Fine-tune the Whisper speech recognition model to support training without timestamp data, training with timestamp data, and training without speech data. Accelerate inference and support Web deployment, Windows desktop deployment, and Android deployment
Apache License 2.0
164
stars
8
forks
source link
Finetune whisper-large-v3 is way slower than Belle-whisper-large-v3-zh-punct #3
I tried to use finetune.py to perform Lora finetuning on both models to see the result.
For Belle-whisper-large-v3-zh-punct, the speed is 1000/68625 [2:35:56<8:55:21, 2.11it/s]
while for whisper-large-v3, it shows [05:33<6351:43:53, 333.21s/it].
Does anyone know what may have caused this issue?
The running command is CUDA_VISIBLE_DEVICES=0 python finetune.py --base_model=openai/whisper-large-v3 --output_dir=output/
I was using RTX4090 card
I tried to use finetune.py to perform Lora finetuning on both models to see the result. For Belle-whisper-large-v3-zh-punct, the speed is 1000/68625 [2:35:56<8:55:21, 2.11it/s] while for whisper-large-v3, it shows [05:33<6351:43:53, 333.21s/it].
Does anyone know what may have caused this issue?
The running command is
CUDA_VISIBLE_DEVICES=0 python finetune.py --base_model=openai/whisper-large-v3 --output_dir=output/I was using RTX4090 card