Ehsan Azar

**Is your feature request related to a problem? Please describe.** What is the best way to run DeepSpeed inference in C++? **Describe the solution you'd like** Documenting if it is...

Document command line to kill language server

1

I first asked [this feature here](https://github.com/microsoft/vscode/issues/97036). I use project with Python/C++ over SSH. After few days of use, I guess VSCode updates and previous language server processes remain dangling. I...

# Ask a Question ### Question This is the explanation [from torchvision nms](https://pytorch.org/vision/stable/generated/torchvision.ops.nms.html) > Returns > int64 tensor with the indices of the elements that have been kept by NMS,...

By default do not choose CUDNN engine in CPU mode

2

even if it is compiled with `USE_CUDNN` Fixes #3317

submission for pip

1

Do you plan to submit the application for inclusion in PyPI package index? This will greatly help users who want to use this nice library in Python.

rusti with Jupyter

8

I want to learn `rust` but in the context of data processing, and `rusti` looks like the way to go. Do you have any plans to support a kernel for...

Keep aspect ratio

1

Option to keep the aspect ratio

When I compile a model that had `nn.Conv2dBiasFewChannels` I get this error: ```text Profile: conv2d_bias_few_channels_1: NI == 1 && HI == 384 && WI == 384 && CI == 4...

nn.Conv2d output mismatch with torch.nn.Conv2d

6

This is a full repro: ```python import torch import numpy as np from collections import OrderedDict from aitemplate.testing import detect_target from aitemplate.frontend import nn, Tensor from aitemplate.compiler import compile_model def...

ops.conv2d(group=256) outputs NaN and Inf

3

This is the UnitTest ```python import unittest import torch from aitemplate.compiler import compile_model, ops from aitemplate.frontend import Tensor from aitemplate.testing import detect_target class ConvGroupTestCase(unittest.TestCase): def test_fp16(self): groups = 256 #...