#6015 ASR Error [Qwen-ASR(Local)] CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasGemmEx( handle, opa, opb, m,

202.150* Posted at: 11 hours ago

ASR Error [Qwen-ASR(Local)] CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasGemmEx( handle, opa, opb, m, n, k, alpha_ptr, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, beta_ptr, c, CUDA_R_16F, ldc, compute_type, CUBLAS_GEMM_DEFAULT_TENSOR_OP)
Traceback (most recent call last):
File "videotrans\task\job.py", line 106, in run
File "videotrans\task\_speech2text.py", line 153, in recogn
File "videotrans\recognition\__init__.py", line 240, in run
File "videotrans\recognition\_base.py", line 142, in run
File "videotrans\recognition\_qwenasrlocal.py", line 52, in _exec
File "videotrans\configure\_base.py", line 290, in _new_process
RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasGemmEx( handle, opa, opb, m, n, k, alpha_ptr, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, beta_ptr, c, CUDA_R_16F, ldc, compute_type, CUBLAS_GEMM_DEFAULT_TENSOR_OP)
TaskCfgSTT(is_cuda=True, uuid='a555d73470', cache_folder='E:/ytrans/tmp/2356/a555d73470', target_dir='E:/ytrans/output/recogn', source_language=None, source_language_code=None, source_sub=None, source_wav=None, source_wav_output=None, target_language=None, target_language_code=None, target_sub='E:/ytrans/output/recogn/11-1. About the Sketch.srt', target_wav=None, target_wav_output=None, name='E:/PH/11-1. About the Sketch.m4a', noextname='11-1. About the Sketch', basename='11-1. About the Sketch.m4a', ext='m4a', dirname='E:/PH', shound_del_name=None, detect_language='ja', recogn_type=2, model_name='0.6B', shibie_audio='E:/ytrans/tmp/2356/a555d73470/11-1. About the Sketch-1784705507.074377.wav', remove_noise=False, enable_diariz=False, nums_diariz=0, rephrase=1, fix_punc=False)

=
system:Windows-10-10.0.26200-SP0
version:v3.97
frozen:True
language:en
root_dir:E:/ytrans

Floor #1 DeepSeek reply at 11 hours ago

This error is caused by a CUDA cuBLAS execution failure, typically due to one of the following: incompatible CUDA Toolkit / cuDNN versions, outdated NVIDIA drivers, insufficient GPU VRAM, or GPU instability during the Qwen-ASR local model inference.

Follow these steps to resolve:

Check CUDA and cuDNN versions – pyVideoTrans requires CUDA Toolkit ≥ 12.8 and cuDNN ≥ 9.11.
- Open Command Prompt and run nvcc -V to verify CUDA version.
- Check that cuDNN DLLs are correctly placed in the CUDA bin and lib directories and accessible in the system PATH.
Update NVIDIA drivers – Ensure your GPU driver is up-to-date (driver version supporting CUDA 12.8+).
Try disabling CUDA acceleration – As a temporary workaround, uncheck the "CUDA acceleration" checkbox in the main interface before starting the task. This forces CPU processing and will confirm whether the problem is CUDA-related.
Reduce model size or precision – If you need CUDA, try selecting the smaller 0.6B model (already selected) or switch to a lower-precision data type:
- Go to Menu → Tools → Advanced Options → Speech Recognition Parameters
- Set Compute type to float16 (or even int8) to lower VRAM usage.
Close other GPU-intensive applications – Ensure no other programs (e.g., other AI tools, games) are consuming GPU memory.
Verify model integrity – The Qwen-ASR model might be corrupted or incomplete. Delete the model folder at models/models--Qwen--Qwen3-ASR-0.6B and let the software re-download it automatically.

#6015 ASR Error [Qwen-ASR(Local)] CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasGemmEx( handle, opa, opb, m,

1 Replies

Please refer to the documentation:

Post Your Reply

Related resource