Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention Article Swipe
Rengan Xu
,
Junjie Yang
,
Yifan Xu
,
Hong Li
,
Xing Liu
,
Devashish Shankar
,
Haoci Zhang
,
M. Liu
,
Boyang Li
,
Yuxi Hu
,
Mingwei Tang
,
Zehua Zhang
,
Tunhou Zhang
,
Dai Li
,
Sijia Chen
,
Gian-Paolo Musumeci
,
Jiaqi Zhai
,
Bill Zhu
,
Hong Yan
,
S. Jaganmohan Reddy
·
YOU?
·
· 2024
· Open Access
·
· DOI: https://doi.org/10.1145/3640457.3688040
YOU?
·
· 2024
· Open Access
·
· DOI: https://doi.org/10.1145/3640457.3688040
The integration of hardware accelerators has significantly advanced the\ncapabilities of modern recommendation systems, enabling the exploration of\ncomplex ranking paradigms previously deemed impractical. However, the GPU-based\ncomputational costs present substantial challenges. In this paper, we\ndemonstrate our development of an efficiency-driven approach to explore these\nparadigms, moving beyond traditional reliance on native PyTorch modules. We\naddress the specific challenges posed by ranking models' dependence on\ncategorical features, which vary in length and complicate GPU utilization. We\nintroduce Jagged Feature Interaction Kernels, a novel method designed to\nextract fine-grained insights from long categorical features through efficient\nhandling of dynamically sized tensors. We further enhance the performance of\nattention mechanisms by integrating Jagged tensors with Flash Attention. Our\nnovel Jagged Flash Attention achieves up to 9x speedup and 22x memory reduction\ncompared to dense attention. Notably, it also outperforms dense flash\nattention, with up to 3x speedup and 53% more memory efficiency. In production\nmodels, we observe 10% QPS improvement and 18% memory savings, enabling us to\nscale our recommendation systems with longer features and more complex\narchitectures.\n
Related Topics
Concepts
Metadata
- Type
- preprint
- Language
- en
- Landing Page
- https://doi.org/10.1145/3640457.3688040
- https://dl.acm.org/doi/pdf/10.1145/3640457.3688040
- OA Status
- gold
- References
- 2
- Related Works
- 10
- OpenAlex ID
- https://openalex.org/W4403220941
All OpenAlex metadata
Raw OpenAlex JSON
- OpenAlex ID
-
https://openalex.org/W4403220941Canonical identifier for this work in OpenAlex
- DOI
-
https://doi.org/10.1145/3640457.3688040Digital Object Identifier
- Title
-
Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash AttentionWork title
- Type
-
preprintOpenAlex work type
- Language
-
enPrimary language
- Publication year
-
2024Year of publication
- Publication date
-
2024-10-08Full publication date if available
- Authors
-
Rengan Xu, Junjie Yang, Yifan Xu, Hong Li, Xing Liu, Devashish Shankar, Haoci Zhang, M. Liu, Boyang Li, Yuxi Hu, Mingwei Tang, Zehua Zhang, Tunhou Zhang, Dai Li, Sijia Chen, Gian-Paolo Musumeci, Jiaqi Zhai, Bill Zhu, Hong Yan, S. Jaganmohan ReddyList of authors in order
- Landing page
-
https://doi.org/10.1145/3640457.3688040Publisher landing page
- PDF URL
-
https://dl.acm.org/doi/pdf/10.1145/3640457.3688040Direct link to full text PDF
- Open access
-
YesWhether a free full text is available
- OA status
-
goldOpen access status per OpenAlex
- OA URL
-
https://dl.acm.org/doi/pdf/10.1145/3640457.3688040Direct OA link when available
- Concepts
-
Scalability, Computer science, Flash (photography), Scale (ratio), Database, Physics, Optics, Quantum mechanicsTop concepts (fields/topics) attached by OpenAlex
- Cited by
-
0Total citation count in OpenAlex
- References (count)
-
2Number of works referenced by this work
- Related works (count)
-
10Other works algorithmically related by OpenAlex
Full payload
| id | https://openalex.org/W4403220941 |
|---|---|
| doi | https://doi.org/10.1145/3640457.3688040 |
| ids.doi | https://doi.org/10.1145/3640457.3688040 |
| ids.openalex | https://openalex.org/W4403220941 |
| fwci | 0.0 |
| type | preprint |
| title | Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention |
| biblio.issue | |
| biblio.volume | |
| biblio.last_page | 780 |
| biblio.first_page | 778 |
| topics[0].id | https://openalex.org/T10101 |
| topics[0].field.id | https://openalex.org/fields/17 |
| topics[0].field.display_name | Computer Science |
| topics[0].score | 0.9973999857902527 |
| topics[0].domain.id | https://openalex.org/domains/3 |
| topics[0].domain.display_name | Physical Sciences |
| topics[0].subfield.id | https://openalex.org/subfields/1710 |
| topics[0].subfield.display_name | Information Systems |
| topics[0].display_name | Cloud Computing and Resource Management |
