Isuxiz Slidder

尝试回答第一个问题： 太长不看版本： 这份代码中参数的随机初始化是在初始化Meta类实例时初始化`self.net`成员变量的时候做的：`self.net = Learner(config, args.imgc, args.imgsz)`，因此整个训练过程中确实只有一次随机取初值。此初值会每个epoch被adam优化器原地优化一次，下一个epoch开始后优化后`self.net`成员变量的内部参数就是上一个epoch更新好的 详细： 82行的地方`vars=None`的含义是不使用外部传入的参数而是使用`self.net`内部自带的参数，而**不是重新随机初始化一份参数给`self.net`**！看下Learner类的源码就知道了： ```python if vars is None: vars = self.vars ``` 只是单纯的让变量`var`指向内部的参数列表。 而真正随机初始化参数是在初始化一个Learner类的实例的时候做的，以一个线性变化层为例： ```python elif name is 'linear': # ↓初始化了一个全1的参数并将其kaiming标准化 w = nn.Parameter(torch.ones(*param))...

I use DeepFM implemented by FuxiCTR(2.0+), dataset is downloaded from link in [this section](https://github.com/openbenchmark/BARS/tree/main/datasets/Avazu#avazu_x4) and don't do any preprocess(nor did the preprocessing of x4_001 or x4_002 below). my setting: dataset_config.yaml...

I figured out a way to avoid this problem. After trying I found out that this OOM is caused by setting the dtype to str, so I changed dataset_config.yaml: ```yaml...

Same issue. It seems to only trigger stably when the input audio is long (in my case, when length >= 5 minutes).

Really need this feature!

I think this is by design, since 24 is not a power of 2. I encountered the same error on linux, and it was solved by using the `normalize=True` parameter...

+1 Please bring multi-line style tabs back, at least provide a param in `gr.Tab` to control.

Same problem. ``` File "/usr/local/lib/python3.12/dist-packages/funasr/models/sense_voice/utils/ctc_alignment.py", line 45, in ctc_forced_align best_score[:, padding_num + 0] = log_probs[:, 0, blank] ~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^ IndexError: index 2 is out of bounds for dimension 1 with size...

Find a temp solution, plz see [this PR](https://github.com/modelscope/FunASR/pull/2454).

In most cases, mp4 / mov just can not be input when using pipe. Because you can not ensure meta data exists at file's begining. See https://github.com/fluent-ffmpeg/node-fluent-ffmpeg/issues/932 for more.