Yaodada12

静态图显存回收

7

模型太大，采用静态图推理时显存占用很高，怎么回收显存？有没有类似这样回收显存的方法： model.to('cpu') oneflow.cuda.empty_cache()

[Bug] TypeError: 'NoneType' object is not subscriptable

4

### Describe the bug when I use tts_models/multilingual/multi-dataset/xtts_v2, speaker cannot be specified. ### To Reproduce ``` from TTS.api import TTS # List available 🐸TTS models print(TTS().list_models()) # Init TTS tts...

ValueError: No matching select or slice.

3

``` import torch import torch.nn as nn import numpy as np import coremltools as ct import librosa from fairseq import checkpoint_utils audio, _ = librosa.load("/Users/admin/Desktop/yao/VC/data/test_women/women.mp3", sr=16000) feats = torch.from_numpy(audio) feats...

models下的三个模型均不能使用

2

``` RuntimeError: Error(s) in loading state_dict for UNet: Missing key(s) in state_dict: "_block2_2.0.weight", "_block2_2.0.bias", "_block4_1.0.weight", "_block4_1.0.bias", "_block4_1.2.weight", "_block4_1.2.bias", "_block4_1.4.weight", "_block4_1.4.bias". size mismatch for _block4.0.weight: copying a param with shape torch.Size([96,...

When testing Chinese, there are no punctuation marks in the results！

4

I use both faster-whisper-v2 and faster-whisper-v3. ``` from faster_whisper import WhisperModel model = WhisperModel("large-v3") segments, info = model.transcribe("zh_audio.mp3") for segment in segments: print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))...

使用sd2.1模型微调特定人物报错OOM

1

显卡是3090，微调sd1.4和sd1.5的fp32模型均没有问题，微调sd2.1的fp16和fp32模型时均显示OOM。

蒸馏后的学生模型是直接在新数据集上微调，还是需要先在新数据集上微调教师模型，再用新数据集去蒸馏学生模型？

关于算法的类型

1

这个算法是属于many-to-many还是any-to-any，可以在不训练的情况下，支持任意人的变声吗？any-to-any可以在不训练的情况下实现任意人的变声吗？

It‘s a awesome work，When is the code open source？