Long Context Evaluation on Long Context benchmark
67.59AccuracyAdaLoRA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AdaLoRABackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.452026.04 | 67.59 | 46.73 | |
| LoRABackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.452026.04 | 67.37 | 52.1 | |
| LARSBackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.672026.04 | 67.18 | 38.33 | |
| LARS-LPBackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.672026.04 | 63.9 | 44.28 | |
| IA3Backbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.012026.04 | 62.02 | 53 | |
| AdaLoRABackbone=Llama-3.2-1B, Trainable Parameters (%)=0.452026.04 | 58.03 | 23.44 | |
| LoRABackbone=Llama-3.2-1B, Trainable Parameters (%)=0.452026.04 | 57.91 | 25.56 | |
| LARSBackbone=Llama-3.2-1B, Trainable Parameters (%)=0.672026.04 | 57.18 | 20.27 | |
| LARS-LPBackbone=Llama-3.2-1B, Trainable Parameters (%)=0.672026.04 | 54.52 | 23.19 | |
| IA3Backbone=Llama-3.2-1B, Trainable Parameters (%)=0.012026.04 | 54.13 | 26.49 | |
| PrefixBackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.032026.04 | 25.95 | 38.89 | |
| PromptBackbone=Qwen2.5-7B-Instruct, Trainable Parameters (%)=0.0032026.04 | 25.88 | 39.04 | |
| PrefixBackbone=Llama-3.2-1B, Trainable Parameters (%)=0.032026.04 | 24.62 | 22.3 | |
| PromptBackbone=Llama-3.2-1B, Trainable Parameters (%)=0.0032026.04 | 24.39 | 20.79 |