[dist] XPU distributed backend selection (XCCL / oneCCL) is unverified — needs XPU hardware
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 28/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
调研方向
Start with get_default_dist_backend() and tests/test_distributed_device_setup.py, which the issue identifies as the implementation and test entry points. The issue requires an exact-head, two-rank torchrun on Intel XPU hardware, checking device binding, all_reduce, DDP forward and backward, and end-of-epoch cache release. Done means the XPU backend selection and test are updated, with the PyTorch, oneAPI, and device versions recorded.
由索引模型根据 Issue 内容生成。
描述
Context
#3771 binds distributed training to the active accelerator instead of assuming CUDA. Writing that change surfaced a second, independent defect on XPU.
get_default_dist_backend() returns "nccl" for every non-NPU device. Since get_accelerator_device_type() does select XPU, an XPU run reaches init_process_group(backend="nccl") and cannot start at all on a build without NCCL — the same class of failure #3771 fixes for Ascend NPU.
Why this is not in #3771
The correct backend for XPU is XCCL (oneCCL; the legacy name on older builds is "ccl"), so the fix is small:
_ACCELERATOR_DIST_BACKENDS = {"npu": ("hccl",), "xpu": ("xccl", "ccl")}
but it advertises a runtime path that nobody here can exercise. We have no Intel XPU hardware in our environment, so under the repository's hardware-validation requirement the XPU arm was the only path #3771 changed without real-device evidence. #3771 therefore keeps the historical "nccl" value for XPU — i.e. exactly the pre-PR behaviour, unchanged — and the selection is tracked here for whoever has the hardware.
What a fix needs
- Add the XPU entry above (keep
"ccl"only for builds that positively confirm it throughtorch.distributed.is_ccl_available()). - Run an exact-head 2-rank
torchrunon Intel XPU with XCCL, showing per-rankxpu:0/xpu:1binding, anall_reduce, DDP forward/backward, the end-of-epoch cache release, plus the exact PyTorch / oneAPI / device versions. - Update
tests/test_distributed_device_setup.pyaccordingly — the current test pins the narrowed behaviour (XPU stays"nccl"even when a build reportsis_xccl_available()), so it needs to be flipped.
Related
- #3771 — the Ascend NPU path, verified end to end on 910B2 hardware.
- Failure mode this follows up on:
init_process_group(backend="nccl")->RuntimeError: Distributed package doesn't have NCCL built in.
- 主要语言
- Python
- 星标
- 20.8k
- 派生
- 2.1k
- 平均合并
- 1 天 46 分钟
- 30 天内合并 PR
- 36
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
modelscope/FunASR 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
modelscope/FunASR#3780 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
modelscope/FunASR#3762 · 1 条评论 ·
维护者通常 1 天内回复
-
Bug: blank lines in wav.scp / jsonl file lists crash prepare_data_iterator (IndexError / JSONDecodeError)可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭needs feedback
难度 2/5 1-3 小时 新手友好度 85/100
modelscope/FunASR#3757 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 88/100
modelscope/FunASR#3730 · 2 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 86/100
modelscope/FunASR#3704 · 1 条评论 ·
维护者通常 1 天内回复
查看 modelscope/FunASR 的全部 Issue
相似的 Issue
-
good first issue tools
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
bug deferred domain:ci P1
难度 2/5 1-3 小时 新手友好度 76/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 68/100
docling-project/docling-mcp#147 ·
-
issue: A pipe model's Valves button ignores the Allow Chat Valves permission可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭bug confirmed issue
难度 2/5 1-3 小时 新手友好度 72/100
open-webui/open-webui#32771 · 1 条评论 ·
维护者通常 1 天内回复
-
Scheduled Security Scan is failing on main可能已有人在做 @pkupt 今天认领。 未关闭security-scan-failure
难度 1/5 1-3 小时 新手友好度 74/100
semantica-agi/semantica#2045 ·
维护者通常 2 天内回复