Benchmark
HELM Lite: WMT 2014 - BLEU-4
Multilingual benchmarks compare model performance across languages. Scores are percentages.
All multilingual benchmarksScores from Stanford HELM Lite data.
- Models measured
- 69
- Scores recorded
- 70
- Sources
- 1
- IQ weight
- 1.81
Score distribution
- 18.5%
- Mean
- 19.2%
- Median
- 4.2%
- Std dev
- 14.2%
- 10th percentile
- 22.5%
- 90th percentile
- 2.3%
- Min
This benchmark has a weight of 1.81 for chat models, measured over 69 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 51 | Yi-34B· Stanford HELM Lite data | 17.2% |
| 52 | Phi-3 Medium 4K Instruct· Stanford HELM Lite data | 17% |
| 53 | Mistral Small· Stanford HELM Lite data | 16.9% |
| 54 | Llama 2 13b· Stanford HELM Lite data | 16.7% |
| 55 | Jamba Instruct· Stanford HELM Lite data | 16.4% |
| 56 | Mistral 7B v0.1· Stanford HELM Lite data | 16% |
| 57 | Qwen2.5 7B Instruct Turbo· Stanford HELM Lite data | 15.5% |
| 58 | Phi-3 Small 8K Instruct· Stanford HELM Lite data | 15.4% |
| 59 | Qwen1.5-7B· Stanford HELM Lite data | 15.3% |
| 60 | Command R+· Stanford HELM Lite data | 14.9% |
| 61 | Claude 3 Haiku· Stanford HELM Lite data | 14.8% |
| 62 | Llama 2 7b· Stanford HELM Lite data | 14.4% |
| 63 | Mistral (7B) Instruct v0.3· Stanford HELM Lite data | 14.2% |
| 64 | Claude Haiku 3.5· Stanford HELM Lite data | 13.5% |
| 65 | DBRX-instruct· Stanford HELM Lite data | 13.1% |
| 66 | Yi-6B· Stanford HELM Lite data | 11.7% |
| 67 | OLMo-7B· Stanford HELM Lite data | 9.7% |
| 68 | command· Stanford HELM Lite data | 8.8% |
| 69 | phi-2· Stanford HELM Lite data | 3.8% |
| 70 | Command Light· Stanford HELM Lite data | 2.3% |
Showing 51–70 of 70 results.
Data credits: Stanford HELM (Stanford CRFM), Apache-2.0 (licence)