Metadata-Version: 2.4
Name: exowindows
Version: 0.1.2
Summary: Fast, distributed ML scaling on Windows. Connect computers over Ethernet, Wi-Fi, USB-C, or Thunderbolt.
License: Apache-2.0
Requires-Python: >=3.11
Requires-Dist: aiofiles>=24.1.0
Requires-Dist: aiohttp>=3.12.14
Requires-Dist: anyio==4.11.0
Requires-Dist: click>=8.1.0
Requires-Dist: fastapi>=0.116.1
Requires-Dist: filelock>=3.18.0
Requires-Dist: httpx>=0.27.0
Requires-Dist: huggingface-hub>=1.8.0
Requires-Dist: hypercorn>=0.18.0
Requires-Dist: loguru>=0.7.3
Requires-Dist: msgspec>=0.19.0
Requires-Dist: openai-harmony>=0.0.8
Requires-Dist: psutil>=6.0.0
Requires-Dist: pydantic>=2.11.7
Requires-Dist: python-multipart>=0.0.21
Requires-Dist: rich>=13.0.0
Requires-Dist: rustworkx>=0.17.1
Requires-Dist: tiktoken>=0.12.0
Requires-Dist: tomlkit>=0.14.0
Requires-Dist: transformers>=5.6.2
Requires-Dist: wmi>=1.5.1; sys_platform == 'win32'
Requires-Dist: zstandard>=0.23.0
Provides-Extra: torch
Requires-Dist: torch>=2.0.0; extra == 'torch'
Description-Content-Type: text/markdown

# exowindows

Fast, distributed ML scaling on Windows. Seamlessly connect machines via Ethernet, USB-C, or Thunderbolt, and run distributed PyTorch training/inference and Ollama LLMs with automatic hardware detection, heterogeneous partitioning, and speedup alerts.

## Installation

```bash
pip install exowindows
```

## Features

- **Distributed Compute & RAM speed filtering**: Automatically runs across local and remote GPUs, CPU, and RAM, with a 3200MHz RAM speed constraint to filter out slow devices.
- **Ollama CLI Integration**: Run `exowindows ollama run llama3.1` to scan for connected worker machines, see performance suggestions, and run distributed inference.
- **PyTorch Hook**: Integrates with PyTorch training to auto-detect optimal cluster nodes and show speedup recommendations.
- **Easy Worker Join**: Run `exowindows-node join` to connect workers.