| topics[1].id | https://openalex.org/T10054 |
| topics[1].field.id | https://openalex.org/fields/17 |
| topics[1].field.display_name | Computer Science |
| topics[1].score | 0.9973000288009644 |
| topics[1].domain.id | https://openalex.org/domains/3 |
| topics[1].domain.display_name | Physical Sciences |
| topics[1].subfield.id | https://openalex.org/subfields/1708 |
| topics[1].subfield.display_name | Hardware and Architecture |
| topics[1].display_name | Parallel Computing and Optimization Techniques |
| topics[2].id | https://openalex.org/T11181 |
| topics[2].field.id | https://openalex.org/fields/17 |
| topics[2].field.display_name | Computer Science |
| topics[2].score | 0.9966999888420105 |
| topics[2].domain.id | https://openalex.org/domains/3 |
| topics[2].domain.display_name | Physical Sciences |
| topics[2].subfield.id | https://openalex.org/subfields/1705 |
| topics[2].subfield.display_name | Computer Networks and Communications |
| topics[2].display_name | Advanced Data Storage Technologies |
| is_xpac | False |
| apc_list | |
| apc_paid | |
| concepts[0].id | https://openalex.org/C48044578 |
| concepts[0].level | 2 |
| concepts[0].score | 0.7890788316726685 |
| concepts[0].wikidata | https://www.wikidata.org/wiki/Q727490 |
| concepts[0].display_name | Scalability |
| concepts[1].id | https://openalex.org/C41008148 |
| concepts[1].level | 0 |
| concepts[1].score | 0.7093556523323059 |
| concepts[1].wikidata | https://www.wikidata.org/wiki/Q21198 |
| concepts[1].display_name | Computer science |
| concepts[2].id | https://openalex.org/C2777526259 |
| concepts[2].level | 2 |
| concepts[2].score | 0.5936803817749023 |
| concepts[2].wikidata | https://www.wikidata.org/wiki/Q221836 |
| concepts[2].display_name | Flash (photography) |
| concepts[3].id | https://openalex.org/C2778755073 |
| concepts[3].level | 2 |
| concepts[3].score | 0.5764390230178833 |
| concepts[3].wikidata | https://www.wikidata.org/wiki/Q10858537 |
| concepts[3].display_name | Scale (ratio) |
| concepts[4].id | https://openalex.org/C77088390 |
| concepts[4].level | 1 |
| concepts[4].score | 0.16413769125938416 |
| concepts[4].wikidata | https://www.wikidata.org/wiki/Q8513 |
| concepts[4].display_name | Database |
| concepts[5].id | https://openalex.org/C121332964 |
| concepts[5].level | 0 |
| concepts[5].score | 0.06407561898231506 |
| concepts[5].wikidata | https://www.wikidata.org/wiki/Q413 |
| concepts[5].display_name | Physics |
| concepts[6].id | https://openalex.org/C120665830 |
| concepts[6].level | 1 |
| concepts[6].score | 0.0 |
| concepts[6].wikidata | https://www.wikidata.org/wiki/Q14620 |
| concepts[6].display_name | Optics |
| concepts[7].id | https://openalex.org/C62520636 |
| concepts[7].level | 1 |
| concepts[7].score | 0.0 |
| concepts[7].wikidata | https://www.wikidata.org/wiki/Q944 |
| concepts[7].display_name | Quantum mechanics |
| keywords[0].id | https://openalex.org/keywords/scalability |
| keywords[0].score | 0.7890788316726685 |
| keywords[0].display_name | Scalability |
| keywords[1].id | https://openalex.org/keywords/computer-science |
| keywords[1].score | 0.7093556523323059 |
| keywords[1].display_name | Computer science |
| keywords[2].id | https://openalex.org/keywords/flash |
| keywords[2].score | 0.5936803817749023 |
| keywords[2].display_name | Flash (photography) |
| keywords[3].id | https://openalex.org/keywords/scale |
| keywords[3].score | 0.5764390230178833 |
| keywords[3].display_name | Scale (ratio) |
| keywords[4].id | https://openalex.org/keywords/database |
| keywords[4].score | 0.16413769125938416 |
| keywords[4].display_name | Database |
| keywords[5].id | https://openalex.org/keywords/physics |
| keywords[5].score | 0.06407561898231506 |
| keywords[5].display_name | Physics |
| language | en |
| locations[0].id | doi:10.1145/3640457.3688040 |
| locations[0].is_oa | True |
| locations[0].source | |
| locations[0].license | |
| locations[0].pdf_url | https://dl.acm.org/doi/pdf/10.1145/3640457.3688040 |
| locations[0].version | publishedVersion |
| locations[0].raw_type | proceedings-article |
| locations[0].license_id | |
| locations[0].is_accepted | True |
| locations[0].is_published | True |
| locations[0].raw_source_name | 18th ACM Conference on Recommender Systems |
| locations[0].landing_page_url | https://doi.org/10.1145/3640457.3688040 |
| locations[1].id | pmh:oai:arXiv.org:2409.15373 |
| locations[1].is_oa | True |
| locations[1].source.id | https://openalex.org/S4306400194 |
| locations[1].source.issn | |
