Daqiu Shi
Daqiu Shi
> 这个有没有自己本地实验 打印一下optim下module的lr 看看是否奏效呀? 我下午测一下~
测了下,warmup_method不是linear的话没问题。 ``` optimizer.state_dict()["param_groups"] [{'_options': {'lr': 9.375e-09, 'eps': 1e-08, 'betas': [0.9, 0.95], 'weight_decay': 0.05, 'bias_correction1': 1.0, 'bias_correction2': 1.0, 'do_bias_correction': True, 'amsgrad': False, 'initial_lr': 9.375e-06}, '_enable_clip_grad': False, 'params': [0, 1, 2, 3,...
from ~~flowvision~~torchvision.models._utils import IntermediateLayerGetter 不支持
> > from ~flowvision~torchvision.models._utils import IntermediateLayerGetter > > 不支持 > > 这个地方flowvision不支持吗,那这边我去flowvision下更新一下然后打个tag包吧 嗯不支持的~ 哈哈行~我刚准备绕一下先
> > > > from ~flowvision~torchvision.models._utils import IntermediateLayerGetter > > > > 不支持 > > > > > > > > > 这个地方flowvision不支持吗,那这边我去flowvision下更新一下然后打个tag包吧 > > > > > > 嗯不支持的~...
**oneflow min/max op 无法在不同数据类型间执行** ``` >>> flow.__version__ '0.8.0.dev20220411+cu102' >>> torch.__version__ '1.11.0+cu102' ``` **最小复现代码** 以float64和float32为例,其他不同类型间同理 **torch** ``` >>> import torch >>> x = torch.randn(5, dtype=torch.float32) >>> y = torch.randn(5, dtype=torch.float64) >>>...
flow.cumsum支持,tensor.cumsum不支持 ``` >>> flow.__version__ '0.8.0.dev20220411+cu102' >>> torch.__version__ '1.11.0+cu102' ``` ``` >>> x = flow.randn(10,10,10) >>> y = flow.cumsum(x,1) >>> y = x.cumsum(1) Traceback (most recent call last): File "", line...
flow.as_tensor从numpy array转换时无法显式指定data type 已同步至:https://github.com/Oneflow-Inc/OneTeam/issues/1207#issuecomment-1073432125 ``` >>> flow.__version__ '0.8.0.dev20220417+cu112' >>> torch.__version__ '1.11.0+cu113' ``` 最小复现代码: flow ``` >>> x=np.random.randn(10) >>> flow.as_tensor(x) tensor([-0.3546, -0.6711, -1.3503, 0.7537, 0.4851, 0.4599, 1.4330, 0.2376, 0.3307, -0.1530], dtype=oneflow.float64)...
**for m in tensor: m[0]=False 并不会改变tensor数值** ``` >>> flow.__version__ '0.8.0.dev20220417+cu112' >>> torch.__version__ '1.11.0+cu113' ``` 最小复现代码: oneflow: ``` >>> mask = flow.ones(10,10) >>> for m in mask: ... m[0]=False ... >>>...
**tensor.copy_()不管用** ``` >>> flow.__version__ '0.8.0.dev20220417+cu112' >>> torch.__version__ '1.11.0+cu113' ``` 最小复现代码: flow: ``` >>> x = flow.ones(5,5) >>> y = flow.zeros(3,3) >>> x[:3,:3].copy_(y) >>> x tensor([[1., 1., 1., 1., 1.], [1.,...