oneflow

oneflow copied to clipboard

Some Modules miss `dtype` property

1

In PyTorch，Conv2D module can pass `dtype` property to create different dtype parameter, but in OneFlow it miss this property.

MARD1NO

oneflow.nn.MovingAverageMinMaxObserver produces "Aborted (core dumped)" bug

1

## Summary The following code snippet produces "Aborted (core dumped)" bug with "dtype=oneflow.float64". However, "dtype=oneflow.float32" won't have this problem. ## Code to reproduce bug input1 = oneflow.rand(9, 40, 20, 10,dtype=oneflow.float64)...

xxxyyyzzz12345

oneflow.nn.conv2D causes "core dumped"

1

## Summary The following code snippet produces "Aborted (core dumped)" bug with "dtype=oneflow.float64". However, "dtype=oneflow.float32" won't have this problem. ## Code to reproduce bug import oneflow input = oneflow.rand(4,10,32,32,dtype=oneflow.float64) m...

xxxyyyzzz12345

Cannot run oneflow.roi_align() properly

1

## Summary Traceback (most recent call last): File "/mnt/sda/yjy/summer/DL-autograd-torch-main/yjy_test.py", line 24, in output = oneflow.roi_align(input0,input1,input2,input3,input4,input5,input6) oneflow._oneflow_internal.exception.OpKernelNotFoundException: File "/home/ci-user/runners/release/_work/oneflow/oneflow/oneflow/core/framework/op_interpreter/op_interpreter_util.cpp", line 139, in Dispatch Dispatch(op_expr, inputs, ctx) File "/home/ci-user/runners/release/_work/oneflow/oneflow/oneflow/core/framework/op_interpreter/op_interpreter_util.cpp", line 131, in...

xxxyyyzzz12345

目前只注册了int64_t，而根据外部反馈有时候也会用到uint8_t这些类型

MARD1NO

profiler 支持按 op attr 而不仅仅是 op type 和 shape 给 op 分组

daquexian

liufengwei0103

Unary math op primitive-based kernel

8

EsdeathYZH

random_tensor() returns [-1, 1) for float and [-10000, 10001) for int

17

修复 random_tensor() 默认范围没有负数的问题，修复 random_tensor(dtype=int64) 只返回 0 的问题

daquexian

DTR

1

基于 add_dtr_feature 分支，仅供 comment 使用

daquexian