Back to Details

triaging-issues

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

PyTorch Issue Triage Skill

PyTorch问题分流技能

This skill helps triage GitHub issues by routing issues, applying labels, and leaving first-line responses.
本技能可通过问题分配、添加标签及留下一线回复来协助完成GitHub问题分流。

Contents

目录

  • MCP Tools Available
  • Labels You Must NEVER Add
  • Issue Triage Steps
    • Step 0: Already Routed — SKIP
    • Step 1: Question vs Bug/Feature
    • Step 1.5: Needs Reproduction — External Files
    • Step 2: Transfer
    • Step 2.5: PT2 Issues — Special Handling
    • Step 3: Redirect to Secondary Oncall
    • Step 4: Label the Issue
    • Step 5: High Priority — REQUIRES HUMAN REVIEW
    • Step 6: bot-triaged (automatic)
    • Step 7: Mark Triaged
  • V1 Constraints
Labels reference: See labels.json for the full catalog of 305 labels suitable for triage. ONLY apply labels that exist in this file. Do not invent or guess label names. This file excludes CI triggers, test configs, release notes, and deprecated labels.
PT2 triage guide: See pt2-triage-rubric.md for detailed labeling guidance when triaging PT2/torch.compile issues.
Response templates: See templates.json for standard response messages.
  • 可用的MCP工具
  • 绝对不能添加的标签
  • 问题分流步骤
    • 步骤0：已分配——跳过
    • 步骤1：区分咨询类问题与Bug/功能请求
    • 步骤1.5：需要复现——外部文件处理
    • 步骤2：转移问题
    • 步骤2.5：PT2问题——特殊处理
    • 步骤3：转至二级值班团队
    • 步骤4：为问题添加标签
    • 步骤5：高优先级——需人工审核
    • 步骤6：bot-triaged（自动添加）
    • 步骤7：标记为已分流
  • V1版本约束
标签参考： 完整的305个适用于分流的标签目录请查看labels.json。仅允许添加此文件中存在的标签，不得自创或猜测标签名称。该文件排除了CI触发标签、测试配置标签、发布说明标签及已废弃标签。
PT2分流指南： 处理PT2/torch.compile相关问题时，详细的标签添加指导请查看pt2-triage-rubric.md。
回复模板： 标准回复消息请查看templates.json。

MCP Tools Available

可用的MCP工具

Use these GitHub MCP tools for triage:
ToolPurpose
mcp__github__issue_read
Get issue details, comments, and existing labels
mcp__github__issue_write
Apply labels or close issues
mcp__github__add_issue_comment
Add comment (only for redirecting questions)
mcp__github__search_issues
Find similar issues for context
分流时可使用以下GitHub MCP工具：
工具用途
mcp__github__issue_read
获取问题详情、评论及现有标签
mcp__github__issue_write
添加标签或关闭问题
mcp__github__add_issue_comment
添加评论（仅用于引导咨询类问题时）
mcp__github__search_issues
查找相似问题以获取上下文信息

Labels You Must NEVER Add

绝对不能添加的标签

Prefix/CategoryReason
Labels not in 
labels.json
Only apply labels that exist in the allowlist
ciflow/*
CI job triggers for PRs only
test-config/*
Test suite selectors for PRs only
release notes: *
Auto-assigned for release notes
ci-*
, 
ci:*
CI infrastructure controls
sev*
Severity labels require human decision
merge blocking
Requires human decision
Any label containing "deprecated"Obsolete
If blocked: When a label is blocked by the hook, add ONLY 
triage review
and stop. A human will handle it.
These rules are enforced by a PreToolUse hook that validates all labels against 
labels.json
.
前缀/类别原因
不在
labels.json
中的标签		仅允许添加白名单中的标签
ciflow/*
仅用于PR的CI任务触发标签
test-config/*
仅用于PR的测试套件选择器标签
release notes: *
用于发布说明的自动分配标签
ci-*
, 
ci:*
CI基础设施控制标签
sev*
严重程度标签需人工决策
merge blocking
需人工决策
任何包含“deprecated”的标签已废弃
若操作被阻止： 当标签被钩子阻止时，仅添加
triage review
标签并停止操作，由人工处理。
这些规则由PreToolUse钩子强制执行，该钩子会根据
labels.json
验证所有标签。

