DistilBERT 66M

Reported on 5 benchmarks across 4 tasks · 1 paper

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Natural Language Processing5 results

Question AnsweringonSQuAD1.1 dev
F1· 2019-10-02
85.8
best: 95.77 (XLNet+DSC)
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter arXiv:1910.01108
Natural Language InferenceonWNLI
Accuracy· 2019-10-02
44.4
best: 95.9 (Turing NLR v5 XXL 5.4B (fine-tuned))
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter arXiv:1910.01108
Semantic Textual SimilarityonSTS Benchmark
Pearson Correlation· 2019-10-02
0.907
best: 0.929 (MT-DNN-SMART)
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter arXiv:1910.01108
Sentiment AnalysisonSST-2 Binary classification
Accuracy· 2019-10-02
91.3
best: 97.5 (T5-11B)
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter arXiv:1910.01108
Sentiment AnalysisonIMDb
Accuracy· 2019-10-02
92.82
best: 96.68 (RoBERTa-large with LlamBERT)
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter arXiv:1910.01108