Сирек қайта даярлау
method
ID: sparsefinetuning
Related Papers
Composable Sparse Fine-Tuning for Cross-Lingual Transfer
AbstractFine-tuning the entire set of parameters of a large pretrained model has become the mainstream approach for transfer learning. To increase its efficiency and prevent catastrophic forgetting and interference, techniques …...
Parameter-Efficient Fine-Tuning without Introducing New Latency
AbstractParameter-efficient fine-tuning (PEFT) of pre-trained language models has recently demonstrated remarkable achievements, effectively matching the performance of full fine-tuning while utilizing significantly fewer trainable parameters, and consequently addressing the storage …...