Gemma 2B.
Google DeepMind Feb 2024 Open weights
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MuSR | 39.8% | #59 of 98 | Official leaderboard | |
| OpenBookQA | 47.8% | #19 of 24 | Lab model card | |
| SIQA | 49.7% | #19 of 24 | Lab model card | |
| TruthfulQA | 44.8% | #25 of 30 | Lab model card | |
| HellaSwag | 71.4% | #47 of 53 | Lab model card | |
| AlpacaEval 2.0 | 5.44% | #43 of 47 | Official leaderboard | |
| PIQA | 77.3% | #34 of 37 | Lab model card | |
| TriviaQA | 53.2% | #37 of 40 | Lab model card | |
| WinoGrande | 65.4% | #44 of 47 | Lab model card | |
| MMLU-Pro | 15.8% | #200 of 212 | Official leaderboard | |
| BIG-Bench Hard | 35.2% | #69 of 73 | Lab model card | |
| Open Arabic LLM Leaderboard v1 | 36% | #52 of 55 | Official leaderboard | |
| CommonSenseQA | 65.3% | #18 of 19 | Lab model card | |
| Natural Questions | 12.5% | #19 of 20 | Lab model card | |
| ARC-Easy | 73.2% | #22 of 23 | Lab model card | |
| ARC-Challenge | 42.1% | #47 of 49 | Lab model card | |
| BoolQ | 69.4% | #27 of 28 | Lab model card | |
| LMArena (Elo) | 1,094 | #325 of 335 | Official leaderboard | |
| MMLU | 42.3% | #129 of 131 | Lab model card | |
| EQ-Bench v2 | 23.3 | #84 of 85 | Official leaderboard | |
| WildBench | -5.25 | #47 of 47 | Official leaderboard | |
| AGIEval | 24.2% | #40 of 40 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HumanEval+ | 20.7% | #43 of 46 | Official leaderboard | |
| HumanEval | 22% | #98 of 101 | Lab model card | |
| MBPP+ | 34.1% | #34 of 35 | Official leaderboard | |
| MBPP | 29.2% | #70 of 71 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ToxiGen | 29.8 | #1 of 4 | Lab model card | |
| CrowS-Pairs | 45.8% | #2 of 4 | Lab model card | |
| CyberMetric | 19.2% | #9 of 10 | Official leaderboard | |
| Winogender | 51.3% | #4 of 4 | Lab model card | |
| BOLD | 45.6 | #2 of 2 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Open LLM Leaderboard v2 | 7.32 | #91 of 99 | Official leaderboard |
The Self-Reported Index fits one score to the 19 capability results Google DeepMind reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 86.1 to 90.9. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Every result in the longer documents the lab published with it. 26 of them aren't in the launch post.