Issue Triage (for each issue)

问题分流（单个问题）

0) Already Routed — SKIP

0) 已分配——跳过

If an issue already has ANY 
oncall:
label, SKIP IT entirely. Do not:
  • Add any labels
  • Add 
    triaged
  • Leave comments
  • Do any triage work
That issue belongs to the sub-oncall team. They own their queue.
若问题已包含任意
oncall:
标签，完全跳过该问题，不得：
  • 添加任何标签
  • 添加
    triaged
    标签
  • 留下评论
  • 执行任何分流操作
该问题属于对应二级值班团队的处理范畴，由他们负责自己的队列。

1) Question vs Bug/Feature

1) 区分咨询类问题与Bug/功能请求

  • If it is a question (not a bug report or feature request): close and use the 
    redirect_to_forum
    template from 
    templates.json
    .
  • If unclear whether it is a bug/feature vs a question: request additional information using the 
    request_more_info
    template and stop.
  • 若为咨询类问题（非Bug报告或功能请求）：关闭问题，并使用
    templates.json
    中的
    redirect_to_forum
    模板回复。
  • 若无法明确区分是Bug/功能请求还是咨询类问题：使用
    request_more_info
    模板请求更多信息，然后停止操作。

1.5) Needs Reproduction — External Files

1.5) 需要复现——外部文件处理

Check if the issue body contains links to external files that users would need to download to reproduce.
Patterns to detect:
  • File attachments: 
    .zip
    , 
    .pt
    , 
    .pth
    , 
    .pkl
    , 
    .safetensors
    , 
    .onnx
    , 
    .bin
    files
  • External storage: Google Drive, Dropbox, OneDrive, Mega, WeTransfer links
  • Model hubs: Hugging Face Hub links to model files
Action:
  1. Edit the issue body to remove/redact the download links
    • Replace with: 
      [Link removed - external file downloads are not permitted for security reasons]
  2. Add 
    needs reproduction
    label
  3. Use the 
    needs_reproduction
    template from 
    templates.json
    to request a self-contained reproduction
  4. Do NOT add 
    triaged
    — wait for the user to provide a reproducible example
检查问题描述中是否包含用户需下载才能复现问题的外部文件链接。
需识别的模式：
  • 文件附件：
    .zip
    , 
    .pt
    , 
    .pth
    , 
    .pkl
    , 
    .safetensors
    , 
    .onnx
    , 
    .bin
    格式文件
  • 外部存储：Google Drive、Dropbox、OneDrive、Mega、WeTransfer链接
  • 模型仓库：Hugging Face Hub中的模型文件链接
操作步骤：
  1. 编辑问题描述以移除/遮蔽下载链接
    • 替换为：
      [链接已移除 - 出于安全考虑，不允许外部文件下载]
  2. 添加
    needs reproduction
    标签
  3. 使用
    templates.json
    中的
    needs_reproduction
    模板请求用户提供可独立复现的示例
  4. 不得添加
    triaged
    标签——等待用户提供可复现的示例

