CUDA Error on code snippet
Hey,
when I'm trying to run the code snippet I always run into a CUDA Error on the tagger.predict(sentence)line.
The error is:
File "/home/johannes/Projects/huggingface/spaces/borrowing-detection-es/venv/lib/python3.10/site-packages/transformers/models/bert/modeling_bert.py", line 327, in forward attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2)) RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling cublasSgemmStridedBatched( handle, opa, opb, m, n, k, &alpha, a, lda, stridea, b, ldb, strideb, &beta, c, ldc, stridec, num_batches)
Any idea why that happens? The inference API seems to run fine, so I was thinking maybe some library version problems, but my experiments didn't give me any results on it.
Hi johko!
It might be an issue coming from Flair library. Could you post the sentence you were trying to predict so I can try to replicate the error?