Aborted (core dumped) in flow.QatConv3d

[x0w3n]

Summary

When I pass a negative value to the flow.QatConv3d parameter, it crashes. and tries to allocate a huge memory size. We believe this is a very serious bug that tries to allocate a large amount of memory and could lead to a memory overflow vulnerability.

Code to reproduce bug

import oneflow as flow
import numpy as np
arr = np.random.randn(20, 16, 50)
input = flow.Tensor(arr)
m = flow.nn.QatConv3d(-16, 33, 3, stride=2, quantization_formula="google", quantization_bit=8, quantization_scheme="symmetric")
output = m(input)

System Information

• What is your OneFlow installation (pip, source, dockerhub): source
• OS: Ubuntu 22.04.3 LTS
• OneFlow version (run python3 -m oneflow --doctor):

  path: ['*/temp/oneflow-1.0.0/python/oneflow']
  version: 0.9.1+cpu.git.f97cd07
  git_commit: f97cd07
  cmake_build_type: Release
  rdma: False
  mlir: False

• Python version: 3.10.0
• CUDA driver version: 12.2
• GPU models: NVIDIA GeForce RTX 3090
• Other info: None