You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
** On entry to HgemmStridedBatched parameter number 10 had an illegal value
cuBLAS API failed with status 7
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/alyssa/anaconda3/envs/lm_fun/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/home/alyssa/anaconda3/envs/lm_fun/lib/python3.10/site-packages/cublas_ops/__init__.py", line 39, in cublas_half_matmul_batched_simple
return _cublas_hgemm_batched_simple(a, b)
RuntimeError: cuBLAS API failed
but this code works:
>>> y = torch.ones([1, 8192, 28672], dtype=torch.float16, device="cuda:0")
>>> z = cublas_ops.cublas_half_matmul_batched_simple(x, y)
The text was updated successfully, but these errors were encountered:
Cool idea! Proud to submit a first bug report :)
This PyTorch code (Ubuntu, CUDA 12.1, Torch 2.2.2, Nvidia 4090):
fails with this stack trace:
but this code works:
The text was updated successfully, but these errors were encountered: