NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks Article Swipe
Yongchang Hao
,
Yanshuai Cao
,
Lili Mou
·
YOU?
·
· 2024
· Open Access
·
· DOI: https://doi.org/10.48550/arxiv.2410.20650
YOU?
·
· 2024
· Open Access
·
· DOI: https://doi.org/10.48550/arxiv.2410.20650
The performance of neural networks improves when more parameters are used. However, the model sizes are constrained by the available on-device memory during training and inference. Although applying techniques like quantization can alleviate the constraint, they suffer from performance degradation. In this work, we introduce NeuZip, a new weight compression scheme based on the entropy of floating-point numbers in neural networks. With NeuZip, we are able to achieve memory-efficient training and inference without sacrificing performance. Notably, we significantly reduce the memory footprint of training a Llama-3 8B model from 31GB to less than 16GB, while keeping the training dynamics fully unchanged. In inference, our method can reduce memory usage by more than half while maintaining near-lossless performance. Our code is publicly available.
Related Topics
Concepts
Metadata
- Type
- preprint
- Language
- en
- Landing Page
- http://arxiv.org/abs/2410.20650
- https://arxiv.org/pdf/2410.20650
- OA Status
- green
- Related Works
- 10
- OpenAlex ID
- https://openalex.org/W4404314261
All OpenAlex metadata
Raw OpenAlex JSON
- OpenAlex ID
-
https://openalex.org/W4404314261Canonical identifier for this work in OpenAlex
- DOI
-
https://doi.org/10.48550/arxiv.2410.20650Digital Object Identifier
- Title
-
NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural NetworksWork title
- Type
-
preprintOpenAlex work type
- Language
-
enPrimary language
- Publication year
-
2024Year of publication
- Publication date
-
2024-10-28Full publication date if available
- Authors
-
Yongchang Hao, Yanshuai Cao, Lili MouList of authors in order
- Landing page
-
https://arxiv.org/abs/2410.20650Publisher landing page
- PDF URL
-
https://arxiv.org/pdf/2410.20650Direct link to full text PDF
- Open access
-
YesWhether a free full text is available
- OA status
-
greenOpen access status per OpenAlex
- OA URL
-
https://arxiv.org/pdf/2410.20650Direct OA link when available
- Concepts
-
Inference, Training (meteorology), Computer science, Artificial neural network, Compression (physics), Artificial intelligence, Machine learning, Geography, Materials science, Meteorology, Composite materialTop concepts (fields/topics) attached by OpenAlex
- Cited by
-
0Total citation count in OpenAlex
- Related works (count)
-
10Other works algorithmically related by OpenAlex
Full payload
| id | https://openalex.org/W4404314261 |
|---|---|
| doi | https://doi.org/10.48550/arxiv.2410.20650 |
| ids.doi | https://doi.org/10.48550/arxiv.2410.20650 |
| ids.openalex | https://openalex.org/W4404314261 |
| fwci | |
| type | preprint |
| title | NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks |
| biblio.issue | |
| biblio.volume | |
| biblio.last_page | |
| biblio.first_page | |
| topics[0].id | https://openalex.org/T10320 |
| topics[0].field.id | https://openalex.org/fields/17 |
| topics[0].field.display_name | Computer Science |
| topics[0].score | 0.883899986743927 |
| topics[0].domain.id | https://openalex.org/domains/3 |
| topics[0].domain.display_name | Physical Sciences |
| topics[0].subfield.id | https://openalex.org/subfields/1702 |
| topics[0].subfield.display_name | Artificial Intelligence |
| topics[0].display_name | Neural Networks and Applications |
| is_xpac | False |
| apc_list | |
| apc_paid | |
| concepts[0].id | https://openalex.org/C2776214188 |
| concepts[0].level | 2 |
| concepts[0].score | 0.7317643165588379 |
| concepts[0].wikidata | https://www.wikidata.org/wiki/Q408386 |
| concepts[0].display_name | Inference |
| concepts[1].id | https://openalex.org/C2777211547 |
| concepts[1].level | 2 |
| concepts[1].score | 0.7114962339401245 |
| concepts[1].wikidata | https://www.wikidata.org/wiki/Q17141490 |
| concepts[1].display_name | Training (meteorology) |
| concepts[2].id | https://openalex.org/C41008148 |
| concepts[2].level | 0 |
| concepts[2].score | 0.6462578773498535 |
| concepts[2].wikidata | https://www.wikidata.org/wiki/Q21198 |
| concepts[2].display_name | Computer science |
| concepts[3].id | https://openalex.org/C50644808 |
| concepts[3].level | 2 |
| concepts[3].score | 0.5890853404998779 |
