I am using the transformer’s trainer API to train a BART model on server. The GPU space is enough, however, the training process only runs on CPU instead of GPU. I tried to use cuda and jit from numba like this example to add function decorators, but it still doesn’t help. What is the reason of i…

Training Model on CPU instead of GPU

sgugger September 8, 2021, 12:21pm 2

The GPU will be automatically used by the Trainer, if that’s not the case, make sure you have properly installed your NVIDIA drivers and PyTorch.
Basically

import torch
torch.cuda.is_available()

should print True.

RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, &fbeta, c, CUDA_R_16F, ldc, CUDA_R_32F, CUBLAS_GEMM_DEFAULT_TENSOR_OP)

Topic		Replies	Views
Training BART Model on CPU instead of GPU 🤗Transformers	0	648	August 26, 2021
Is Transformers using GPU by default? Beginners	6	145124	December 11, 2023
How to get the Trainer API to use GPU? Beginners	0	1497	May 21, 2021
Using trainer to train a bart model on 4 gpus failed 🤗Transformers	0	334	March 16, 2022
Can I use CUDA with Trainer.train? Beginners	3	7368	May 10, 2022

Training Model on CPU instead of GPU

Related topics