CUBLAS_STATUS_EXECUTION_FAILED on 1650 Ti

[TURING-V2]

Ryzen 7 4800H 16gb, 1650 ti mobile 4gb vram , Gentoo , Nvidia driver 515.76
It loads 2gb on vram nvidia then crashes

PYTHONUSERBASE=/media/hdd/DeadCode/whisper/ bin/whisper /media/hdd/Videos/tmp/test102.aac --model small --language Japanese --task translate --output_dir /media/hdd/Videos/tmp/
Traceback (most recent call last):
File "/media/hdd/DeadCode/whisper/bin/whisper", line 8, in
sys.exit(cli())
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/transcribe.py", line 275, in cli
result = transcribe(model, audio_path, temperature=temperature, **args)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/transcribe.py", line 163, in transcribe
result = decode_with_fallback(segment)[0]
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/transcribe.py", line 101, in decode_with_fallback
results = model.decode(segment, options)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/decoding.py", line 700, in decode
result = DecodingTask(model, options).run(mel)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/decoding.py", line 616, in run
audio_features: Tensor = self._get_audio_features(mel) # encoder forward pass
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/decoding.py", line 560, in _get_audio_features
audio_features = self.model.encoder(mel)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/model.py", line 156, in forward
x = block(x)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/model.py", line 127, in forward
x = x + self.mlp(self.mlp_ln(x))
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/nn/modules/container.py", line 139, in forward
input = module(input)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/media/hdd/DeadCode/whisper/lib/python3.10/site-packages/whisper/model.py", line 36, in forward
return F.linear(
RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, &fbeta, c, CUDA_R_16F, ldc, CUDA_R_32F, CUBLAS_GEMM_DFALT_TENSOR_OP)

[TURING-V2]

WORKING Now with --fp16 False

[gosuimbalyndh]

Thank you
Is your GTX 1650 TI enough for OpenAI Whisper or other speech-to-text transcribe, other AI applications?
Best wishes

[xuanyinmu]

i am the beginer，what meaning of the "--fp16 False", can you be more specific?

[TURING-V2]

floating point operation 16 bit is set false will choose 32 bit , which is slower taking double space for same data storage on vram . 16 bit precision will be less but enough to range to represent values in neural net operation .
on fp16 single gpu clock cycle will process two 16bit values at a time in single core. fp32 have to do it in 2 cycles.
Although my gpu supports fp16 and its a bug .