| locations[1].source.type | repository |
| locations[1].source.is_oa | True |
| locations[1].source.issn_l | |
| locations[1].source.is_core | False |
| locations[1].source.is_in_doaj | False |
| locations[1].source.display_name | arXiv (Cornell University) |
| locations[1].source.host_organization | https://openalex.org/I205783295 |
| locations[1].source.host_organization_name | Cornell University |
| locations[1].source.host_organization_lineage | https://openalex.org/I205783295 |
| locations[1].license | |
| locations[1].pdf_url | https://arxiv.org/pdf/2409.15373 |
| locations[1].version | submittedVersion |
| locations[1].raw_type | text |
| locations[1].license_id | |
| locations[1].is_accepted | False |
| locations[1].is_published | False |
| locations[1].raw_source_name | |
| locations[1].landing_page_url | http://arxiv.org/abs/2409.15373 |
| indexed_in | arxiv, crossref |
| authorships[0].author.id | https://openalex.org/A5083192531 |
| authorships[0].author.orcid | https://orcid.org/0000-0002-5230-5530 |
| authorships[0].author.display_name | Rengan Xu |
| authorships[0].countries | US |
| authorships[0].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[0].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[0].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[0].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[0].institutions[0].type | company |
| authorships[0].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[0].institutions[0].country_code | US |
| authorships[0].institutions[0].display_name | Meta (United States) |
| authorships[0].author_position | first |
| authorships[0].raw_author_name | Rengan Xu |
| authorships[0].is_corresponding | False |
| authorships[0].raw_affiliation_strings | Meta Platforms, USA |
| authorships[1].author.id | https://openalex.org/A5005149567 |
| authorships[1].author.orcid | https://orcid.org/0000-0003-0603-4482 |
| authorships[1].author.display_name | Junjie Yang |
| authorships[1].countries | US |
| authorships[1].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[1].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[1].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[1].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[1].institutions[0].type | company |
| authorships[1].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[1].institutions[0].country_code | US |
| authorships[1].institutions[0].display_name | Meta (United States) |
| authorships[1].author_position | middle |
| authorships[1].raw_author_name | Junjie Yang |
| authorships[1].is_corresponding | False |
| authorships[1].raw_affiliation_strings | Meta Platforms, USA |
| authorships[2].author.id | https://openalex.org/A5057666775 |
| authorships[2].author.orcid | https://orcid.org/0000-0003-2676-6613 |
| authorships[2].author.display_name | Yifan Xu |
| authorships[2].countries | US |
| authorships[2].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[2].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[2].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[2].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[2].institutions[0].type | company |
| authorships[2].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[2].institutions[0].country_code | US |
| authorships[2].institutions[0].display_name | Meta (United States) |
| authorships[2].author_position | middle |
| authorships[2].raw_author_name | Yifan Xu |
| authorships[2].is_corresponding | False |
| authorships[2].raw_affiliation_strings | Meta Platforms, USA |
| authorships[3].author.id | https://openalex.org/A5112833347 |
| authorships[3].author.orcid | |
| authorships[3].author.display_name | Hong Li |
| authorships[3].countries | US |
| authorships[3].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[3].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[3].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[3].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[3].institutions[0].type | company |
| authorships[3].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[3].institutions[0].country_code | US |
| authorships[3].institutions[0].display_name | Meta (United States) |
| authorships[3].author_position | middle |
| authorships[3].raw_author_name | Hong Li |
| authorships[3].is_corresponding | False |
| authorships[3].raw_affiliation_strings | Meta Platforms, USA |
| authorships[4].author.id | https://openalex.org/A5100381490 |
| authorships[4].author.orcid | |
| authorships[4].author.display_name | Xing Liu |
