Add repository documentation

README.md

@@ -0,0 +1,135 @@
+---
+license: mit
+title: Triton Windows Pre-built Binaries
+sdk: static
+emoji: ⚡
+colorFrom: blue
+colorTo: green
+tags:
+  - triton
+  - windows
+  - cuda
+  - deep-learning
+  - gpu
+---
+
+# Triton for Windows - Pre-built Binaries
+
+**Developed by Dr. Ankush Rai** | [GitHub](https://github.com/RaiAnk)
+
+This repository hosts pre-built Windows binaries for [OpenAI Triton](https://github.com/triton-lang/triton), a language and compiler for writing highly efficient custom Deep-Learning primitives.
+
+## Available Downloads
+
+### Triton Wheels (Python)
+
+| Python | CUDA | Wheel | Size |
+|--------|------|-------|------|
+| 3.11 | 12.x | `wheels/triton-3.2.0+windows-cp311-cp311-win_amd64.whl` | ~100MB |
+
+### Pre-built LLVM (Optional)
+
+If you want to build Triton from source or develop custom kernels:
+
+| File | Description | Size |
+|------|-------------|------|
+| `llvm/llvm-triton-windows-x64.zip` | LLVM with MLIR support | ~800MB |
+
+## Installation
+
+### Quick Install
+
+```bash
+# 1. Download the wheel for your Python version
+# 2. Install with pip:
+pip install triton-3.2.0+windows-cp311-cp311-win_amd64.whl
+```
+
+### Using the Installer Script
+
+```bash
+# Clone the source repository
+git clone https://github.com/RaiAnk/triton-windows.git
+cd triton-windows
+
+# Run installer (downloads from this HuggingFace repo)
+python install_triton_windows.py
+```
+
+## Requirements
+
+- **Windows 10/11 x64**
+- **Python 3.10, 3.11, or 3.12**
+- **NVIDIA CUDA Toolkit 12.x**
+- **PyTorch 2.4+ with CUDA support**
+
+## Verify Installation
+
+```python
+import triton
+import triton.language as tl
+import torch
+
+@triton.jit
+def add_kernel(x_ptr, y_ptr, output_ptr, n_elements, BLOCK_SIZE: tl.constexpr):
+    pid = tl.program_id(axis=0)
+    block_start = pid * BLOCK_SIZE
+    offsets = block_start + tl.arange(0, BLOCK_SIZE)
+    mask = offsets < n_elements
+    x = tl.load(x_ptr + offsets, mask=mask)
+    y = tl.load(y_ptr + offsets, mask=mask)
+    output = x + y
+    tl.store(output_ptr + offsets, output, mask=mask)
+
+# Test
+size = 1024
+x = torch.rand(size, device='cuda')
+y = torch.rand(size, device='cuda')
+output = torch.empty_like(x)
+
+grid = lambda meta: (triton.cdiv(size, meta['BLOCK_SIZE']),)
+add_kernel[grid](x, y, output, size, BLOCK_SIZE=256)
+
+print("Triton is working!" if torch.allclose(output, x + y) else "Error!")
+```
+
+## Source Code
+
+The modified Triton source code with Windows support is available at:
+- **GitHub**: https://github.com/RaiAnk/triton-windows
+
+## Build from Source
+
+If you need to build from source (for unsupported Python versions or custom modifications):
+
+1. Download pre-built LLVM from this repository
+2. Extract to `C:\llvm-triton`
+3. Set environment variables:
+   ```cmd
+   set LLVM_SYSPATH=C:\llvm-triton
+   set LLVM_LIBRARY_DIR=C:\llvm-triton\lib
+   set LLVM_INCLUDE_DIRS=C:\llvm-triton\include
+   ```
+4. Build Triton:
+   ```cmd
+   pip install -e .
+   ```
+
+## Known Limitations
+
+- AMD ROCm is not supported on Windows
+- Some advanced features may have limited functionality
+- Proton profiler has limited Windows support
+
+## License
+
+Triton is licensed under the MIT License. See the [original repository](https://github.com/triton-lang/triton) for details.
+
+## Credits
+
+- **Original Triton**: OpenAI and the Triton team
+- **Windows Port**: Dr. Ankush Rai
+
+## Support
+
+For issues with Windows support, please open an issue on the GitHub repository.