### Repro On H100 run: `pytest test/prototype/moe_training/test_scaled_grouped_mm.py -v -rs` ### Error `RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling ::cublasLtMatmulDescSetAttribute(descriptor(), attr, &value, sizeof(value))`
Repro
On H100 run:
pytest test/prototype/moe_training/test_scaled_grouped_mm.py -v -rsError
RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling ::cublasLtMatmulDescSetAttribute(descriptor(), attr, &value, sizeof(value))