Qwen2.5 Coder 14B Instruct
Alibaba · 2024-11-12 · 14.8B parameters
Qwen2.5-Coder-14B-Instruct is the instruction-tuned 14.7-billion-parameter model of Alibaba's Qwen2.5-Coder family, released on November 12, 2024 and post-trained for code generation, repair and reasoning. The released configuration is set to 32,768 tokens and supports extrapolation to 131,072 tokens with YaRN.
Benchmark scores
| Benchmark | Score |
|---|---|
| AIME 2024 | 10.0 |
| BigCodeBench Instruct | 48.4 |
| BigCodeBench-Hard Instruct | 22.2 |
| C-Eval | 66.2 |
| Capability | 119.5 |
| GPQA - release-reported, subset unspecified | 36.8 |
| GSM8K | 94.2 |
| HumanEval | 89.6 |
| HumanEval+ | 87.2 |
| IFEval | 66.5 |
| LiveCodeBench - July to November 2024 | 23.4 |
| LiveCodeBench - July to September 2024 | 23.4 |
| MATH | 66.8 |
| MBPP | 86.2 |
| MBPP+ | 72.8 |
| MMLU | 71.7 |
| MMLU Pro | 55.6 |
| MultiPL-E (HumanEval, 8-language average) | 79.6 |