| concepts[3].wikidata | https://www.wikidata.org/wiki/Q192776 |
| concepts[3].display_name | Artificial neural network |
| concepts[4].id | https://openalex.org/C180016635 |
| concepts[4].level | 2 |
| concepts[4].score | 0.5213414430618286 |
| concepts[4].wikidata | https://www.wikidata.org/wiki/Q2712821 |
| concepts[4].display_name | Compression (physics) |
| concepts[5].id | https://openalex.org/C154945302 |
| concepts[5].level | 1 |
| concepts[5].score | 0.47112107276916504 |
| concepts[5].wikidata | https://www.wikidata.org/wiki/Q11660 |
| concepts[5].display_name | Artificial intelligence |
| concepts[6].id | https://openalex.org/C119857082 |
| concepts[6].level | 1 |
| concepts[6].score | 0.38877755403518677 |
| concepts[6].wikidata | https://www.wikidata.org/wiki/Q2539 |
| concepts[6].display_name | Machine learning |
| concepts[7].id | https://openalex.org/C205649164 |
| concepts[7].level | 0 |
| concepts[7].score | 0.040221452713012695 |
| concepts[7].wikidata | https://www.wikidata.org/wiki/Q1071 |
| concepts[7].display_name | Geography |
| concepts[8].id | https://openalex.org/C192562407 |
| concepts[8].level | 0 |
| concepts[8].score | 0.03392484784126282 |
| concepts[8].wikidata | https://www.wikidata.org/wiki/Q228736 |
| concepts[8].display_name | Materials science |
| concepts[9].id | https://openalex.org/C153294291 |
| concepts[9].level | 1 |
| concepts[9].score | 0.0 |
| concepts[9].wikidata | https://www.wikidata.org/wiki/Q25261 |
| concepts[9].display_name | Meteorology |
| concepts[10].id | https://openalex.org/C159985019 |
| concepts[10].level | 1 |
| concepts[10].score | 0.0 |
| concepts[10].wikidata | https://www.wikidata.org/wiki/Q181790 |
| concepts[10].display_name | Composite material |
| keywords[0].id | https://openalex.org/keywords/inference |
| keywords[0].score | 0.7317643165588379 |
| keywords[0].display_name | Inference |
| keywords[1].id | https://openalex.org/keywords/training |
| keywords[1].score | 0.7114962339401245 |
| keywords[1].display_name | Training (meteorology) |
| keywords[2].id | https://openalex.org/keywords/computer-science |
| keywords[2].score | 0.6462578773498535 |
| keywords[2].display_name | Computer science |
| keywords[3].id | https://openalex.org/keywords/artificial-neural-network |
| keywords[3].score | 0.5890853404998779 |
| keywords[3].display_name | Artificial neural network |
| keywords[4].id | https://openalex.org/keywords/compression |
| keywords[4].score | 0.5213414430618286 |
| keywords[4].display_name | Compression (physics) |
| keywords[5].id | https://openalex.org/keywords/artificial-intelligence |
| keywords[5].score | 0.47112107276916504 |
| keywords[5].display_name | Artificial intelligence |
| keywords[6].id | https://openalex.org/keywords/machine-learning |
| keywords[6].score | 0.38877755403518677 |
| keywords[6].display_name | Machine learning |
| keywords[7].id | https://openalex.org/keywords/geography |
| keywords[7].score | 0.040221452713012695 |
| keywords[7].display_name | Geography |
| keywords[8].id | https://openalex.org/keywords/materials-science |
| keywords[8].score | 0.03392484784126282 |
| keywords[8].display_name | Materials science |
| language | en |
| locations[0].id | pmh:oai:arXiv.org:2410.20650 |
| locations[0].is_oa | True |
| locations[0].source.id | https://openalex.org/S4306400194 |
| locations[0].source.issn | |
| locations[0].source.type | repository |
| locations[0].source.is_oa | True |
| locations[0].source.issn_l | |
| locations[0].source.is_core | False |
| locations[0].source.is_in_doaj | False |
| locations[0].source.display_name | arXiv (Cornell University) |
| locations[0].source.host_organization | https://openalex.org/I205783295 |
| locations[0].source.host_organization_name | Cornell University |
| locations[0].source.host_organization_lineage | https://openalex.org/I205783295 |
| locations[0].license | |
| locations[0].pdf_url | https://arxiv.org/pdf/2410.20650 |
| locations[0].version | submittedVersion |
| locations[0].raw_type | text |
| locations[0].license_id | |
| locations[0].is_accepted | False |
| locations[0].is_published | False |
| locations[0].raw_source_name | |
| locations[0].landing_page_url | http://arxiv.org/abs/2410.20650 |
| locations[1].id | doi:10.48550/arxiv.2410.20650 |
| 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 | |
| locations[1].version | |
| locations[1].raw_type | article |
