search

Const-me / Whisper

High-performance GPGPU inference of OpenAI's Whisper automatic speech recognition (ASR) model
Mozilla Public License 2.0
7.65k stars 664 forks source link

runFullImpl: failed to generate timestamp token - skipping one second while using ggml-large-v3_2 #190

Open awaqawaqawaq opened 7 months ago

awaqawaqawaq commented 7 months ago

runFullImpl: failed to generate timestamp token - skipping one second while using ggml-large-v3_2 but there is no problems when i use ggml-large-v2 and i also want to ask why cant i use ggml-large-v3-q5_0.bin

leixuexin commented 7 months ago

the same problem

datan79 commented 7 months ago

yup exact same problem

EinProfispieler commented 7 months ago

same here

emcodem commented 7 months ago

solution here: https://github.com/Const-me/Whisper/issues/188

ThisIsLoui commented 6 months ago

solution here: #188

It dosen't work. And the same error exists when using ggml-medium. I think it was caused by a high sampling rate of the target video.

jwlrxy commented 6 months ago

i cant use v3 model. is it not supported by this?

ThisIsLoui commented 6 months ago

i cant use v3 model. is it not supported by this?

You can try the custom version made by @emcodem, which can be found in this issue: #188 (see the fourth comment)

thisanickname commented 5 months ago

same

PandaBear1314 commented 3 months ago

same here