tileops 0.0.1.dev0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
|
@@ -0,0 +1,82 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: tileops
|
|
3
|
+
Version: 0.0.1.dev0
|
|
4
|
+
Summary: TileOPs kernels for efficient inference
|
|
5
|
+
Author: Tile-AI
|
|
6
|
+
License-Expression: MIT
|
|
7
|
+
Project-URL: Homepage, https://github.com/tile-ai/tileops
|
|
8
|
+
Project-URL: Bug Tracker, https://github.com/tile-ai/tileops/issues
|
|
9
|
+
Classifier: Development Status :: 3 - Alpha
|
|
10
|
+
Classifier: Programming Language :: Python :: 3
|
|
11
|
+
Requires-Python: >=3.8
|
|
12
|
+
Description-Content-Type: text/markdown
|
|
13
|
+
License-File: LICENSE
|
|
14
|
+
Requires-Dist: torch
|
|
15
|
+
Requires-Dist: tilelang
|
|
16
|
+
Dynamic: license-file
|
|
17
|
+
|
|
18
|
+
# TileOPs (TOP)
|
|
19
|
+
|
|
20
|
+
**TileOPs (TOP)** is a high-performance machine learning operator collections built to run on the [TileLang](https://github.com/tile-ai/tilelang) backend. It offers efficient, modular, and composable implementations optimized for AI workloads.
|
|
21
|
+
|
|
22
|
+
---
|
|
23
|
+
|
|
24
|
+
|
|
25
|
+
## 📦 Installation
|
|
26
|
+
|
|
27
|
+
### Requirements
|
|
28
|
+
|
|
29
|
+
- Python 3.8+
|
|
30
|
+
- PyTorch >= 2.1
|
|
31
|
+
- [TileLang](https://github.com/tilelang/tilelang)
|
|
32
|
+
- Triton (optional, for selected fast kernels)
|
|
33
|
+
|
|
34
|
+
### Install (editable mode for development)
|
|
35
|
+
|
|
36
|
+
```bash
|
|
37
|
+
git clone https://github.com/tile-ai/TileOPs
|
|
38
|
+
cd TileOPs
|
|
39
|
+
git submodule update --init --recursive
|
|
40
|
+
pip install -e .
|
|
41
|
+
```
|
|
42
|
+
|
|
43
|
+
## 🚀 Quick Usage
|
|
44
|
+
|
|
45
|
+
```python
|
|
46
|
+
import torch
|
|
47
|
+
import top
|
|
48
|
+
from top import MLAKernel
|
|
49
|
+
|
|
50
|
+
device = "cuda"
|
|
51
|
+
dtype = torch.float16
|
|
52
|
+
|
|
53
|
+
batch = 128
|
|
54
|
+
heads = 64
|
|
55
|
+
kv_heads = 1
|
|
56
|
+
kv_ctx = 8192
|
|
57
|
+
dim = 512
|
|
58
|
+
pe_dim = 64
|
|
59
|
+
|
|
60
|
+
# Query input: [batch, heads, dim]
|
|
61
|
+
q = torch.randn(batch, heads, dim, device=device, dtype=dtype)
|
|
62
|
+
|
|
63
|
+
# Query positional encoding: [batch, heads, pe_dim]
|
|
64
|
+
q_pe = torch.randn(batch, heads, pe_dim, device=device, dtype=dtype)
|
|
65
|
+
|
|
66
|
+
# KV cache input: [batch, kv_ctx, kv_heads, dim]
|
|
67
|
+
kv = torch.randn(batch, kv_ctx, kv_heads, dim, device=device, dtype=dtype)
|
|
68
|
+
|
|
69
|
+
# KV positional encoding: [batch, kv_ctx, kv_heads, pe_dim]
|
|
70
|
+
k_pe = torch.randn(batch, kv_ctx, kv_heads, pe_dim, device=device, dtype=dtype)
|
|
71
|
+
|
|
72
|
+
# Use MLA kernel
|
|
73
|
+
block_N = 64
|
|
74
|
+
block_H = 64
|
|
75
|
+
num_split = 1
|
|
76
|
+
|
|
77
|
+
mla = MLAKernel(batch, heads, kv_heads, kv_ctx, dim, pe_dim, block_N, block_H, num_split)
|
|
78
|
+
|
|
79
|
+
out = mla(q, q_pe, kv, k_pe)
|
|
80
|
+
```
|
|
81
|
+
|
|
82
|
+
## Acknowledgments
|
|
@@ -0,0 +1,6 @@
|
|
|
1
|
+
tileops-0.0.1.dev0.dist-info/licenses/LICENSE,sha256=hr_8lcbfSW8PTqqDHjB_EfPueXTmbaAz7nDy2F97vl0,1126
|
|
2
|
+
top/__init__.py,sha256=HZiXk6QxdAcFqHnRvXeVE797P22yDK7MlLDvkk4V5F0,847
|
|
3
|
+
tileops-0.0.1.dev0.dist-info/METADATA,sha256=Cwnh1OJkSy9JsrwE_cSXM7ASakyt2VWB2l7sLh69rGA,2008
|
|
4
|
+
tileops-0.0.1.dev0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
|
|
5
|
+
tileops-0.0.1.dev0.dist-info/top_level.txt,sha256=994pR8ZMtkNeFfsr7zWdHtX2NWsq67eyBTXjdykE5ts,4
|
|
6
|
+
tileops-0.0.1.dev0.dist-info/RECORD,,
|
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) Tile-AI.
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
top
|
top/__init__.py
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
# Copyright (c) Tile-AI.
|
|
2
|
+
# Licensed under the MIT License.
|
|
3
|
+
|
|
4
|
+
from .kernel.mha import MHAKernel, MHADecodeKernel
|
|
5
|
+
from .kernel.mla import MLAKernel
|
|
6
|
+
from .kernel.gqa import GQAKernel, GQADecodeKernel
|
|
7
|
+
from .kernel.mamba_chunk_scan import MambaChunkScanKernel
|
|
8
|
+
from .kernel.mamba_chunk_state import MambaChunkStateKernel
|
|
9
|
+
from .kernel.blocksparse_attention import BlockSparseAttentionKernel
|
|
10
|
+
from .kernel.linear_attention.linear_attn import LinearAttentionFusedChunkKernel, LinearAttentionFusedRecurrentKernel
|
|
11
|
+
from .kernel.bitnet import Bitnet_158_int8xint2_kernel
|
|
12
|
+
|
|
13
|
+
__all__ = [
|
|
14
|
+
"MHAKernel", "MHADecodeKernel", "MLAKernel", "GQAKernel", "GQADecodeKernel",
|
|
15
|
+
"MambaChunkScanKernel", "MambaChunkStateKernel", "BlockSparseAttentionKernel",
|
|
16
|
+
"LinearAttentionFusedChunkKernel", "LinearAttentionFusedRecurrentKernel",
|
|
17
|
+
"Bitnet_158_int8xint2_kernel"
|
|
18
|
+
]
|