Browse › Natural Language Processing › Sentence Completion › HellaSwag
HellaSwag Benchmark (Sentence Completion)
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: Accuracy (higher is better). Points are placed at the row's paper date; 86 of 89 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||
|---|---|---|---|---|---|---|---|---|
| 1 | CompassMTL 567M with Tailor | 96.1 | – | Paper | Code | 2022 | linked, not harvested | report |
| 2 | CompassMTL 567M | 95.6 | – | Paper | Code | 2022 | linked, not harvested | report |
| 3 | DeBERTa-Large 304M (classification-based) | 95.6 | – | Paper | Code | 2022 | 1 of 7 ran · 6 unverified | report |
| 4 | GPT-4 (10-shot) | 95.3 | – | Paper | Code | 2023 | 2 of 5 ran · 3 unverified | report |
| 5 | LLaMA3+MoSLoRA | 95.0 | – | Paper | Code | 2024 | 4 of 6 ran · 2 unverified | report |
| 6 | DeBERTa-Large 304M | 94.7 | – | Paper | Code | 2022 | 1 of 7 ran · 6 unverified | report |
| 7 | LLaMA-2 13B + MixLoRA | 94.7 | – | Paper | Code | 2024 | 6 of 11 ran · 5 unverified | report |
| 8 | Unicorn 11B (fine-tuned) | 93.9 | ✓ | Paper | Code | 2021 | 0 of 5 ran · 5 unverified | report |
| 9 | LLaMA-3 8B + MixLoRA | 93.3 | – | Paper | Code | 2024 | 6 of 11 ran · 5 unverified | report |
| 10 | LLaMA-2 7B + MixLoRA | 93.1 | – | Paper | Code | 2024 | 6 of 11 ran · 5 unverified | report |
| 11 | DeBERTa++ | 93 | – | Paper | Code | 2020 | 4 of 13 ran · 9 unverified | report |
| 12 | ELECTRA-Large 335M (fine-tuned on DiscoSense and HellaSwag) | 91.5 | – | Paper | Code | 2022 | linked, not harvested | report |
| 13 | DBRX Instruct 132B (10-shot) | 89 | – | – | – | not matched | report | |
| 14 | TheBloke/llama-2-70b-Guanaco-QLoRA-fp16 (10-shot) | 88.3 | – | – | – | not matched | report | |
| 15 | ALBERT-XXL 235M | 88 | – | – | – | not matched | report | |
| 16 | PaLM 2-L (1-shot) | 87.4 | – | Paper | Code | 2023 | linked, not harvested | report |
| 17 | ELECTRA-Large 335M (fine-tuned on HellaSwag) | 86.9 | – | Paper | Code | 2022 | linked, not harvested | report |
| 18 | PaLM 2-M (1-shot) | 86.7 | – | Paper | Code | 2023 | linked, not harvested | report |
| 19 | MUPPET Roberta Large | 86.4 | – | Paper | Code | 2021 | linked, not harvested | report |
| 20 | LLaMA 65B + CFG (0-shot) | 86.3 | – | Paper | – | 2023 | no code linked | report |
| 21 | Falcon-180B (0-shot) | 85.9 | – | Paper | – | 2023 | no code linked | report |
| 22 | PaLM 2-S (1-shot) | 85.6 | – | Paper | Code | 2023 | linked, not harvested | report |
| 23 | GPT-3.5 (10-shot) | 85.5 | – | Paper | Code | 2023 | 2 of 5 ran · 3 unverified | report |
| 24 | RoBERTa-Large Ensemble | 85.5 | – | Paper | Code | 2019 | 22 of 48 ran · 26 unverified | report |
| 25 | LLaMA 30B + CFG (0-shot) | 85.3 | – | Paper | – | 2023 | no code linked | report |
| 26 | LLaMA 2 70B (0-shot) | 85.3 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 27 | HyKAS+CSKG | 85.0 | – | Paper | – | 2019 | no code linked | report |
| 28 | LLaMA 65B (0-shot) | 84.2 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 29 | PaLM-540B (Few-Shot) | 83.8 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 30 | PaLM-540B (1-shot) | 83.6 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 31 | ExDeBERTa 567M | 83.6 | – | Paper | Code | 2022 | linked, not harvested | report |
| 32 | PaLM-540B (0-shot) | 83.4 | – | Paper | Code | 2022 | 30 of 37 ran · 7 unverified | report |
| 33 | LLaMA 2 34B (0-shot) | 83.3 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 34 | Camelidae-8×34B (10-shot) | 83.2 | – | Paper | Code | 2024 | linked, not harvested | report |
| 35 | LLaMA 33B (0-shot) | 82.8 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 36 | Falcon-40B (0-shot) | 82.7 | – | Paper | – | 2023 | no code linked | report |
| 37 | Megatron-Turing NLG 530B (Few-Shot) | 82.4 | – | Paper | Code | 2022 | linked, not harvested | report |
| 38 | Qwen2idae-16x14B (10-shot) | 82.3 | – | Paper | Code | 2024 | linked, not harvested | report |
| 39 | LLaMA 13B + CFG (0-shot) | 82.1 | – | Paper | – | 2023 | no code linked | report |
| 40 | RoBERTa-Large 355M | 81.7 | – | Paper | Code | 2019 | 22 of 48 ran · 26 unverified | report |
| 41 | Mistral 7B (0-shot) | 81.3 | – | Paper | Code | 2023 | 9 of 11 ran · 2 unverified | report |
| 42 | Chinchilla 70B (0-shot) | 80.8 | – | Paper | Code | 2022 | 8 of 11 ran · 3 unverified | report |
| 43 | LLaMA 2 13B (0-shot) | 80.7 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 44 | Megatron-Turing NLG 530B (1-shot) | 80.2 | – | Paper | Code | 2022 | linked, not harvested | report |
