| Name | mase-triton JSON |
| Version |
0.0.3
JSON |
| download |
| home_page | None |
| Summary | Triton kernels for MASE |
| upload_time | 2025-07-09 10:24:16 |
| maintainer | None |
| docs_url | None |
| author | None |
| requires_python | >=3.11 |
| license | None |
| keywords |
deep-learning
mase
triton
|
| VCS |
 |
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# MASE-Triton
Software-emulation & acceleration triton kernels for [MASE](https://github.com/DeepWok/mase).
## Install
Please ensure you are using Python 3.11 or later, and run MASE-Triton on **CUDA-enabled GPU**.
### PyPI
```bash
pip install mase-triton
```
### Build from Source
1. Install tox
```bash
pip install tox
```
2. Build & Install
```bash
tox -e build
```
Then the wheel file will be generated in `dist/` folder.
You can install it by `pip install path/to/wheel/file.whl`
## Functionality
- Random Bitflip
- [`random_bitflip_fn`](/src/mase_triton/random_bitflip/core.py): random bitflip function with backward support.
- [`layers.py`](/src/mase_triton/random_bitflip/layers.py): subclasses of `torch.nn.Module` that can be used in neural networks.
- `RandomBitflipDropout`
- `RandomBitflipLinear`
- MXFP: Simulate MXFP formats (Note that subnormal numbers are flushed to zero)
- [`functional`](/src/mase_triton/mxfp/functional/__init__.py)
- `extract_mxfp_tensor`: Cast a tensor to MXFP format (extracting the shared exponent and Minifloat elements).
- `compose_mxfp_tensor`: Cast an MXFP tensor to FP format (composing MXFP components).
- `mxfp_linear`: functional linear operation with MXFP support.
- `mxfp_matmul`: functional matrix multiplication with MXFP support.
- [`layers`](/src/mase_triton/mxfp/layers.py)
- `MXFPLinearPTQ`: Linear layer with MXFP support for post-training quantization (no back propagation support).
## Dev
1. Install [tox](https://tox.wiki/en/latest/index.html)
```
pip install tox
```
2. Create Dev Environment
```
tox -e dev
```
Raw data
{
"_id": null,
"home_page": null,
"name": "mase-triton",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.11",
"maintainer_email": "Cheng Zhang <chengzhang98@outlook.com>",
"keywords": "deep-learning, mase, triton",
"author": null,
"author_email": "Cheng Zhang <chengzhang98@outlook.com>",
"download_url": "https://files.pythonhosted.org/packages/09/47/1e7c773e455599c7e74d27e361da91fe90d28bb203d04263fa42997860ee/mase_triton-0.0.3.tar.gz",
"platform": null,
"description": "# MASE-Triton\n\nSoftware-emulation & acceleration triton kernels for [MASE](https://github.com/DeepWok/mase).\n\n## Install\n\nPlease ensure you are using Python 3.11 or later, and run MASE-Triton on **CUDA-enabled GPU**.\n\n### PyPI\n\n```bash\npip install mase-triton\n```\n\n### Build from Source\n\n1. Install tox\n\n ```bash\n pip install tox\n ```\n\n2. Build & Install\n\n ```bash\n tox -e build\n ```\n\n Then the wheel file will be generated in `dist/` folder.\n You can install it by `pip install path/to/wheel/file.whl`\n\n\n## Functionality\n- Random Bitflip\n - [`random_bitflip_fn`](/src/mase_triton/random_bitflip/core.py): random bitflip function with backward support.\n - [`layers.py`](/src/mase_triton/random_bitflip/layers.py): subclasses of `torch.nn.Module` that can be used in neural networks.\n - `RandomBitflipDropout`\n - `RandomBitflipLinear`\n- MXFP: Simulate MXFP formats (Note that subnormal numbers are flushed to zero)\n - [`functional`](/src/mase_triton/mxfp/functional/__init__.py)\n - `extract_mxfp_tensor`: Cast a tensor to MXFP format (extracting the shared exponent and Minifloat elements).\n - `compose_mxfp_tensor`: Cast an MXFP tensor to FP format (composing MXFP components).\n - `mxfp_linear`: functional linear operation with MXFP support.\n - `mxfp_matmul`: functional matrix multiplication with MXFP support.\n - [`layers`](/src/mase_triton/mxfp/layers.py)\n - `MXFPLinearPTQ`: Linear layer with MXFP support for post-training quantization (no back propagation support).\n\n\n## Dev\n\n1. Install [tox](https://tox.wiki/en/latest/index.html)\n ```\n pip install tox\n ```\n\n2. Create Dev Environment\n ```\n tox -e dev\n ```",
"bugtrack_url": null,
"license": null,
"summary": "Triton kernels for MASE",
"version": "0.0.3",
"project_urls": {
"Repository": "https://github.com/DeepWok/mase-triton.git"
},
"split_keywords": [
"deep-learning",
" mase",
" triton"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "6280d1d6cbbfa7e5870b8ffac6b5ee64636bb5bb897a16e5111e93ac735319dd",
"md5": "f6acb392db94a3312739d3fb7e981ab7",
"sha256": "8566f49f3a5b15a7ea523eef065c522f51e833181f8c8f08d8a8285fb075c6c6"
},
"downloads": -1,
"filename": "mase_triton-0.0.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "f6acb392db94a3312739d3fb7e981ab7",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.11",
"size": 34961,
"upload_time": "2025-07-09T10:24:14",
"upload_time_iso_8601": "2025-07-09T10:24:14.544602Z",
"url": "https://files.pythonhosted.org/packages/62/80/d1d6cbbfa7e5870b8ffac6b5ee64636bb5bb897a16e5111e93ac735319dd/mase_triton-0.0.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "09471e7c773e455599c7e74d27e361da91fe90d28bb203d04263fa42997860ee",
"md5": "d2bcea9156d72fdb4b627226d9af7e96",
"sha256": "aae3eff70dd8efcfb4a4f575a1b79e121c03e0912746d23d2c7b127c1bc6815c"
},
"downloads": -1,
"filename": "mase_triton-0.0.3.tar.gz",
"has_sig": false,
"md5_digest": "d2bcea9156d72fdb4b627226d9af7e96",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.11",
"size": 30660,
"upload_time": "2025-07-09T10:24:16",
"upload_time_iso_8601": "2025-07-09T10:24:16.599677Z",
"url": "https://files.pythonhosted.org/packages/09/47/1e7c773e455599c7e74d27e361da91fe90d28bb203d04263fa42997860ee/mase_triton-0.0.3.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-07-09 10:24:16",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "DeepWok",
"github_project": "mase-triton",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "mase-triton"
}
