【操作步骤&问题现象】
在使用Mindspore-GPU跑程序的时候出现报错：
[CRITICAL] KERNEL(1084,7f5e5ffff700,python3):2022-06-22-19:46:23.385.199 [mindspore/ccsrc/plugin/device/gpu/kernel/math/matmul_gpu_kernel.h:88] Launch] cuBLAS Error: cublasGemmStridedBatchedEx failed. Possible reasons: the GPU is occupied by other processes. | Error Number: 15 CUBLAS_STATUS_NOT_SUPPORTED: The functionality requested is not supported.
The function call stack:
In file /mnt/d/Anaconda3/envs/mindspore/lib/python3.9/site-packages/mindspore/ops/_grad/grad_math_ops.py(277)/ dw = mul2(x, dout)/
Corresponding forward node candidate:
- In file /mnt/e/课题/dockerfile/alchemy/cybertron/base.py(386)/ context = self.bmm(attention_probs, V)/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/interaction.py(378)/ v = self.multi_head_attention(/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/interaction.py(475)/ xx = self._encoder(xx, neighbours, g_ii,/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/interaction.py(475)/ xx = self._encoder(xx, neighbours, g_ii,/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/model.py(431)/ n_interaction = len(self.interactions)/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/cybertron.py(568)/ x, xlist = self.model(distances, atom_types, atom_mask,/
In file /mnt/e/课题/dockerfile/alchemy/cybertron/train.py(468)/ out = self._network(/
In file /mnt/d/Anaconda3/envs/mindspore/lib/python3.9/site-packages/mindspore/nn/wrap/cell_wrapper.py(373)/ loss = self.network(*inputs)/
【截图信息】
是cuda版本太低了，换成11.1就可以了
