{"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/smart-robust-and-efficient-fine-tuning-for","title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","arxiv_id":"1911.03437","date":"2019-11-08","proceeding":"ACL 2020 6","authors":["Haoming Jiang","Pengcheng He","Weizhu Chen","Xiaodong Liu","Jianfeng Gao","Tuo Zhao"],"abstract":"Transfer learning has fundamentally changed the landscape of natural language processing (NLP) research. Many existing state-of-the-art models are first pre-trained on a large text corpus and then fine-tuned on downstream tasks. However, due to limited data resources from downstream tasks and the extremely large capacity of pre-trained models, aggressive fine-tuning often causes the adapted model to overfit the data of downstream tasks and forget the knowledge of the pre-trained model. To address the above issue in a more principled manner, we propose a new computational framework for robust and efficient fine-tuning for pre-trained language models. Specifically, our proposed framework contains two important ingredients: 1. Smoothness-inducing regularization, which effectively manages the capacity of the model; 2. Bregman proximal point optimization, which is a class of trust-region methods and can prevent knowledge forgetting. Our experiments demonstrate that our proposed method achieves the state-of-the-art performance on multiple NLP benchmarks.","url_abs":"https://arxiv.org/abs/1911.03437v5","url_pdf":"https://arxiv.org/pdf/1911.03437v5.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":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/namisan/mt-dnn","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/chunhuililili/mt_dnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/cliang1453/camero","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/microsoft/MT-DNN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/archinetai/smart-pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null},{"paper_slug":"smart-robust-and-efficient-fine-tuning-for","repo_url":"https://github.com/archinetai/vat-pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"linguistic-acceptability","task_name":"Linguistic Acceptability"},{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"natural-language-understanding","task_name":"Natural Language Understanding"},{"task_slug":"paraphrase-identification","task_name":"Paraphrase Identification"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/natural-language-inference-on-ax","task":"Natural Language Inference","dataset":"AX","model":"T5","rank_in_archive_order":1,"of":1,"metrics":{"Accuracy":"53.1"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-mnli-snli-anli","task":"Natural Language Inference","dataset":"MNLI + SNLI + ANLI + FEVER","model":"SMARTRoBERTa-LARGE","rank_in_archive_order":1,"of":1,"metrics":{"% Dev Accuracy":"57.1","% Test Accuracy":"57.1"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"T5","rank_in_archive_order":3,"of":67,"metrics":{"Matched":"92.0","Mismatched":"91.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"MT-DNN-SMARTv0","rank_in_archive_order":62,"of":67,"metrics":{"Accuracy":"85.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"MT-DNN-SMART","rank_in_archive_order":63,"of":67,"metrics":{"Accuracy":"85.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"SMART+BERT-BASE","rank_in_archive_order":64,"of":67,"metrics":{"Accuracy":"85.6"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"SMARTRoBERTa","rank_in_archive_order":66,"of":67,"metrics":{"Dev Matched":"91.1","Dev Mismatched":"91.3"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"SMART-BERT","rank_in_archive_order":67,"of":67,"metrics":{"Dev Matched":"85.6","Dev Mismatched":"86.0"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"ALICE","rank_in_archive_order":3,"of":43,"metrics":{"Accuracy":"99.2%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"MT-DNN-SMART","rank_in_archive_order":4,"of":43,"metrics":{"Accuracy":"99.2%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"SMART-BERT","rank_in_archive_order":42,"of":43,"metrics":{},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"SMARTRoBERTa","rank_in_archive_order":43,"of":43,"metrics":{},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"T5-XXL 11B","rank_in_archive_order":8,"of":90,"metrics":{"Accuracy":"92.5%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"SMARTRoBERTa","rank_in_archive_order":12,"of":90,"metrics":{"Accuracy":"92.0%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"SMART-BERT","rank_in_archive_order":50,"of":90,"metrics":{"Accuracy":"71.2%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"SMART","rank_in_archive_order":51,"of":90,"metrics":{"Accuracy":"71.2%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"MT-DNN-SMARTLARGEv0","rank_in_archive_order":9,"of":98,"metrics":{"% Dev Accuracy":"92.6","% Test Accuracy":"91.