Flash Attention
Optimization Algorithm
Overview
Developed byDan Fu
Founded2022
LicenseBSD-3-Clause
Open source✓ Open Source
Use casememory-efficient attention computation
Also see
Based onattention mechanism
Knowledge graph stats
Claims25
Avg confidence94%
Avg freshness100%
Last updatedUpdated 21 days ago
Trust distribution
100% unverified
Flash Attention
concept
Memory-efficient attention algorithm that reduces memory usage during transformer inference
Compare with...supports model
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| transformer architectures | ○Unverified | High | Fresh | 1 |
| transformer models | ○Unverified | High | Fresh | 1 |
| BERT | ○Unverified | High | Fresh | 1 |
| GPT | ○Unverified | High | Fresh | 1 |
primary use case
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| memory-efficient attention computation | ○Unverified | High | Fresh | 1 |
| memory-efficient attention computation for transformers | ○Unverified | High | Fresh | 1 |
| accelerating transformer training | ○Unverified | High | Fresh | 1 |
| accelerating transformer training and inference | ○Unverified | High | Fresh | 1 |
based on
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| attention mechanism | ○Unverified | High | Fresh | 1 |
| tiling algorithm | ○Unverified | High | Fresh | 1 |
| tiling and recomputation techniques | ○Unverified | High | Fresh | 1 |
integrates with
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| CUDA | ○Unverified | High | Fresh | 1 |
| PyTorch | ○Unverified | High | Fresh | 1 |
developed by
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| Dan Fu | ○Unverified | High | Fresh | 1 |
| Atri Rudra | ○Unverified | High | Fresh | 1 |
| Stefano Ermon | ○Unverified | High | Fresh | 1 |
| Christopher Ré | ○Unverified | High | Fresh | 1 |
| Daniel Y. Fu | ○Unverified | High | Fresh | 1 |
| Tri Dao | ○Unverified | High | Fresh | 1 |
alternative to
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| standard attention mechanism | ○Unverified | High | Fresh | 1 |
| standard attention implementation | ○Unverified | High | Fresh | 1 |
founded year
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| 2022 | ○Unverified | High | Fresh | 1 |
requires
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| CUDA | ○Unverified | High | Fresh | 1 |
open source
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| true | ○Unverified | High | Fresh | 1 |
license type
| Value | Trust | Confidence | Freshness | Sources |
|---|---|---|---|---|
| BSD-3-Clause | ○Unverified | Moderate | Fresh | 1 |