Downstream Task
Benchmarks
TaskDatasetSOTA resultTrendResults
Downstream Task Llama-2-7B to Llama-2-13B
80Relative Performance vs Retrained LoRA
2
Downstream Task GPT-2 to GPT-2-Medium
90Relative Performance vs Retrained LoRA
2
Downstream Task Llama-2-7B to Mistral-7B
85Relative Performance vs Retrained LoRA
2