#7643 ASR Error [Qwen-ASR(Local)] CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasGemmEx( handle, opa, opb, m,

202.150**5 Posted at: 1 hour 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='e3dc4414b7', cache_folder='E:/ytrans/tmp/20092/e3dc4414b7', 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/20 Series Illustration 03 Color Placement.srt', target_wav=None, target_wav_output=None, name='E:/2023/personal/inspo2/ZQ/updated subs/20 Series Illustration 03 Color Placement.m4a', noextname='20 Series Illustration 03 Color Placement', basename='20 Series Illustration 03 Color Placement.m4a', ext='m4a', dirname='E:/2023/personal/inspo2/ZQ/updated subs', shound_del_name=None, detect_language='ko', recogn_type=2, model_name='0.6B', shibie_audio='E:/ytrans/tmp/20092/e3dc4414b7/20 Series Illustration 03 Color Placement-1790485988.9937654.wav', remove_noise=False, enable_diariz=False, nums_diariz=0, rephrase=0, fix_punc=False)


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

1 Replies

#1 DeepSeek reply at 1 hour ago

The underlying cause is a GPU-side computation failure: your GPU/CUDA environment could not complete the FP16 (half-precision) matrix multiplication required by the Qwen-ASR local model.

This is typically caused by one of the following:

  1. Insufficient VRAM — the model tried to run in FP16, but free VRAM was not enough.
  2. CUDA/cuDNN environment mismatch — the bundled CUDA runtime does not match your driver/cuDNN version.
  3. Driver too old for the CUDA 12.8 runtime used by the software.
  4. Software version too old — you are on v3.97, which is far behind the current v4.14.

Fix steps, in order:

  1. Fall back to a lighter ASR channel first (fastest workaround). Switch the Speech Recognition channel to faster-whisper(内置) with model large-v3-turbo or medium, which is the default recommended channel and much more stable.
  2. If you must use Qwen-ASR, switch to the smaller 0.6B model (you already have 0.6B selected, so instead try disabling CUDA acceleration once to confirm whether it is a GPU-specific issue).
  3. Update your NVIDIA driver to the latest version, and verify:

    • nvcc -V shows CUDA 12.8+
    • cuDNN 9.11+ is installed, with its bin and lib folders added to the system PATH
  4. Update the software. v3.97 is outdated.

    Download the latest full package plus patch from the official site and overwrite your directory (files such as params.json / cfg.json will be preserved).
  5. Reduce GPU load: in Menu → Tools/Options → Advanced Options → General, set GPU同时任务数[重启生效] to 1, then restart the software.
  6. If the error persists after all of the above, use the CPU path for this recognition (uncheck 启用CUDA加速) — Qwen-ASR will still run, just slower.

Note: your input file name contains spaces (20 Series Illustration 03 Color Placement.m4a).

While not the cause of this CUDA error, renaming to a simple ASCII name (e.g.

myvideo.m4a) is recommended to avoid other FFmpeg-related failures.

请查阅相关文档:

Thinking...
Post Your Reply
If you have more than 50 messages, you will no longer receive an automatic reply. Please ask a new question!
Open source and free maintenance is not easy. If this project is helpful to you, please consider making a small donation to help the project continue to maintain and update.

Dondate via ko-fi.com

Related resource