Installation.md 5.11 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
# Installation Guide

## Installing with pip

**Prerequisites for installation via wheel or PyPI:**

- **Operating System**: Ubuntu 20.04 or later
- **Python Version**: >= 3.8
- **CUDA Version**: >= 11.0

11
The easiest way to install **tile-lang** is directly from PyPI using pip. To install the latest version, run the following command in your terminal:
12
13
14
15
16

```bash
pip install tilelang
```

17
Alternatively, you may choose to install **tile-lang** using prebuilt packages available on the Release Page:
18
19
20
21
22

```bash
pip install tilelang-0.0.0.dev0+ubuntu.20.4.cu120-py3-none-any.whl
```

23
To install the latest version of **tile-lang** from the GitHub repository, you can run the following command:
24
25
26
27
28

```bash
pip install git+https://github.com/tile-ai/tilelang.git
```

29
After installing **tile-lang**, you can verify the installation by running:
30
31
32
33
34
35
36
37
38
39
40
41

```bash
python -c "import tilelang; print(tilelang.__version__)"
```

## Building from Source

**Prerequisites for building from source:**

- **Operating System**: Linux
- **Python Version**: >= 3.7
- **CUDA Version**: >= 10.0
42
- **LLVM**: < 20 if you are using the bundled TVM submodule
43

44
We recommend using a Docker container with the necessary dependencies to build **tile-lang** from source. You can use the following command to run a Docker container with the required dependencies:
45
46
47
48
49

```bash
docker run --gpus all -it --rm --ipc=host nvcr.io/nvidia/pytorch:23.01-py3
```

50
To build and install **tile-lang** directly from source, follow these steps. This process requires certain pre-requisites from Apache TVM, which can be installed on Ubuntu/Debian-based systems using the following commands:
51
52

```bash
53
54
apt-get update
apt-get install -y python3 python3-dev python3-setuptools gcc zlib1g-dev build-essential cmake libedit-dev
55
56
```

57
After installing the prerequisites, you can clone the **tile-lang** repository and install it using pip:
58
59
60

```bash
git clone --recursive https://github.com/tile-ai/tilelang.git
61
cd tilelang
62
63
64
pip install .  # Please be patient, this may take some time.
```

65
If you want to install **tile-lang** in development mode, you can run the following command:
66
67
68
69
70

```bash
pip install -e .
```

71
We currently provide three methods to install **tile-lang**:
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135

1. [Install from Source (using your own TVM installation)](#install-method-1)
2. [Install from Source (using the bundled TVM submodule)](#install-method-2)
3. [Install Using the Provided Script](#install-method-3)

(install-method-1)=

### Method 1: Install from Source (Using Your Own TVM Installation)

If you already have a compatible TVM installation, follow these steps:

1. **Clone the Repository**:

```bash
git clone --recursive https://github.com/tile-ai/tilelang
cd tilelang
```

**Note**: Use the `--recursive` flag to include necessary submodules.

2. **Configure Build Options**:

Create a build directory and specify your existing TVM path:

```bash
mkdir build
cd build
cmake .. -DTVM_PREBUILD_PATH=/your/path/to/tvm/build  # e.g., /workspace/tvm/build
make -j 16
```

3. **Set Environment Variables**:

Update `PYTHONPATH` to include the `tile-lang` Python module:

```bash
export PYTHONPATH=/your/path/to/tilelang/:$PYTHONPATH
# TVM_IMPORT_PYTHON_PATH is used by 3rd-party frameworks to import TVM
export TVM_IMPORT_PYTHON_PATH=/your/path/to/tvm/python
```

(install-method-2)=

### Method 2: Install from Source (Using the Bundled TVM Submodule)

If you prefer to use the built-in TVM version, follow these instructions:

1. **Clone the Repository**:

```bash
git clone --recursive https://github.com/tile-ai/tilelang
cd tilelang
```

**Note**: Ensure the `--recursive` flag is included to fetch submodules.

2. **Configure Build Options**:

Copy the configuration file and enable the desired backends (e.g., LLVM and CUDA):

```bash
mkdir build
cp 3rdparty/tvm/cmake/config.cmake build
cd build
136
# echo "set(USE_LLVM ON)"  # set USE_LLVM to ON if using LLVM
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
echo "set(USE_CUDA ON)" >> config.cmake 
# or echo "set(USE_ROCM ON)" >> config.cmake to enable ROCm runtime
cmake ..
make -j 16
```

The build outputs (e.g., `libtilelang.so`, `libtvm.so`, `libtvm_runtime.so`) will be generated in the `build` directory.

3. **Set Environment Variables**:

Ensure the `tile-lang` Python package is in your `PYTHONPATH`:

```bash
export PYTHONPATH=/your/path/to/tilelang/:$PYTHONPATH
```

(install-method-3)=

### Method 3: Install Using the Provided Script

For a simplified installation, use the provided script:

1. **Clone the Repository**:

```bash
git clone --recursive https://github.com/tile-ai/tilelang
cd tilelang
```

2. **Run the Installation Script**:

```bash
bash install_cuda.sh
# or bash `install_amd.sh` if you want to enable ROCm runtime
171
172
173
174
175
176
177
178
179
180
181
```

## Install with Nightly Version

For users who want access to the latest features and improvements before official releases, we provide nightly builds of **tile-lang**.

```bash
pip install tilelang -f https://tile-ai.github.io/whl/nightly/cu121/
# or pip install tilelang --find-links https://tile-ai.github.io/whl/nightly/cu121/
```

182
> **Note:** Nightly builds contain the most recent code changes but may be less stable than official releases. They're ideal for testing new features or if you need a specific bugfix that hasn't been released yet.