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"MT-DNN-SMART_100%ofTrainingData","rank_in_archive_order":94,"of":98,"metrics":{"Dev Accuracy":"91.6"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"MT-DNN-SMART_10%ofTrainingData","rank_in_archive_order":95,"of":98,"metrics":{"Dev Accuracy":"88.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"MT-DNN-SMART_1%ofTrainingData","rank_in_archive_order":96,"of":98,"metrics":{"Dev Accuracy":"86"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-snli","task":"Natural Language Inference","dataset":"SNLI","model":"MT-DNN-SMART_0.1%ofTrainingData","rank_in_archive_order":97,"of":98,"metrics":{"Dev Accuracy":"82.7"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-scitail","task":"Natural Language Inference","dataset":"SciTail","model":"MT-DNN-SMART_100%ofTrainingData","rank_in_archive_order":9,"of":13,"metrics":{"Dev Accuracy":"96.1"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-scitail","task":"Natural Language Inference","dataset":"SciTail","model":"MT-DNN-SMART_10%ofTrainingData","rank_in_archive_order":10,"of":13,"metrics":{"Dev Accuracy":"91.3"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-scitail","task":"Natural Language Inference","dataset":"SciTail","model":"MT-DNN-SMART_1%ofTrainingData","rank_in_archive_order":11,"of":13,"metrics":{"Dev Accuracy":"88.6"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-scitail","task":"Natural Language Inference","dataset":"SciTail","model":"MT-DNN-SMART_0.1%ofTrainingData","rank_in_archive_order":12,"of":13,"metrics":{"Dev Accuracy":"82.3"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-scitail","task":"Natural Language Inference","dataset":"SciTail","model":"MT-DNN-SMARTLARGEv0","rank_in_archive_order":13,"of":13,"metrics":{"% Dev Accuracy":"96.6","% Test Accuracy":"95.2"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-understanding-on-glue","task":"Natural Language Understanding","dataset":"GLUE","model":"MT-DNN-SMART","rank_in_archive_order":1,"of":2,"metrics":{"Average":"89.9"},"uses_additional_data":false},{"leaderboard":"/sota/paraphrase-identification-on-quora-question","task":"Paraphrase Identification","dataset":"Quora Question Pairs","model":"ALICE","rank_in_archive_order":1,"of":31,"metrics":{"F1":"90.7"},"uses_additional_data":false},{"leaderboard":"/sota/paraphrase-identification-on-quora-question","task":"Paraphrase Identification","dataset":"Quora Question Pairs","model":"FreeLB","rank_in_archive_order":28,"of":31,"metrics":{"Accuracy":"74.8","Dev Accuracy":"92.6"},"uses_additional_data":false},{"leaderboard":"/sota/paraphrase-identification-on-quora-question","task":"Paraphrase Identification","dataset":"Quora Question Pairs","model":"SMART-BERT","rank_in_archive_order":30,"of":31,"metrics":{"Dev Accuracy":"91.5","Dev F1":"88.5"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-mrpc","task":"Semantic Textual Similarity","dataset":"MRPC","model":"MT-DNN-SMART","rank_in_archive_order":1,"of":45,"metrics":{"Accuracy":"93.7%","F1":"91.7"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-mrpc","task":"Semantic Textual Similarity","dataset":"MRPC","model":"SMART","rank_in_archive_order":6,"of":45,"metrics":{"Accuracy":"91.3%"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-mrpc","task":"Semantic Textual Similarity","dataset":"MRPC","model":"SMART-BERT","rank_in_archive_order":44,"of":45,"metrics":{},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-mrpc","task":"Semantic Textual Similarity","dataset":"MRPC","model":"SMARTRoBERTa","rank_in_archive_order":45,"of":45,"metrics":{},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-sts-benchmark","task":"Semantic Textual Similarity","dataset":"STS Benchmark","model":"MT-DNN-SMART","rank_in_archive_order":1,"of":66,"metrics":{"Pearson Correlation":"0.929","Spearman Correlation":"0.925"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-sts-benchmark","task":"Semantic Textual Similarity","dataset":"STS Benchmark","model":"SMARTRoBERTa","rank_in_archive_order":65,"of":66,"metrics":{"Dev Pearson Correlation":"92.8","Dev Spearman Correlation":"92.6"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-sts-benchmark","task":"Semantic