I am trying to use the accelerate library with transformers.Trainer. However, I always get an error message when using multiple GPUs. Here is the minimum version of my train.py file: import os import hydra from omegaconf import DictConfig from peft import LoraConfig, get_peft_model, TaskType from …

Accelerator() causes Error

cemde April 12, 2024, 4:57pm 3

Thank you. It works now. Any idea why this error occurs?

Topic		Replies	Views
Code terminates without training while using accelerate 🤗Accelerate	3	183	April 13, 2024
Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! 🤗Accelerate	1	892	September 20, 2023
Struggle with training on TPU using 'accelerate' library 🤗Accelerate	3	1716	March 7, 2022
How to properly wrap a model for training with accelerate? 🤗Accelerate	1	1318	September 20, 2023
Can't pickle error using accelerate multi-GPU 🤗Accelerate	6	10082	March 7, 2023