1.6) Edge Cases & Numerical Accuracy

1.6) 边缘情况与数值精度问题

If the issue involves extremal values or numerical precision differences:
Patterns to detect:
  • Values near 
    torch.finfo(dtype).max
    or 
    torch.finfo(dtype).min
  • NaN/Inf appearing in outputs from valid (but extreme) inputs
  • Differences between CPU and GPU results
  • Precision differences between dtypes (e.g., fp32 vs fp16)
  • Fuzzer-generated edge cases
Action:
  1. Add 
    module: edge cases
    label
  2. If from a fuzzer, also add 
    topic: fuzzer
  3. Use the 
    numerical_accuracy
    template from 
    templates.json
    to link to the docs
  4. If the issue is clearly expected behavior per the docs, close it with the template comment
若问题涉及极值或数值精度差异：
需识别的模式：
  • 数值接近
    torch.finfo(dtype).max
    或
    torch.finfo(dtype).min
  • 有效（但极端）输入产生NaN/Inf输出
  • CPU与GPU结果存在差异
  • 不同数据类型间的精度差异（如fp32 vs fp16）
  • 模糊测试生成的边缘情况
操作步骤：
  1. 添加
    module: edge cases
    标签
  2. 若来自模糊测试工具，同时添加
    topic: fuzzer
    标签
  3. 使用
    templates.json
    中的
    numerical_accuracy
    模板链接至相关文档
  4. 若根据文档判断该问题属于预期行为，使用模板回复并关闭问题

2) Transfer (domain library or ExecuTorch)

2) 转移问题（至领域库或ExecuTorch）

If the issue belongs in another repo (vision/text/audio/RL/ExecuTorch/etc.), transfer the issue and STOP.
若问题属于其他仓库（视觉/文本/音频/RL/ExecuTorch等），转移问题并停止后续操作。

2.5) PT2 Issues — Special Handling

2.5) PT2问题——特殊处理

When triaging PT2 issues (torch.compile, dynamo, inductor), see pt2-triage-rubric.md for detailed labeling decisions.
Key differences from general triage:
  • For PT2 issues, you MAY apply 
    module:
    labels (e.g., 
    module: dynamo
    , 
    module: inductor
    , 
    module: dynamic shapes
    )
  • Use the rubric to determine the correct component labels
  • Only redirect to 
    oncall: cpu inductor
    for MKLDNN-specific issues; otherwise keep in PT2 queue
处理PT2问题（torch.compile、dynamo、inductor）时，请查看pt2-triage-rubric.md获取详细的标签添加指导。
与通用分流的主要区别：
  • 对于PT2问题，可添加
    module:
    标签（如
    module: dynamo
    , 
    module: inductor
    , 
    module: dynamic shapes
    ）
  • 根据指南确定正确的组件标签
  • 仅针对MKLDNN相关问题转至
    oncall: cpu inductor
    ，其余PT2问题留在PT2队列中

3) Redirect to Secondary Oncall

3) 转至二级值班团队

CRITICAL: When redirecting issues to an oncall queue (critical with the exception of PT2), apply exactly one 
oncall: ...
label and STOP. Do NOT:
  • Add any 
    module:
    labels
  • Mark it 
    triaged
  • Do any further triage work
The sub-oncall team will handle their own triage. Your job is only to route it to them.
关键注意事项： 将问题转至值班队列时（PT2问题除外，此规则同样关键），仅添加一个
oncall: ...
标签并停止后续操作。不得：
  • 添加任何
    module:
    标签
  • 标记为
    triaged
  • 执行任何其他分流操作
二级值班团队会自行处理后续分流，你的任务仅为将问题路由至对应团队。

Oncall Redirect Labels

值班团队分流标签

LabelWhen to use
oncall: jit
TorchScript issues
oncall: distributed
Distributed training (DDP, FSDP, RPC, c10d, DTensor, DeviceMesh, symmetric memory, context parallel, pipelining)
oncall: export
torch.export issues
oncall: quantization
Quantization issues
oncall: mobile
Mobile (iOS/Android), excludes ExecuTorch
oncall: profiler
Profiler issues (CPU, GPU, Kineto)
oncall: visualization
TensorBoard integration
Note: 
oncall: cpu inductor
is a sub-queue of PT2. For general triage, just use 
oncall: pt2
.
标签使用场景
oncall: jit
TorchScript相关问题
oncall: distributed
分布式训练相关问题（DDP、FSDP、RPC、c10d、DTensor、DeviceMesh、对称内存、上下文并行、流水线）
oncall: export
torch.export相关问题
oncall: quantization
量化相关问题
oncall: mobile
移动端（iOS/Android）相关问题，不含ExecuTorch
oncall: profiler
性能分析工具相关问题（CPU、GPU、Kineto）
oncall: visualization
TensorBoard集成相关问题
注意： 
oncall: cpu inductor
是PT2的子队列。对于通用分流，仅使用
oncall: pt2
即可。

