{"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/unit-scaling-out-of-the-box-low-precision","title":"Unit Scaling: Out-of-the-Box Low-Precision Training","arxiv_id":"2303.11257","date":"2023-03-20","proceeding":null,"authors":["Charlie Blake","Douglas Orr","Carlo Luschi"],"abstract":"We present unit scaling, a paradigm for designing deep learning models that simplifies the use of low-precision number formats. Training in FP16 or the recently proposed FP8 formats offers substantial efficiency gains, but can lack sufficient range for out-of-the-box training. Unit scaling addresses this by introducing a principled approach to model numerics: seeking unit variance of all weights, activations and gradients at initialisation. Unlike alternative methods, this approach neither requires multiple training runs to find a suitable scale nor has significant computational overhead. We demonstrate the efficacy of unit scaling across a range of models and optimisers. We further show that existing models can be adapted to be unit-scaled, training BERT-Large in FP16 and then FP8 with no degradation in accuracy.","url_abs":"https://arxiv.org/abs/2303.11257v2","url_pdf":"https://arxiv.org/pdf/2303.11257v2.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":"unit-scaling-out-of-the-box-low-precision","repo_url":"https://github.com/graphcore-research/unit-scaling-demo","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"unit-scaling-out-of-the-box-low-precision","repo_url":"https://github.com/graphcore-research/unit-scaling","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2303.11257","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2303.11257"}},"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/graphcore-research/unit-scaling","reach":{"status":"ok","spdx":"Apache-2.0"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/graphcore-research/unit-scaling-demo","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"unverified":14},"by_repo_kind":{"official":{"samples":14,"ran":0,"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":"2f8c9dc17d697ee0","entry":"batched_gather","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/layers.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/layers.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2f8c9dc17d697ee0"}},{"code_sha256_prefix":"b910812684c6f413","entry":"context","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/pedal/xpu.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/pedal/xpu.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"b910812684c6f413"}},{"code_sha256_prefix":"922e7a65963e6b06","entry":"conv1d","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/uscale/ops.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/uscale/ops.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"922e7a65963e6b06"}},{"code_sha256_prefix":"05df62f6648200c1","entry":"load_character","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/datasets.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/datasets.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"05df62f6648200c1"}},{"code_sha256_prefix":"73c30a2cc2fee462","entry":"pointwise","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/uscale/ops.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/uscale/ops.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"73c30a2cc2fee462"}},{"code_sha256_prefix":"8cae868db43ba55a","entry":"printing","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/uscale/utility.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/uscale/utility.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"8cae868db43ba55a"}},{"code_sha256_prefix":"f525cde07e4a95c4","entry":"remove_keys","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/pedal/utility.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/pedal/utility.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f525cde07e4a95c4"}},{"code_sha256_prefix":"55834bdb013ab248","entry":"scaling","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/uscale/ops.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/uscale/ops.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"55834bdb013ab248"}},{"code_sha256_prefix":"288d2d846f7173a7","entry":"sinusoid_embedding","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/layers.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/layers.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"288d2d846f7173a7"}},{"code_sha256_prefix":"d380f15ca83de673","entry":"softmax_cross_entropy","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/layers.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/layers.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"d380f15ca83de673"}},{"code_sha256_prefix":"9e76a013a5849311","entry":"split_seed","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/pedal/utility.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/pedal/utility.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"9e76a013a5849311"}},{"code_sha256_prefix":"c273c7f894008b87","entry":"to_ids","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/datasets.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/datasets.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c273c7f894008b87"}},{"code_sha256_prefix":"7b5e4d5d64c189c4","entry":"to_jsonable","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/pedal/utility.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/pedal/utility.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"7b5e4d5d64c189c4"}},{"code_sha256_prefix":"e30037d5501d3b17","entry":"to_str","repo":"graphcore-research/unit-scaling-demo","repo_kind":"official","path":"scmm/datasets.py","file_url":"https://github.com/graphcore-research/unit-scaling-demo/blob/HEAD/scmm/datasets.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"e30037d5501d3b17"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}