Qwen2.5 7B
Alibaba · 2024-09-19 · 7.6B parameters
Qwen2.5-7B is the 7B-class dense base (pretrained) checkpoint of Alibaba's Qwen2.5 series, released on September 19, 2024. It has 7,615,616,512 parameters with grouped-query attention, and Qwen documents a 131,072-token context length; Qwen2.5-7B-Instruct is its instruction-tuned counterpart.
Benchmark scores
| Benchmark | Score |
|---|---|
| ARC-Challenge | 63.7 |
| BBH (BIG-Bench Hard) | 70.4 |
| Capability | 109.8 |
| GPQA - release-reported, subset unspecified | 36.4 |
| GSM8K | 85.4 |
| HellaSwag (Unspecified Scoring Protocol) | 80.2 |
| HumanEval | 57.9 |
| HumanEval+ | 50.6 |
| MATH | 49.8 |
| MBPP | 74.9 |
| MBPP+ | 62.9 |
| MMLU | 74.2 |
| MMLU Pro | 45.0 |
| MMLU STEM | 72.3 |
| MMLU-Redux | 71.1 |
| MultiPL-E (HumanEval, 8-language average) | 50.3 |
| TheoremQA | 36.0 |
| TruthfulQA (Unspecified Metric) | 56.4 |
| WinoGrande | 75.9 |