Instruction Following on X-AlpacaEval
70.9Chinese Win RatePLUG
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PLUGFoundation LLM=LLaMA-2-13B, Comparison Baseline=Pivot-Only2023.11 | 70.9 | 19.1 | 51.8 | 76.5 | 12.7 | 63.9 | 67.6 | 17.8 | 49.8 | 64 | 20.9 | 43.1 | |
| PLUGFoundation LLM=LLaMA-2-13B, Comparison Baseline=Mono. Response2023.11 | 58 | 25.2 | 32.8 | 64.1 | 19.9 | 44.2 | 50.3 | 25.8 | 24.5 | 53 | 27.6 | 25.5 | |
| PLUGFoundation LLM=PolyLM-13B, Comparison Baseline=Pivot-Only2023.11 | 53.2 | 32.3 | 20.9 | 79.9 | 11.1 | 68.8 | 65.7 | 18.5 | 47.2 | 57.4 | 24.1 | 33.3 | |
| PLUGFoundation LLM=LLaMA-2-13B, Comparison Baseline=Mono. + Translation2023.11 | 53 | 28 | 25.1 | 62.7 | 20.1 | 42.6 | 50.1 | 26.6 | 23.5 | 51.3 | 25.6 | 25.7 | |
| PLUGFoundation LLM=PolyLM-Instruct-13B, Comparison Baseline=Pivot-Only2023.11 | 52.8 | 31.9 | 20.9 | 77.1 | 12.9 | 64.2 | 62 | 20.1 | 41.9 | 56.7 | 26.3 | 30.4 | |
| PLUGFoundation LLM=LLaMA-2-13B, Comparison Baseline=Mono. + Code-Switch2023.11 | 50.2 | 31.6 | 18.6 | 55.2 | 25.6 | 29.6 | 46.2 | 30.9 | 15.3 | 48.4 | 29.9 | 18.5 | |
| PLUGFoundation LLM=PolyLM-Instruct-13B, Comparison Baseline=Mono. Response2023.11 | 48.5 | 32.1 | 16.4 | 64.5 | 19 | 45.5 | 54.2 | 22.9 | 31.3 | 44.8 | 32.1 | 12.7 | |
| PLUGFoundation LLM=PolyLM-13B, Comparison Baseline=Mono. + Translation2023.11 | 47 | 34.3 | 12.7 | 67.3 | 20.9 | 46.5 | 51.9 | 27.5 | 24.5 | 50.2 | 31.2 | 19 | |
| PLUGFoundation LLM=PolyLM-13B, Comparison Baseline=Mono. + Code-Switch2023.11 | 47 | 37.8 | 11.2 | 57.5 | 25.1 | 32.4 | 48.8 | 29.4 | 19.4 | 45.8 | 34 | 11.8 | |
| PLUGFoundation LLM=PolyLM-Instruct-13B, Comparison Baseline=Mono. + Translation2023.11 | 46.8 | 33.5 | 13.3 | 65 | 21.8 | 43.3 | 51.1 | 29 | 22.1 | 48.3 | 32.6 | 15.7 | |
| PLUGFoundation LLM=PolyLM-Instruct-13B, Comparison Baseline=Mono. + Code-Switch2023.11 | 46.1 | 32.8 | 13.3 | 57.8 | 23.9 | 33.9 | 49.6 | 29.8 | 19.8 | 45.5 | 32.9 | 12.5 | |
| PLUGFoundation LLM=PolyLM-13B, Comparison Baseline=Mono. Response2023.11 | 45.5 | 34.5 | 10.9 | 67.3 | 18.4 | 48.9 | 59.3 | 22.1 | 37.1 | 44.5 | 30.7 | 13.8 |