Resource exhausted: OOM when allocating tensor with shape on 125M with 8GB VRAM

[Fqlox]

Hi,
I pull this issues because I got a issue: I wanted to train my own model locally. but got this issues 2022-01-08 14:44:23.733680: W tensorflow/core/framework/op_kernel.cc:1692] OP_REQUIRES failed at matmul_op_impl.h:681 : Resource exhausted: OOM when allocating tensor with shape[1024,50257] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc

I used the 125M model and tried with a 30MB and 0.3MB dataset, both get the same issues.

I'm on windows 10 with a RTX2070 (8GB) with tensorflow 2.6

Thanks in advance

[darylwavy]

Looks like 8GB of VRAM is not enough for your current model settings. I ran the default code from his Google Colab on the lowest spec Colab GPU (12GB VRAM) without any issues. If your data is not sensitive, then you could give that a try.