While the Zero Redundancy Optimizer (ZeRO) excels in training large-scale models, it struggles to achieve good throughput in environments with limited band-width or small batches where communication becomes a major bottleneck. Inspired by the principles of fine-grained quantization in machine learning algorithms, we designed ZeRO++, an optimizer robust to quantization effects that allows for significant communication volume reduction using low-precision quantization techniques. ZeRO++ composes of three communication volume reduction techniques (low-precision all-gather, data remapping, and low-precision gradient averaging) to significantly reduce the communication volume up to 4x that enables up to 2.16x better throughput at 384 GPU scale. Our results also show ZeRO++ can speedup the RLHF by 3.3x compared to vanilla ZeRO. To verify the convergence of ZeRO++, we test up to 13B model for pretraining with 8\/6-bits all gather and up to 30B model for finetuning with 4-bit or 2-bit all gather, and demonstrate on-par accuracy as original ZeRO (aka standard training). As a byproduct, the model trained with ZeRO++ is weight-quantized, which can be directly used for inference without post-training quantization or quantization-aware training.<\/p>\n","protected":false},"excerpt":{"rendered":"
While the Zero Redundancy Optimizer (ZeRO) excels in training large-scale models, it struggles to achieve good throughput in environments with limited band-width or small batches where communication becomes a major bottleneck. Inspired by the principles of fine-grained quantization in machine learning algorithms, we designed ZeRO++, an optimizer robust to quantization effects that allows for significant […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"footnotes":""},"msr-content-type":[3],"msr-research-highlight":[],"research-area":[13556],"msr-publication-type":[193716],"msr-product-type":[],"msr-focus-area":[],"msr-platform":[],"msr-download-source":[],"msr-locale":[268875],"msr-field-of-study":[246691],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1083120","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-locale-en_us","msr-field-of-study-computer-science"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2024-3-9","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":1,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/dblp.org\/rec\/conf\/iclr\/WangQJWHYRR00H24.html","label_id":"243109","label":0},{"type":"url","viewUrl":"false","id":"false","title":"https:\/\/openreview.net\/forum?id=gx2BT0a9MQ","label_id":"243109","label":0}],"msr_related_uploader":"","msr_attachments":[],"msr-author-ordering":[{"type":"user_nicename","value":"Guanhua Wang","user_id":42816,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Guanhua Wang"},{"type":"user_nicename","value":"Heyang Qin","user_id":43116,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Heyang Qin"},{"type":"user_nicename","value":"Sam Ade Jacobs","user_id":43503,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Sam Ade Jacobs"},{"type":"text","value":"Xiaoxia Wu","user_id":0,"rest_url":false},{"type":"guest","value":"connor-holmes","user_id":909312,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=connor-holmes"},{"type":"user_nicename","value":"Zhewei Yao","user_id":41967,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Zhewei Yao"},{"type":"user_nicename","value":"Samyam Rajbhandari","user_id":39363,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Samyam Rajbhandari"},{"type":"user_nicename","value":"Olatunji Ruwase","user_id":33157,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Olatunji Ruwase"},{"type":"guest","value":"feng-yang","user_id":844600,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=feng-yang"},{"type":"text","value":"Lei Yang","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Yuxiong He","user_id":35084,"rest_url":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Yuxiong He"}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[1014303],"msr_group":[],"msr_project":[678390],"publication":[],"video":[],"download":[],"msr_publication_type":"inproceedings","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1083120"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":2,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1083120\/revisions"}],"predecessor-version":[{"id":1083492,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1083120\/revisions\/1083492"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1083120"}],"wp:term":[{"taxonomy":"msr-content-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-content-type?post=1083120"},{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1083120"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1083120"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1083120"},{"taxonomy":"msr-product-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-product-type?post=1083120"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1083120"},{"taxonomy":"msr-platform","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-platform?post=1083120"},{"taxonomy":"msr-download-source","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-download-source?post=1083120"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1083120"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1083120"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1083120"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1083120"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1083120"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1083120"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}