| locations[1].license_id | |
| locations[1].is_accepted | False |
| locations[1].is_published | |
| locations[1].raw_source_name | |
| locations[1].landing_page_url | https://doi.org/10.48550/arxiv.2410.20650 |
| indexed_in | arxiv, datacite |
| authorships[0].author.id | https://openalex.org/A5111588412 |
| authorships[0].author.orcid | |
| authorships[0].author.display_name | Yongchang Hao |
| authorships[0].author_position | first |
| authorships[0].raw_author_name | Hao, Yongchang |
| authorships[0].is_corresponding | False |
| authorships[1].author.id | https://openalex.org/A5112457697 |
| authorships[1].author.orcid | |
| authorships[1].author.display_name | Yanshuai Cao |
| authorships[1].author_position | middle |
| authorships[1].raw_author_name | Cao, Yanshuai |
| authorships[1].is_corresponding | False |
| authorships[2].author.id | https://openalex.org/A5024821632 |
| authorships[2].author.orcid | https://orcid.org/0000-0001-7753-4295 |
| authorships[2].author.display_name | Lili Mou |
| authorships[2].author_position | last |
| authorships[2].raw_author_name | Mou, Lili |
| authorships[2].is_corresponding | False |
| has_content.pdf | False |
| has_content.grobid_xml | False |
| is_paratext | False |
| open_access.is_oa | True |
| open_access.oa_url | https://arxiv.org/pdf/2410.20650 |
| open_access.oa_status | green |
| open_access.any_repository_has_fulltext | False |
| created_date | 2024-11-14T00:00:00 |
| display_name | NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks |
| has_fulltext | False |
| is_retracted | False |
| updated_date | 2025-11-06T06:51:31.235846 |
| primary_topic.id | https://openalex.org/T10320 |
| primary_topic.field.id | https://openalex.org/fields/17 |
| primary_topic.field.display_name | Computer Science |
| primary_topic.score | 0.883899986743927 |
| primary_topic.domain.id | https://openalex.org/domains/3 |
| primary_topic.domain.display_name | Physical Sciences |
| primary_topic.subfield.id | https://openalex.org/subfields/1702 |
| primary_topic.subfield.display_name | Artificial Intelligence |
| primary_topic.display_name | Neural Networks and Applications |
| related_works | https://openalex.org/W230091440, https://openalex.org/W2233261550, https://openalex.org/W2810751659, https://openalex.org/W258997015, https://openalex.org/W2997094352, https://openalex.org/W3216976533, https://openalex.org/W100620283, https://openalex.org/W2495260952, https://openalex.org/W4366179611, https://openalex.org/W2996078371 |
| cited_by_count | 0 |
| locations_count | 2 |
| best_oa_location.id | pmh:oai:arXiv.org:2410.20650 |
| best_oa_location.is_oa | True |
| best_oa_location.source.id | https://openalex.org/S4306400194 |
| best_oa_location.source.issn | |
| best_oa_location.source.type | repository |
| best_oa_location.source.is_oa | True |
| best_oa_location.source.issn_l | |
| best_oa_location.source.is_core | False |
| best_oa_location.source.is_in_doaj | False |
| best_oa_location.source.display_name | arXiv (Cornell University) |
| best_oa_location.source.host_organization | https://openalex.org/I205783295 |
| best_oa_location.source.host_organization_name | Cornell University |
| best_oa_location.source.host_organization_lineage | https://openalex.org/I205783295 |
| best_oa_location.license | |
| best_oa_location.pdf_url | https://arxiv.org/pdf/2410.20650 |
| best_oa_location.version | submittedVersion |
| best_oa_location.raw_type | text |
| best_oa_location.license_id | |
| best_oa_location.is_accepted | False |
| best_oa_location.is_published | False |
| best_oa_location.raw_source_name | |
| best_oa_location.landing_page_url | http://arxiv.org/abs/2410.20650 |
| primary_location.id | pmh:oai:arXiv.org:2410.20650 |
| primary_location.is_oa | True |
| primary_location.source.id | https://openalex.org/S4306400194 |
| primary_location.source.issn | |
| primary_location.source.type | repository |
| primary_location.source.is_oa | True |
| primary_location.source.issn_l | |
| primary_location.source.is_core | False |
| primary_location.source.is_in_doaj | False |
| primary_location.source.display_name | arXiv (Cornell University) |
| primary_location.source.host_organization | https://openalex.org/I205783295 |
| primary_location.source.host_organization_name | Cornell University |
| primary_location.source.host_organization_lineage | https://openalex.org/I205783295 |