| authorships[4].countries | US |
| authorships[4].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[4].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[4].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[4].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[4].institutions[0].type | company |
| authorships[4].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[4].institutions[0].country_code | US |
| authorships[4].institutions[0].display_name | Meta (United States) |
| authorships[4].author_position | middle |
| authorships[4].raw_author_name | Xing Liu |
| authorships[4].is_corresponding | False |
| authorships[4].raw_affiliation_strings | Meta Platforms, USA |
| authorships[5].author.id | https://openalex.org/A5057026035 |
| authorships[5].author.orcid | |
| authorships[5].author.display_name | Devashish Shankar |
| authorships[5].countries | US |
| authorships[5].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[5].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[5].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[5].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[5].institutions[0].type | company |
| authorships[5].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[5].institutions[0].country_code | US |
| authorships[5].institutions[0].display_name | Meta (United States) |
| authorships[5].author_position | middle |
| authorships[5].raw_author_name | Devashish Shankar |
| authorships[5].is_corresponding | False |
| authorships[5].raw_affiliation_strings | Meta Platforms, USA |
| authorships[6].author.id | https://openalex.org/A5055539155 |
| authorships[6].author.orcid | |
| authorships[6].author.display_name | Haoci Zhang |
| authorships[6].countries | US |
| authorships[6].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[6].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[6].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[6].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[6].institutions[0].type | company |
| authorships[6].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[6].institutions[0].country_code | US |
| authorships[6].institutions[0].display_name | Meta (United States) |
| authorships[6].author_position | middle |
| authorships[6].raw_author_name | Haoci Zhang |
| authorships[6].is_corresponding | False |
| authorships[6].raw_affiliation_strings | Meta Platforms, USA |
| authorships[7].author.id | https://openalex.org/A5112557620 |
| authorships[7].author.orcid | https://orcid.org/0000-0002-1039-1565 |
| authorships[7].author.display_name | M. Liu |
| authorships[7].countries | US |
| authorships[7].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[7].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[7].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[7].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[7].institutions[0].type | company |
| authorships[7].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[7].institutions[0].country_code | US |
| authorships[7].institutions[0].display_name | Meta (United States) |
| authorships[7].author_position | middle |
| authorships[7].raw_author_name | Meng Liu |
| authorships[7].is_corresponding | False |
| authorships[7].raw_affiliation_strings | Meta Platforms, USA |
| authorships[8].author.id | https://openalex.org/A5007325176 |
| authorships[8].author.orcid | |
| authorships[8].author.display_name | Boyang Li |
| authorships[8].countries | US |
| authorships[8].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[8].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[8].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[8].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[8].institutions[0].type | company |
| authorships[8].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[8].institutions[0].country_code | US |
| authorships[8].institutions[0].display_name | Meta (United States) |
| authorships[8].author_position | middle |
| authorships[8].raw_author_name | Boyang Li |
| authorships[8].is_corresponding | False |
| authorships[8].raw_affiliation_strings | Meta Platforms, USA |
| authorships[9].author.id | https://openalex.org/A5073566291 |
| authorships[9].author.orcid | |
| authorships[9].author.display_name | Yuxi Hu |
| authorships[9].countries | US |
| authorships[9].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[9].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[9].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[9].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[9].institutions[0].type | company |
