oneflow

oneflow copied to clipboard

Flowingsun007

## Summary When dealing with inf and nan, "oneflow.quantile" gets different output on the cpu and gpu ## Code to reproduce bug ``` import oneflow as flow x1 = flow.tensor([0,...

Redempt1onzzZZ

Dev fused nms

1

This PR is mainly to support the NPU-compatible SortedNMS operator, so sorted_score and input_indices have been added.

ShawnXuan

Npu use cpu argwhere

4

Add NPU support for ArgWhereFunctor: - When the input tensor is on NPU, it is first moved to CPU to perform the argwhere operation; - The result is then moved...

ShawnXuan

ShawnXuan

## Summary Using `flow.nn.BCELoss` on unbounded input logits (e.g., raw outputs from `flow.randn()`) leads to a **segmentation fault (core dumped)**. The BCELoss operator internally expects the input tensor to contain...

deepliao

## Summary When applying `flow.nn.Conv2d` to an input whose spatial dimensions are too small for the given kernel size, stride, and padding, OneFlow attempts to compute a convolution with a...

deepliao

## Summary Calling `unbind()` on a **zero-dimensional scalar tensor** in OneFlow results in a **segmentation fault** (core dumped), instead of raising a proper `RuntimeError` or `ValueError`. This crash occurs because...

deepliao

## Summary Creating a `DoubleTensor` from a Python generator object (`generator`) causes a **segmentation fault** in OneFlow. This indicates that the tensor constructor does not validate input types properly. In...

deepliao

## Summary OneFlow does not properly validate data types during tensor arithmetic involving `int8` tensors. Specifically, performing an operation like `int8_tensor + "a"` (string) leads to a **segmentation fault**, rather...

deepliao