add in gradio finetune 8bit value fix some stuff and add new transcripe into easy

[lpscr]

@SWivid hi i make some new fix update
and a add 8bit , and also i add in utlis infer new function for get easy transcripe
also fix sime values

[SWivid]

Yes, thanks!

[lpscr]

@SWivid Yes, I was sick the last few days :( , but I’m finally better now !! , so I can focus on some things here.

what about to load data for hugging face. arrow?
like this

also i wonder about the bigvgan ?any news i try finetune last days and like i say results in my case very good quality

[SWivid]

.arrow loading is great, an easy to understand and easy to use designed ui is the key for that while the loading stuff behind is straightforward

bigvgan is good at quality though tend to produce less diverse output to some extend, we can say it's good but still not perfect a vocoder (or maybe reasons lay in the stft part).
So in general, finetuning on vocos or bigvgan both ok.

[justinjohn0306]

@lpscr I can't test the model during the training, got this msg: Traceback (most recent call last): File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\gradio\queueing.py", line 536, in process_events response = await route_utils.call_process_api( File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\gradio\route_utils.py", line 322, in call_process_api output = await app.get_blocks().process_api( File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\gradio\blocks.py", line 1935, in process_api result = await self.call_function( File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\gradio\blocks.py", line 1520, in call_function prediction = await anyio.to_thread.run_sync( # type: ignore File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\anyio\to_thread.py", line 56, in run_sync return await get_async_backend().run_sync_in_worker_thread( File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\anyio\_backends\_asyncio.py", line 2441, in run_sync_in_worker_thread return await future File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\anyio\_backends\_asyncio.py", line 943, in run result = context.run(func, *args) File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\gradio\utils.py", line 826, in wrapper response = f(*args, **kwargs) File "E:\REPOS\F5-TTS\src\f5_tts\train\finetune_gradio.py", line 1224, in infer tts_api = F5TTS( File "E:\REPOS\F5-TTS\src\f5_tts\api.py", line 52, in __init__ self.load_ema_model( File "E:\REPOS\F5-TTS\src\f5_tts\api.py", line 82, in load_ema_model self.ema_model = load_model( File "E:\REPOS\F5-TTS\src\f5_tts\infer\utils_infer.py", line 256, in load_model model = load_checkpoint(model, ckpt_path, device, dtype=dtype, use_ema=use_ema) File "E:\REPOS\F5-TTS\src\f5_tts\infer\utils_infer.py", line 178, in load_checkpoint if torch.cuda.is_available() and torch.cuda.get_device_properties(device).major >= 6 File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\torch\cuda\__init__.py", line 466, in get_device_properties device = _get_device_index(device, optional=True) File "E:\Programs\anaconda3\envs\f5-tts\lib\site-packages\torch\cuda\_utils.py", line 34, in _get_device_index raise ValueError(f"Expected a cuda device, but got: {device}") ValueError: Expected a cuda device, but got: cpu

Seems like it no longer can switch between cuda and cpu

[SWivid]

@justinjohn0306 thanks~
fixed in last commit 84db002