| primary_location.license | |
| primary_location.pdf_url | https://arxiv.org/pdf/2410.20650 |
| primary_location.version | submittedVersion |
| primary_location.raw_type | text |
| primary_location.license_id | |
| primary_location.is_accepted | False |
| primary_location.is_published | False |
| primary_location.raw_source_name | |
| primary_location.landing_page_url | http://arxiv.org/abs/2410.20650 |
| publication_date | 2024-10-28 |
| publication_year | 2024 |
| referenced_works_count | 0 |
| abstract_inverted_index.a | 46, 84 |
| abstract_inverted_index.8B | 86 |
| abstract_inverted_index.In | 40, 101 |
| abstract_inverted_index.by | 17, 109 |
| abstract_inverted_index.in | 58 |
| abstract_inverted_index.is | 119 |
| abstract_inverted_index.of | 2, 55, 82 |
| abstract_inverted_index.on | 52 |
| abstract_inverted_index.to | 66, 90 |
| abstract_inverted_index.we | 43, 63, 76 |
| abstract_inverted_index.Our | 117 |
| abstract_inverted_index.The | 0 |
| abstract_inverted_index.and | 24, 70 |
| abstract_inverted_index.are | 9, 15, 64 |
| abstract_inverted_index.can | 31, 105 |
| abstract_inverted_index.new | 47 |
| abstract_inverted_index.our | 103 |
| abstract_inverted_index.the | 12, 18, 33, 53, 79, 96 |
| abstract_inverted_index.31GB | 89 |
| abstract_inverted_index.With | 61 |
| abstract_inverted_index.able | 65 |
| abstract_inverted_index.code | 118 |
| abstract_inverted_index.from | 37, 88 |
| abstract_inverted_index.half | 112 |
| abstract_inverted_index.less | 91 |
| abstract_inverted_index.like | 29 |
| abstract_inverted_index.more | 7, 110 |
| abstract_inverted_index.than | 92, 111 |
| abstract_inverted_index.they | 35 |
| abstract_inverted_index.this | 41 |
| abstract_inverted_index.when | 6 |
| abstract_inverted_index.16GB, | 93 |
| abstract_inverted_index.based | 51 |
| abstract_inverted_index.fully | 99 |
| abstract_inverted_index.model | 13, 87 |
| abstract_inverted_index.sizes | 14 |
| abstract_inverted_index.usage | 108 |
| abstract_inverted_index.used. | 10 |
| abstract_inverted_index.while | 94, 113 |
| abstract_inverted_index.work, | 42 |
| abstract_inverted_index.during | 22 |
| abstract_inverted_index.memory | 21, 80, 107 |
| abstract_inverted_index.method | 104 |
| abstract_inverted_index.neural | 3, 59 |
| abstract_inverted_index.reduce | 78, 106 |
| abstract_inverted_index.scheme | 50 |
| abstract_inverted_index.suffer | 36 |
| abstract_inverted_index.weight | 48 |
| abstract_inverted_index.Llama-3 | 85 |
| abstract_inverted_index.NeuZip, | 45, 62 |
| abstract_inverted_index.achieve | 67 |
| abstract_inverted_index.entropy | 54 |
| abstract_inverted_index.keeping | 95 |
| abstract_inverted_index.numbers | 57 |
| abstract_inverted_index.without | 72 |
| abstract_inverted_index.Although | 26 |
| abstract_inverted_index.However, | 11 |
| abstract_inverted_index.Notably, | 75 |
| abstract_inverted_index.applying | 27 |
| abstract_inverted_index.dynamics | 98 |
| abstract_inverted_index.improves | 5 |
| abstract_inverted_index.networks | 4 |
| abstract_inverted_index.publicly | 120 |
| abstract_inverted_index.training | 23, 69, 83, 97 |
| abstract_inverted_index.alleviate | 32 |
| abstract_inverted_index.available | 19 |
| abstract_inverted_index.footprint | 81 |
| abstract_inverted_index.inference | 71 |
| abstract_inverted_index.introduce | 44 |
| abstract_inverted_index.networks. | 60 |
| abstract_inverted_index.on-device | 20 |
| abstract_inverted_index.available. | 121 |
| abstract_inverted_index.inference, | 102 |
| abstract_inverted_index.inference. | 25 |
| abstract_inverted_index.parameters | 8 |
| abstract_inverted_index.techniques | 28 |
| abstract_inverted_index.unchanged. | 100 |
| abstract_inverted_index.compression | 49 |
| abstract_inverted_index.constrained | 16 |
| abstract_inverted_index.constraint, | 34 |
| abstract_inverted_index.maintaining | 114 |
| abstract_inverted_index.performance | 1, 38 |
| abstract_inverted_index.sacrificing | 73 |
| abstract_inverted_index.degradation. | 39 |
| abstract_inverted_index.performance. | 74, 116 |
| abstract_inverted_index.quantization | 30 |
| abstract_inverted_index.near-lossless | 115 |
| abstract_inverted_index.significantly | 77 |
| abstract_inverted_index.floating-point | 56 |
| abstract_inverted_index.memory-efficient | 68 |
| cited_by_percentile_year | |
| countries_distinct_count | 0 |
| institutions_distinct_count | 3 |
| citation_normalized_percentile |