| 45 | GPT-3 175B (few-shot, k=32) | 79.3 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 46 | Gopher 280B (0-shot) | 79.2 | – | Paper | Code | 2021 | linked, not harvested | report |
| 47 | LLaMA 13B (0-shot) | 79.2 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 48 | GPT-3 (0-shot) | 78.9 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 49 | LLaMA 2 7B (0-shot) | 77.2 | – | Paper | Code | 2023 | 31 of 52 ran · 21 unverified | report |
| 50 | Falcon-7B (0-shot) | 76.3 | – | Paper | – | 2023 | no code linked | report |
| 51 | LLaMA 7B (0-shot) | 76.1 | – | Paper | Code | 2023 | 26 of 58 ran · 32 unverified | report |
| 52 | BlooombergGPT 50B (1-shot) | 73.9 | – | Paper | Code | 2023 | linked, not harvested | report |
| 53 | OPT 66B (1-shot) | 73.5 | – | Paper | Code | 2023 | linked, not harvested | report |
| 54 | BLOOM 176B (1-shot) | 73.2 | – | Paper | Code | 2023 | linked, not harvested | report |
| 55 | Sheared-LLaMA-2.7B (50B) | 70.8 | – | Paper | Code | 2023 | 3 of 3 ran · 0 unverified | report |
| 56 | GPT-NeoX 20B (1-shot) | 68.4 | – | Paper | Code | 2023 | linked, not harvested | report |
| 57 | Open-LLaMA-3B-v2 | 67.6 | – | Paper | Code | 2023 | 3 of 3 ran · 0 unverified | report |
| 58 | Mamba-2.8B | 66.1 | – | Paper | Code | 2023 | 18 of 62 ran · 44 unverified | report |
| 59 | Sheared-LLaMA-1.3B (50B) | 60.7 | – | Paper | Code | 2023 | 3 of 3 ran · 0 unverified | report |
| 60 | FLAN 137B (3-shot) | 59.2 | – | Paper | Code | 2021 | 0 of 1 ran · 1 unverified | report |
| 61 | Mamba-1.4B | 59.1 | – | Paper | Code | 2023 | 18 of 62 ran · 44 unverified | report |
| 62 | FLAN 137B (0-shot) | 56.7 | – | Paper | Code | 2021 | 0 of 1 ran · 1 unverified | report |
| 63 | sMLP – deterministic 9.4B (0-shot) | 54.5 | – | Paper | – | 2022 | no code linked | report |
| 64 | Switch Transformer 9B | 52.5 | – | Paper | – | 2022 | no code linked | report |
| 65 | GPT-3 Large 760M (0-shot) | 51.0 | – | Paper | Code | 2020 | 15 of 65 ran · 50 unverified | report |
| 66 | GPT-2-XL 1.5B | 50.9 | – | Paper | Code | 2023 | linked, not harvested | report |
| 67 | OPT-6.7B | 50.3 | – | Paper | – | 2023 | no code linked | report |
| 68 | LLM in a Flash (OPT-6.7B with Predictor) | 49.8 | – | Paper | – | 2023 | no code linked | report |
| 69 | FLAN-T5-Large 783M | 48.7 | – | Paper | Code | 2023 | linked, not harvested | report |
| 70 | LaMini-GPT 1.5B | 48.3 | – | Paper | Code | 2023 | linked, not harvested | report |
| 71 | BERT-Large 340M | 47.3 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 72 | LaMini-F-T5 783M | 43.7 | – | Paper | Code | 2023 | linked, not harvested | report |
| 73 | GPT-1 117M | 41.7 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 74 | Flipped-3B | 41.6 | – | Paper | Code | 2022 | linked, not harvested | report |
| 75 | T0-3B (CoT fine-tuned) | 41.1 | – | Paper | Code | 2023 | linked, not harvested | report |
| 76 | LaMini-T5 738M | 40.6 | – | Paper | Code | 2023 | linked, not harvested | report |
| 77 | BERT-Base 110M | 40.5 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 78 | T5-Large 738M | 38.9 | – | Paper | Code | 2023 | linked, not harvested | report |
| 79 | Gshard 9B | 38 | – | Paper | – | 2022 | no code linked | report |
| 80 | LSTM + BERT-Base | 36.2 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 81 | RoE-3B | 34.6 | – | Paper | Code | 2023 | 3 of 3 ran · 0 unverified | report |
| 82 | ESIM + ElMo | 33.3 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 83 | HASH Layers 10B (0-shot) | 33 | – | Paper | – | 2022 | no code linked | report |
| 84 | LSTM + GloVe | 31.7 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 85 | fastText | 31.6 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 86 | LSTM + ElMo | 31.4 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
| 87 | Base Layers 10B (0-shot) | 30.2 | – | Paper | – | 2022 | no code linked | report |
| 88 | KiC-770M | 29.6 | – | Paper | – | 2022 | no code linked | report |
| 89 | Random chance baseline | 25 | – | Paper | Code | 2019 | 0 of 6 ran · 6 unverified | report |
All 89 rows shown. 86 link to a paper page on this site; 1 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 45 rows have a graph line, from 18 distinct papers; 33 rows (15 papers) have at least one sample that ran. Counting each paper once: Syntology ran 182 of 404 samples; 222 unverified. Separately, 97 of those 404 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections