Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.

[WorkAlbertX]

I tried to run Whisper on command line with Python3.10.9,PyTorch for CUDA11.7 and CUDA11.7,but every time I run it shows "Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model" and redownload the pt file.It bothered me for a long time,is there an enthusiastic person who can help me?Appreciate.

Total error message:
PS D:\Test_for_Whisper> whisper audio.mp3
C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper_init_.py:48: UserWarning: C:\Users\Albert_Administrator.cache\whisper\small.pt exists, but the SHA256 checksum does not match; re-downloading the file
warnings.warn(f"{download_target} exists, but the SHA256 checksum does not match; re-downloading the file")
99%|██████████████████████████████████████▊| 459M/461M [00:55<00:00, 8.65MiB/s]
Traceback (most recent call last):
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 196, in _run_module_as_main
return run_code(code, main_globals, None,
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 86, in run_code
exec(code, run_globals)
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\Scripts\whisper.exe_main.py", line 7, in
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\transcribe.py", line 310, in cli
model = load_model(model_name, device=device, download_root=model_dir)
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper_init.py", line 108, in load_model
checkpoint_file = _download(MODELS[name], download_root, in_memory)
File "C:\Users\Albert_Administrator\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper_init.py", line 62, in _download
raise RuntimeError("Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.")
RuntimeError: Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.

[WorkAlbertX]

I dont kown what happened,but the error has been fixed automatically after restarting.

[EtienneAb3d]

I had the same problem on an unstable machine with a lot of disk read/write errors. After a (very) long search, I found this was due to a SWAP configured on a SSD disk (I didn't have this problem before with some other SSD).

[adhiraj2001]

@EtienneAb3d I seem to have the same problem as you, what workaround did you use for this ?

[NIXON-L]

My problem was resolved by removing the ~/.cache/torch/whisperx-vad-segmentation.bin file

[Yongle-Fu]

same issue

[zhuiyi6]

C:\Users\Administrator.DESKTOP-DHKFNAB\Desktop>whisper test.mp3 --model large-v2 --language Chinese --output_format srt
C:\Users\Administrator.DESKTOP-DHKFNAB\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper_init_.py:63: UserWarning: C:\Users\Administrator.DESKTOP-DHKFNAB.cache\whisper\large-v2.pt exists, but the SHA256 checksum does not match; re-downloading the file
warnings.warn(
15%|█████▍ | 435M/2.87G [59:59<5:46:02, 127kiB/s]
Traceback (most recent call last):
File "", line 198, in run_module_as_main
File "", line 88, in run_code
File "C:\Users\Administrator.DESKTOP-DHKFNAB\AppData\Local\Programs\Python\Python311\Scripts\whisper.exe_main.py", line 7, in
File "C:\Users\Administrator.DESKTOP-DHKFNAB\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper\transcribe.py", line 440, in cli
model = load_model(model_name, device=device, download_root=model_dir)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Administrator.DESKTOP-DHKFNAB\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper_init.py", line 131, in load_model
checkpoint_file = _download(MODELS[name], download_root, in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Administrator.DESKTOP-DHKFNAB\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper_init.py", line 85, in _download
raise RuntimeError(
RuntimeError: Model has been downloaded but the SHA256 checksum does not not match. Please retry loading the model.

how to solve this problem

[Bailey-24]

restart what?

[Slepetys]

I solve it under WINDOWS 11 miniconda environment deleting the whole huggingface directory under default user/.cache.
Was a kind of brute force, implying that I needed to redownload many hugging face models, but it worked.

[search620]

remove those files and run the script again
"C:\Users\.cache\torch\whisperx-vad-segmentation.bin"
"C:\Users\USER-NAME.cache\torch\whisperx-vad-segmentation.bak"

[fraschm1998]

This fixed it for me on linux

[sdekna]

I had the same problem, my device is linux based, so rm ~/.cache/torch/whisperx-vad-segmentation.bin worked for me.
Thanks for the help.

[QuietWinterX]

this

[kovogo]

it works for me

[sirEgghead]

I have this same issue. I am running on Ubuntu 22.04. Fresh Whisper install. Have not yet had a successful transcribe. Download of the appropriate .pt file gets to 100% then sits there. I left one for 48 hours. Upon restarting Whisper, it says that the SHA checksum does not match and will download again. Same results. Tried manually deleting the .pt files from the .cache. No change in result.

[misutoneko]

Hi,

A late answer for sure, but here are some misc ideas for Linux...

The models are located in ~/.cache/whisper by default.
You could take a look in there and even check the SHA sums manually, if need be.
The tiny models are about ~73MB in size so should be pretty fast to dl unless you're on dialup :D

It's best to start out with tiny.en or other small models first.
Otherwise you could be in for a looong wait.
Besides, you should see good results with the jfk sample, even with the tiny models.

You could also try running it with CPU only by setting the env variable CUDA_VISIBLE_DEVICES=-1
That would be slow, but it'll tell you if your problem is CUDA-related.

GL

[royallenppt]

it seems related to network as well. i tried 3 times, and it can work

[lemo366]

我遇到了同样的问题，我用的google colab尝试了下面的两个命令就可以顺利运行了
! rm -rf ~/.cache/whisper
! rm -rf ~/.cache/torch

[medford-transcripts]

FYI, this happened to me, and I know what caused it. While transcribing, I got this error message:

Lightning automatically upgraded your loaded checkpoint from v1.5.4 to v2.4.0. To apply the upgrade to your files permanently, run python -m pytorch_lightning.utilities.upgrade_checkpoint %USERPROFILE%\.cache\torch\whisperx-vad-segmentation.bin

I issued the command, but I was using a CPU, not GPU, so I added the --map-to-cpu option:

python -m pytorch_lightning.utilities.upgrade_checkpoint --map-to-cpu %USERPROFILE%.cache\torch\whisperx-vad-segmentation.bin

But then I got this error. I fixed it by deleting %USERPROFILE%.cache\torch\whisperx-vad-segmentation.bin, and just accepting the original error message.

[NIXON-L]

It seems that my problem was resolved by removing the ~/.cache/torch/whisperx-vad-segmentation.bin file