| authorships[9].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[9].institutions[0].country_code | US |
| authorships[9].institutions[0].display_name | Meta (United States) |
| authorships[9].author_position | middle |
| authorships[9].raw_author_name | Yuxi Hu |
| authorships[9].is_corresponding | False |
| authorships[9].raw_affiliation_strings | Meta Platforms, USA |
| authorships[10].author.id | https://openalex.org/A5063149441 |
| authorships[10].author.orcid | |
| authorships[10].author.display_name | Mingwei Tang |
| authorships[10].countries | US |
| authorships[10].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[10].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[10].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[10].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[10].institutions[0].type | company |
| authorships[10].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[10].institutions[0].country_code | US |
| authorships[10].institutions[0].display_name | Meta (United States) |
| authorships[10].author_position | middle |
| authorships[10].raw_author_name | Mingwei Tang |
| authorships[10].is_corresponding | False |
| authorships[10].raw_affiliation_strings | Meta Platforms, USA |
| authorships[11].author.id | https://openalex.org/A5113198800 |
| authorships[11].author.orcid | |
| authorships[11].author.display_name | Zehua Zhang |
| authorships[11].countries | US |
| authorships[11].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[11].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[11].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[11].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[11].institutions[0].type | company |
| authorships[11].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[11].institutions[0].country_code | US |
| authorships[11].institutions[0].display_name | Meta (United States) |
| authorships[11].author_position | middle |
| authorships[11].raw_author_name | Zehua Zhang |
| authorships[11].is_corresponding | False |
| authorships[11].raw_affiliation_strings | Meta Platforms, USA |
| authorships[12].author.id | https://openalex.org/A5052236334 |
| authorships[12].author.orcid | https://orcid.org/0000-0001-9590-9433 |
| authorships[12].author.display_name | Tunhou Zhang |
| authorships[12].countries | US |
| authorships[12].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[12].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[12].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[12].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[12].institutions[0].type | company |
| authorships[12].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[12].institutions[0].country_code | US |
| authorships[12].institutions[0].display_name | Meta (United States) |
| authorships[12].author_position | middle |
| authorships[12].raw_author_name | Tunhou Zhang |
| authorships[12].is_corresponding | False |
| authorships[12].raw_affiliation_strings | Meta Platforms, USA |
| authorships[13].author.id | https://openalex.org/A5102999050 |
| authorships[13].author.orcid | https://orcid.org/0009-0000-6645-5620 |
| authorships[13].author.display_name | Dai Li |
| authorships[13].countries | US |
| authorships[13].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[13].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[13].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[13].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[13].institutions[0].type | company |
| authorships[13].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[13].institutions[0].country_code | US |
| authorships[13].institutions[0].display_name | Meta (United States) |
| authorships[13].author_position | middle |
| authorships[13].raw_author_name | Dai Li |
| authorships[13].is_corresponding | False |
| authorships[13].raw_affiliation_strings | Meta Platforms, USA |
| authorships[14].author.id | https://openalex.org/A5072440584 |
| authorships[14].author.orcid | https://orcid.org/0000-0002-8098-5852 |
| authorships[14].author.display_name | Sijia Chen |
| authorships[14].countries | US |
| authorships[14].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[14].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[14].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[14].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[14].institutions[0].type | company |
