README.md 1.09 KB
Newer Older
Casper's avatar
Casper committed
1
2
3
4
5
6
7
# AutoAWQ Kernels

AutoAWQ Kernels is a new package that is split up from the [main repository](https://github.com/casper-hansen/AutoAWQ) in order to avoid compilation times.

## Requirements

- Windows: Must use WSL2.
Ilyas Moutawwakil's avatar
Ilyas Moutawwakil committed
8
9
10
11
12
13

- NVIDIA:
  - GPU: Must be compute capability 7.5 or higher.
  - CUDA Toolkit: Must be 11.8 or higher.
- AMD:
  - ROCm: Must be 5.6 or higher.
Casper's avatar
Casper committed
14
15
16
17
18

## Install

### Install from PyPi

Ilyas Moutawwakil's avatar
Ilyas Moutawwakil committed
19
20
The package is available on PyPi with CUDA 12.1.1 wheels:

Casper's avatar
Casper committed
21
22
23
24
```
pip install autoawq-kernels
```

Ilyas Moutawwakil's avatar
Ilyas Moutawwakil committed
25
26
27
28
29
30
31
32
### Install release wheels

For ROCm and other CUDA versions, you can use the wheels published at each [release](https://github.com/casper-hansen/AutoAWQ_kernels/releases/):

```
pip install https://github.com/casper-hansen/AutoAWQ_kernels/releases/download/v0.0.2/autoawq_kernels-0.0.2+rocm561-cp310-cp310-linux_x86_64.whl
```

Casper's avatar
Casper committed
33
### Build from source
Ilyas Moutawwakil's avatar
Ilyas Moutawwakil committed
34
You can also build from source:
Casper's avatar
Casper committed
35
36
37
38
39
40

```
git clone https://github.com/casper-hansen/AutoAWQ_kernels
cd AutoAWQ_kernels
pip install -e .
```
Ilyas Moutawwakil's avatar
Ilyas Moutawwakil committed
41
42

To build for ROCm, you need to first install the following packages `rocsparse-dev hipsparse-dev rocthrust-dev rocblas-dev hipblas-dev`.