Add oneCCL backend for idist

描述

貢獻者指南

技術棧

pythonpytorch

領域

backendmachine learning

議題類型

feature

難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。

3

預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。

over 1 week

活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。

stale

清晰度議題是否清楚說明預期改動、驗收標準和下一步。

mostly clear

前置要求

PythonPyTorchDistributed training basicsoneCCL knowledge

新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。

20

研究方向

The issue requests adding a oneCCL backend by extending the torch native distributed model in `ignite/distributed/comp models/native.py`. Investigate the current implementation and the references to Intel's torch ccl and PyTorch with oneCCL backend. The maintainers have not yet provided implementation guidance, so a contributor should start by understanding the architecture of the existing comp models and how to integrate a new backend similar to the NCCL or Gloo backends. Check the linked references for details on oneCCL integration with PyTorch.