{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/accelerating-transformer-pre-training-with-2","title":"Accelerating Transformer Pre-training with 2:4 Sparsity","arxiv_id":"2404.01847","date":"2024-04-02","proceeding":null,"authors":["Yuezhou Hu","Kang Zhao","Weiyu Huang","Jianfei Chen","Jun Zhu"],"abstract":"Training large transformers is slow, but recent innovations on GPU architecture give us an advantage. NVIDIA Ampere GPUs can execute a fine-grained 2:4 sparse matrix multiplication twice as fast as its dense equivalent. In the light of this property, we comprehensively investigate the feasibility of accelerating feed-forward networks (FFNs) of transformers in pre-training. First, we define a ``flip rate'' to monitor the stability of a 2:4 training process. Utilizing this metric, we propose three techniques to preserve accuracy: to modify the sparse-refined straight-through estimator by applying the masked decay term on gradients, to determine a feasible decay factor in warm-up stage, and to enhance the model's quality by a dense fine-tuning procedure near the end of pre-training. Besides, we devise two techniques to practically accelerate training: to calculate transposable 2:4 masks by convolution, and to accelerate gated activation functions by reducing GPU L2 cache miss. Experiments show that our 2:4 sparse training algorithm achieves similar convergence to dense training algorithms on several transformer pre-training tasks, while actual acceleration can be observed on different shapes of transformer block apparently. Our toolkit is available at https://github.com/huyz2023/2by4-pretrain.","url_abs":"https://arxiv.org/abs/2404.01847v3","url_pdf":"https://arxiv.org/pdf/2404.01847v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"accelerating-transformer-pre-training-with-2","repo_url":"https://github.com/huyz2023/2by4-pretrain","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"BSD-3-Clause"}},{"paper_slug":"accelerating-transformer-pre-training-with-2","repo_url":"https://github.com/thu-ml/2by4-pretrain-acc-examples","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"tasks":[{"task_slug":null,"task_name":"GPU"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2404.01847","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2404.01847"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/huyz2023/2by4-pretrain","reach":{"status":"ok","spdx":"BSD-3-Clause"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/thu-ml/2by4-pretrain-acc-examples","reach":{"status":"ok","spdx":"BSD-3-Clause"}}],"summary":{"ran":4,"unverified":3},"by_repo_kind":{"official":{"samples":7,"ran":4,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"3d4c7fb4a6904bf4","entry":"get_higher_dtype","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/matmul.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/matmul.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"3d4c7fb4a6904bf4"}},{"code_sha256_prefix":"8c0bc604664a85d2","entry":"rand24","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/legacy.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/legacy.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"8c0bc604664a85d2"}},{"code_sha256_prefix":"4c291c5bf1d8324b","entry":"sparse12","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/legacy.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/legacy.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"4c291c5bf1d8324b"}},{"code_sha256_prefix":"7376a432c0164615","entry":"sparse24_torch","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/legacy.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/legacy.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"7376a432c0164615"}},{"code_sha256_prefix":"34eef7118623d4c6","entry":"sparse_semi_structured_from_dense","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/_semi_structured_conversions.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/_semi_structured_conversions.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"34eef7118623d4c6"}},{"code_sha256_prefix":"8e2170c920f2feab","entry":"sparse_semi_structured_to_dense","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/_semi_structured_conversions.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/_semi_structured_conversions.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"8e2170c920f2feab"}},{"code_sha256_prefix":"978da202e72c36c3","entry":"to_sparse_semi_structured","repo":"huyz2023/2by4-pretrain","repo_kind":"official","path":"sparse/semi_structured.py","file_url":"https://github.com/huyz2023/2by4-pretrain/blob/HEAD/sparse/semi_structured.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"BSD-3-Clause","inline_ok":true,"mcp_get_code":{"code_sha256":"978da202e72c36c3"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}