and recommendation systems by effectively capturing contextual knowledge from the entire input sequence. However, this success comes with excessive computational costs due to an over-provisioned parameter space, making attention layers a significant contributor to inference time in Transformers. In this paper, we propose a software-hardware co-design approach to exploit sparsity in the attention mechanism and accelerate the execution of Transformers on GPUs equipped with Tensor Cores (TCs). The software component prunes attention matrices into a structured pattern to improve workload balance. The pruning algorithm provides a flexible trade-off between accuracy and sparsity. Additionally, we introduce microarchitectural support and an instruction set extension in TCs to enable efficient execution of sparse attention layers on GPUs. Our evaluations show that optimizing both hardware and software for attention layers achieves an average speed-up of 8.7% and energy savings of 54.3% with negligible impact on accuracy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.005 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".