platform-dev

name	platform-dev
description	Develop HyperParallel platform abstraction: new Platform APIs, torch/mindspore backends, FSDP/HSDP/Pipeline/Activation Checkpoint, DTensorBase, collectives. Use when changing platform/ or cross-backend features. Not for shard op YAML/impl (use dist-op-dev) or git/CI (autogit / gate-doctor).

Platform Development

Guides cross-platform work under platform/. This file is the index — load the matching workflow / reference before that step.

When to use

• New method on platform/platform.py, or impl in platform/torch|mindspore/
• FSDP / HSDP / Pipeline / Activation Checkpoint platform code
• DTensorBase extensions; collectives; stream sync / memory lifecycle patterns

Execution (autonomous checklist)

1. workflows/01-scope-analysis.md
2. workflows/02-base-class-api.md
3. workflows/03-backend-implementation.md
4. workflows/04-cross-platform-verification.md
5. workflows/05-testing.md
6. workflows/06-commit.md → autogit

Tree, API map, patterns: references/architecture.md, references/quick-reference.md, references/decisions.md.

Hard rules

1. No bare import torch / mindspore in platform-agnostic code — get_platform()
2. New Platform APIs land in base class first (platform/platform.py)
3. Consider both backends — implement or NotImplementedError
4. Type gaps: torch device/ProcessGroup vs mindspore str groups — check signatures
5. Lazy framework imports inside methods under platform/torch|mindspore/ + # pylint: disable=C0415
6. Stream/memory: handle.wait(), event.record→wait, resize_(0) — see .agent/rules/distributed.md

Out of scope

• Distributed op YAML / parallel_*.py → dist-op-dev
• Commit / PR / CI → autogit / gate-doctor
• Full distributed audit after change → code-review