| authorships[14].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[14].institutions[0].country_code | US |
| authorships[14].institutions[0].display_name | Meta (United States) |
| authorships[14].author_position | middle |
| authorships[14].raw_author_name | Sijia Chen |
| authorships[14].is_corresponding | False |
| authorships[14].raw_affiliation_strings | Meta Platforms, USA |
| authorships[15].author.id | https://openalex.org/A5107823466 |
| authorships[15].author.orcid | |
| authorships[15].author.display_name | Gian-Paolo Musumeci |
| authorships[15].countries | US |
| authorships[15].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[15].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[15].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[15].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[15].institutions[0].type | company |
| authorships[15].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[15].institutions[0].country_code | US |
| authorships[15].institutions[0].display_name | Meta (United States) |
| authorships[15].author_position | middle |
| authorships[15].raw_author_name | Gian-Paolo Musumeci |
| authorships[15].is_corresponding | False |
| authorships[15].raw_affiliation_strings | Meta Platforms, USA |
| authorships[16].author.id | https://openalex.org/A5100349821 |
| authorships[16].author.orcid | https://orcid.org/0009-0004-7279-3318 |
| authorships[16].author.display_name | Jiaqi Zhai |
| authorships[16].countries | US |
| authorships[16].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[16].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[16].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[16].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[16].institutions[0].type | company |
| authorships[16].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[16].institutions[0].country_code | US |
| authorships[16].institutions[0].display_name | Meta (United States) |
| authorships[16].author_position | middle |
| authorships[16].raw_author_name | Jiaqi Zhai |
| authorships[16].is_corresponding | False |
| authorships[16].raw_affiliation_strings | Meta Platforms, USA |
| authorships[17].author.id | https://openalex.org/A5104325322 |
| authorships[17].author.orcid | |
| authorships[17].author.display_name | Bill Zhu |
| authorships[17].countries | US |
| authorships[17].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[17].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[17].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[17].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[17].institutions[0].type | company |
| authorships[17].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[17].institutions[0].country_code | US |
| authorships[17].institutions[0].display_name | Meta (United States) |
| authorships[17].author_position | middle |
| authorships[17].raw_author_name | Bill Zhu |
| authorships[17].is_corresponding | False |
| authorships[17].raw_affiliation_strings | Meta Platforms, USA |
| authorships[18].author.id | https://openalex.org/A5076961426 |
| authorships[18].author.orcid | |
| authorships[18].author.display_name | Hong Yan |
| authorships[18].countries | US |
| authorships[18].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[18].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[18].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[18].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[18].institutions[0].type | company |
| authorships[18].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[18].institutions[0].country_code | US |
| authorships[18].institutions[0].display_name | Meta (United States) |
| authorships[18].author_position | middle |
| authorships[18].raw_author_name | Hong Yan |
| authorships[18].is_corresponding | False |
| authorships[18].raw_affiliation_strings | Meta Platforms, USA |
| authorships[19].author.id | https://openalex.org/A5101926449 |
| authorships[19].author.orcid | https://orcid.org/0009-0001-1504-9013 |
| authorships[19].author.display_name | S. Jaganmohan Reddy |
| authorships[19].countries | US |
| authorships[19].affiliations[0].institution_ids | https://openalex.org/I4210114444 |
| authorships[19].affiliations[0].raw_affiliation_string | Meta Platforms, USA |
| authorships[19].institutions[0].id | https://openalex.org/I4210114444 |
| authorships[19].institutions[0].ror | https://ror.org/01zbnvs85 |
| authorships[19].institutions[0].type | company |
| authorships[19].institutions[0].lineage | https://openalex.org/I4210114444 |