Textual Similarity","dataset":"STS Benchmark","model":"SMART-BERT","rank_in_archive_order":66,"of":66,"metrics":{"Dev Pearson Correlation":"90.0","Dev Spearman Correlation":"89.4"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"MT-DNN-SMART","rank_in_archive_order":2,"of":87,"metrics":{"Accuracy":"97.5"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"MT-DNN","rank_in_archive_order":39,"of":87,"metrics":{"Accuracy":"93.6"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"SMART+BERT-BASE","rank_in_archive_order":45,"of":87,"metrics":{"Accuracy":"93"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"SMARTRoBERTa","rank_in_archive_order":84,"of":87,"metrics":{"Dev Accuracy":"96.9"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"SMART-MT-DNN","rank_in_archive_order":85,"of":87,"metrics":{"Dev Accuracy":"96.1"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"SMART-BERT","rank_in_archive_order":86,"of":87,"metrics":{"Dev Accuracy":"93.0"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1911.03437","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1911.03437"}},"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/namisan/mt-dnn","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/archinetai/vat-pytorch","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/archinetai/smart-pytorch","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/cliang1453/camero","reach":{"status":"ok"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/microsoft/MT-DNN","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/chunhuililili/mt_dnn","reach":null}],"summary":{"ran_honours":2,"ran_violates":1,"ran_fixture":3,"unverified":2},"by_repo_kind":{"listed":{"samples":7,"ran":5,"repositories":2}},"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":1,"samples":[{"code_sha256_prefix":"c3932e0eff6d7c92","entry":"default","repo":"archinetai/smart-pytorch","repo_kind":"listed","path":"smart_pytorch/smart_pytorch.py","file_url":"https://github.com/archinetai/smart-pytorch/blob/HEAD/smart_pytorch/smart_pytorch.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c3932e0eff6d7c92"}},{"code_sha256_prefix":"aa5486a3650902d8","entry":"exists","repo":null,"repo_kind":null,"path":null,"file_url":null,"link_basis":"identical_code_first_harvested_elsewhere","language":"python","status":"ran_violates","verification_level":1,"contract_check":"VIOLATES","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":null,"inline_ok":false,"mcp_get_code":{"code_sha256":"aa5486a3650902d8"}},{"code_sha256_prefix":"43812e8f014de7b9","entry":"inf_norm","repo":"archinetai/smart-pytorch","repo_kind":"listed","path":"smart_pytorch/smart_pytorch.py","file_url":"https://github.com/archinetai/smart-pytorch/blob/HEAD/smart_pytorch/smart_pytorch.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"43812e8f014de7b9"}},{"code_sha256_prefix":"acf00a3e56df0585","entry":"js_loss","repo":"archinetai/smart-pytorch","repo_kind":"listed","path":"smart_pytorch/loss.py","file_url":"https://github.com/archinetai/smart-pytorch/blob/HEAD/smart_pytorch/loss.py","link_basis":"first_harvest_node","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"acf00a3e56df0585"}},{"code_sha256_prefix":"9f77e2bc0dc49ca0","entry":"kl_loss","repo":"archinetai/smart-pytorch","repo_kind":"listed","path":"smart_pytorch/loss.py","file_url":"https://github.com/archinetai/smart-pytorch/blob/HEAD/smart_pytorch/loss.py","link_basis":"first_harvest_node","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"9f77e2bc0dc49ca0"}},{"code_sha256_prefix":"e84d702dcea1d1de","entry":"sym_kl_loss","repo":"archinetai/smart-pytorch","repo_kind":"listed","path":"smart_pytorch/loss.py","file_url":"https://github.com/archinetai/smart-pytorch/blob/HEAD/smart_pytorch/loss.py","link_basis":"first_harvest_node","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"e84d702dcea1d1de"}},{"code_sha256_prefix":"c1c06b9077cc6e02","entry":"default","repo":"archinetai/vat-pytorch","repo_kind":"listed","path":"vat_pytorch/utils.py","file_url":"https://github.com/archinetai/vat-pytorch/blob/HEAD/vat_pytorch/utils.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":"c1c06b9077cc6e02"}},{"code_sha256_prefix":"3c4599eeaaa1d994","entry":"exists","repo":"archinetai/vat-pytorch","repo_kind":"listed","path":"vat_pytorch/utils.py","file_url":"https://github.com/archinetai/vat-pytorch/blob/HEAD/vat_pytorch/utils.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":"3c4599eeaaa1d994"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}