4) Label the issue (if NOT transferred/redirected)

4) 为问题添加标签（若未转移/分流）

Only if the issue stays in the general queue:
  • Add 1+ 
    module: ...
    labels based on the affected area
  • If feature request: add 
    feature
    (or 
    function request
    for a new function or new arguments/modes)
  • If small improvement: add 
    enhancement
仅当问题留在通用队列时：
  • 根据受影响的领域添加1个或多个
    module: ...
    标签
  • 若为功能请求：添加
    feature
    标签（若为新函数或新参数/模式的请求，添加
    function request
    标签）
  • 若为小改进：添加
    enhancement
    标签

5) High Priority — REQUIRES HUMAN REVIEW

5) 高优先级——需人工审核

CRITICAL: If you believe an issue is high priority, you MUST:
  1. Add 
    triage review
    label and do not add 
    triaged
Do NOT directly add 
high priority
without human confirmation.
High priority criteria:
  • Crash / segfault / illegal memory access
  • Silent correctness issue (wrong results without error)
  • Regression from a prior version
  • Internal assert failure
  • Many users affected
  • Core component or popular model impact
关键注意事项： 若认为问题为高优先级，必须：
  1. 添加
    triage review
    标签，不得添加
    triaged
    标签
不得在未经人工确认的情况下直接添加
high priority
标签。
高优先级判定标准：
  • 崩溃/段错误/非法内存访问
  • 静默正确性问题（无错误提示但结果错误）
  • 相较于之前版本出现回归
  • 内部断言失败
  • 影响大量用户
  • 核心组件或热门模型受影响

6) bot-triaged (automatic)

6) bot-triaged（自动添加）

The 
bot-triaged
label is automatically applied by a post-hook after any issue mutation. You do not need to add it manually.
bot-triaged
标签会在问题发生任何变更后由后置钩子自动添加，无需手动添加。

7) Mark triaged

7) 标记为已分流

If not transferred/redirected and not flagged for review, add 
triaged
.
若问题未被转移/分流，也未标记为需要审核，添加
triaged
标签。

V1 Constraints

V1版本约束

DO NOT:
  • Close bug reports or feature requests automatically
  • Close issues unless they are clear usage questions per Step 1
  • Assign issues to users
  • Add 
    high priority
    directly without human confirmation
  • Add module labels when redirecting to oncall
  • Add comments to bug reports or feature requests, except a single info request when classification is unclear
DO:
  • Close clear usage questions and point to discuss.pytorch.org (per step 1)
  • Be conservative - when in doubt, add 
    triage review
    for human attention
  • Apply type labels (
    feature
    , 
    enhancement
    , 
    function request
    ) when confident
  • Add 
    triaged
    label when classification is complete
Note: 
bot-triaged
is automatically applied by a post-hook after any issue mutation.
禁止操作：
  • 自动关闭Bug报告或功能请求
  • 除非是步骤1中明确的使用类咨询问题，否则不得关闭问题
  • 将问题分配给用户
  • 未经人工确认直接添加
    high priority
    标签
  • 分流至值班团队时添加module标签
  • 为Bug报告或功能请求添加评论，除非在分类不明确时发送一次信息请求
允许操作：
  • 关闭明确的使用类咨询问题，并引导至discuss.pytorch.org（按照步骤1）
  • 保持谨慎——若存在疑问，添加
    triage review
    标签交由人工处理
  • 确认分类后添加类型标签（
    feature
    , 
    enhancement
    , 
    function request
    ）
  • 分类完成后添加
    triaged
    标签
注意： 
bot-triaged
标签会在问题发生任何变更后由后置钩子自动添加。