| authorships[19].institutions[0].country_code | US |
| authorships[19].institutions[0].display_name | Meta (United States) |
| authorships[19].author_position | last |
| authorships[19].raw_author_name | Srihari Reddy |
| authorships[19].is_corresponding | False |
| authorships[19].raw_affiliation_strings | Meta Platforms, USA |
| has_content.pdf | True |
| has_content.grobid_xml | False |
| is_paratext | False |
| open_access.is_oa | True |
| open_access.oa_url | https://dl.acm.org/doi/pdf/10.1145/3640457.3688040 |
| open_access.oa_status | gold |
| open_access.any_repository_has_fulltext | False |
| created_date | 2025-10-10T00:00:00 |
| display_name | Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention |
| has_fulltext | False |
| is_retracted | False |
| updated_date | 2025-11-06T06:51:31.235846 |
| primary_topic.id | https://openalex.org/T10101 |
| primary_topic.field.id | https://openalex.org/fields/17 |
| primary_topic.field.display_name | Computer Science |
| primary_topic.score | 0.9973999857902527 |
| primary_topic.domain.id | https://openalex.org/domains/3 |
| primary_topic.domain.display_name | Physical Sciences |
| primary_topic.subfield.id | https://openalex.org/subfields/1710 |
| primary_topic.subfield.display_name | Information Systems |
| primary_topic.display_name | Cloud Computing and Resource Management |
| related_works | https://openalex.org/W4391375266, https://openalex.org/W2899084033, https://openalex.org/W2748952813, https://openalex.org/W2390279801, https://openalex.org/W4391913857, https://openalex.org/W2358668433, https://openalex.org/W4396701345, https://openalex.org/W2376932109, https://openalex.org/W2001405890, https://openalex.org/W2389214306 |
| cited_by_count | 0 |
| locations_count | 2 |
| best_oa_location.id | doi:10.1145/3640457.3688040 |
| best_oa_location.is_oa | True |
| best_oa_location.source | |
| best_oa_location.license | |
| best_oa_location.pdf_url | https://dl.acm.org/doi/pdf/10.1145/3640457.3688040 |
| best_oa_location.version | publishedVersion |
| best_oa_location.raw_type | proceedings-article |
| best_oa_location.license_id | |
| best_oa_location.is_accepted | True |
| best_oa_location.is_published | True |
| best_oa_location.raw_source_name | 18th ACM Conference on Recommender Systems |
| best_oa_location.landing_page_url | https://doi.org/10.1145/3640457.3688040 |
| primary_location.id | doi:10.1145/3640457.3688040 |
| primary_location.is_oa | True |
| primary_location.source | |
| primary_location.license | |
| primary_location.pdf_url | https://dl.acm.org/doi/pdf/10.1145/3640457.3688040 |
| primary_location.version | publishedVersion |
| primary_location.raw_type | proceedings-article |
| primary_location.license_id | |
| primary_location.is_accepted | True |
| primary_location.is_published | True |
| primary_location.raw_source_name | 18th ACM Conference on Recommender Systems |
| primary_location.landing_page_url | https://doi.org/10.1145/3640457.3688040 |
| publication_date | 2024-10-08 |
| publication_year | 2024 |
| referenced_works | https://openalex.org/W4296591817, https://openalex.org/W2954698171 |
| referenced_works_count | 2 |
| abstract_inverted_index.a | 74 |
| abstract_inverted_index.3x | 130 |
| abstract_inverted_index.9x | 112 |
| abstract_inverted_index.In | 29, 137 |
| abstract_inverted_index.We | 91 |
| abstract_inverted_index.an | 36 |
| abstract_inverted_index.by | 55, 98 |
| abstract_inverted_index.in | 63 |
| abstract_inverted_index.it | 122 |
| abstract_inverted_index.of | 2, 9, 35, 87 |
| abstract_inverted_index.on | 46 |
| abstract_inverted_index.to | 39, 111, 118, 129 |
| abstract_inverted_index.up | 110, 128 |
| abstract_inverted_index.us | 149 |
| abstract_inverted_index.we | 139 |
| abstract_inverted_index.10% | 141 |
| abstract_inverted_index.18% | 145 |
| abstract_inverted_index.22x | 115 |
| abstract_inverted_index.53% | 133 |
| abstract_inverted_index.GPU | 67 |
| abstract_inverted_index.QPS | 142 |
| abstract_inverted_index.The | 0 |
| abstract_inverted_index.and | 65, 114, 132, 144, 157 |
| abstract_inverted_index.has | 5 |
| abstract_inverted_index.our | 33, 151 |
| abstract_inverted_index.the | 14, 23, 51, 94 |
| abstract_inverted_index.also | 123 |
| abstract_inverted_index.from | 81 |
| abstract_inverted_index.long | 82 |
| abstract_inverted_index.more | 134, 158 |
| abstract_inverted_index.this | 30 |
| abstract_inverted_index.vary | 62 |
| abstract_inverted_index.with | 102, 127, 154 |
| abstract_inverted_index.Flash | 103, 107 |
| abstract_inverted_index.costs | 25 |
| abstract_inverted_index.dense | 119, 125 |
| abstract_inverted_index.novel | 75 |
| abstract_inverted_index.posed | 54 |
| abstract_inverted_index.sized | 89 |
| abstract_inverted_index.which | 61 |
| abstract_inverted_index.Jagged | 70, 100, 106 |
| abstract_inverted_index.beyond | 43 |
| abstract_inverted_index.deemed | 20 |
| abstract_inverted_index.length | 64 |
| abstract_inverted_index.longer | 155 |
| abstract_inverted_index.memory | 116, 135, 146 |
| abstract_inverted_index.method | 76 |
| abstract_inverted_index.modern | 10 |
| abstract_inverted_index.moving | 42 |
| abstract_inverted_index.native | 47 |
| abstract_inverted_index.paper, | 31 |
| abstract_inverted_index.Feature | 71 |
| abstract_inverted_index.PyTorch | 48 |
| abstract_inverted_index.enhance | 93 |
| abstract_inverted_index.explore | 40 |
| abstract_inverted_index.further | 92 |
| abstract_inverted_index.models' | 57 |
| abstract_inverted_index.observe | 140 |
| abstract_inverted_index.present | 26 |
| abstract_inverted_index.ranking | 17, 56 |
| abstract_inverted_index.speedup | 113, 131 |
| abstract_inverted_index.systems | 153 |
| abstract_inverted_index.tensors | 101 |
| abstract_inverted_index.through | 85 |
| abstract_inverted_index.However, | 22 |
| abstract_inverted_index.Kernels, | 73 |
| abstract_inverted_index.Notably, | 121 |
| abstract_inverted_index.achieves | 109 |
| abstract_inverted_index.advanced | 7 |
| abstract_inverted_index.approach | 38 |
| abstract_inverted_index.designed | 77 |
| abstract_inverted_index.enabling | 13, 148 |
| abstract_inverted_index.features | 84, 156 |
| abstract_inverted_index.hardware | 3 |
| abstract_inverted_index.insights | 80 |
| abstract_inverted_index.modules. | 49 |
| abstract_inverted_index.reliance | 45 |
| abstract_inverted_index.savings, | 147 |
| abstract_inverted_index.specific | 52 |
| abstract_inverted_index.systems, | 12 |
| abstract_inverted_index.tensors. | 90 |
| abstract_inverted_index.Attention | 108 |
| abstract_inverted_index.features, | 60 |
| abstract_inverted_index.paradigms | 18 |
| abstract_inverted_index.to\nscale | 150 |
| abstract_inverted_index.Attention. | 104 |
| abstract_inverted_index.Our\nnovel | 105 |
| abstract_inverted_index.attention. | 120 |
| abstract_inverted_index.challenges | 53 |
| abstract_inverted_index.complicate | 66 |
| abstract_inverted_index.dependence | 58 |
| abstract_inverted_index.mechanisms | 97 |
| abstract_inverted_index.previously | 19 |
| abstract_inverted_index.Interaction | 72 |
| abstract_inverted_index.We\naddress | 50 |
| abstract_inverted_index.categorical | 83 |
| abstract_inverted_index.challenges. | 28 |
| abstract_inverted_index.development | 34 |
| abstract_inverted_index.dynamically | 88 |
| abstract_inverted_index.efficiency. | 136 |
| abstract_inverted_index.exploration | 15 |
| abstract_inverted_index.improvement | 143 |
| abstract_inverted_index.integrating | 99 |
| abstract_inverted_index.integration | 1 |
| abstract_inverted_index.of\ncomplex | 16 |
| abstract_inverted_index.outperforms | 124 |
| abstract_inverted_index.performance | 95 |
| abstract_inverted_index.substantial | 27 |
| abstract_inverted_index.to\nextract | 78 |
| abstract_inverted_index.traditional | 44 |
| abstract_inverted_index.accelerators | 4 |
| abstract_inverted_index.fine-grained | 79 |
| abstract_inverted_index.impractical. | 21 |
| abstract_inverted_index.utilization. | 68 |
| abstract_inverted_index.We\nintroduce | 69 |
| abstract_inverted_index.of\nattention | 96 |
| abstract_inverted_index.significantly | 6 |
| abstract_inverted_index.recommendation | 11, 152 |
| abstract_inverted_index.on\ncategorical | 59 |
| abstract_inverted_index.we\ndemonstrate | 32 |
| abstract_inverted_index.efficiency-driven | 37 |
| abstract_inverted_index.flash\nattention, | 126 |
| abstract_inverted_index.the\ncapabilities | 8 |
| abstract_inverted_index.these\nparadigms, | 41 |
| abstract_inverted_index.efficient\nhandling | 86 |
| abstract_inverted_index.production\nmodels, | 138 |
| abstract_inverted_index.reduction\ncompared | 117 |
| abstract_inverted_index.GPU-based\ncomputational | 24 |
| abstract_inverted_index.complex\narchitectures.\n | 159 |
| cited_by_percentile_year | |
| countries_distinct_count | 1 |
| institutions_distinct_count | 20 |
| citation_normalized_percentile.value | 0.2956519 |
| citation_normalized_percentile.is_in_top_1_percent | False |
| citation_normalized_percentile.is_in_top_10_percent | False |