| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
feat(sapp-nd): add pp_modeling with YAML-driven ILP optimization and import path cleanup - Add pp_modeling module with YAML-driven ILP load balancing, per-stage recompute support, and pipeline visualization - Fix 2 bugs in sapp_ppb_adapter: _parse_yaml_offset() extra arg (TypeError), and early return skipping recompute parsing when offset=0 - Unify all sapp_ppb imports to long-path form (hyper_parallel.auto_parallel.sapp_ppb), eliminating dual try/except strategy and removing 6 pylint E0401 import-error warnings - Reduce pp_modeling tests from ~6600→3019 lines by removing 7 redundant test files, 9 unused YAML fixtures, and trimming 8 remaining files - Refactor 5 high-CCN functions below CCN-15 threshold - Add docstrings and type hints to 5 test files - Clean up CI exemption rules in filter_pylint.txt and whitelizard.txt | 2 个月前 | |
fix: 修复静态分析告警，补充模块文档与返回值校验 | 3 个月前 | |
feat(dmodule): add M1 declarative Configurable and Module layer Introduce Configurable in hyper_parallel/config/ and declarative Module, BaseModel, ShardingConfig, and ModelSpec under hyper_parallel.dmodule. ModelSpec supports Config-based registration alongside legacy build_model_fn. Module.parallelize applies DTensor sharding on a single DeviceMesh; local_map is deferred to M9. Expand Google-style docstrings with usage examples; add resolve_placements unit tests (case sensitivity, empty mesh, Partial). Add tests/ut/dmodule and dist_module parallelize coverage without wiring the new path into LLMTrainer or package __init__ exports. | 4 个月前 | |
fix: handle equivalent meshes and empty AGM outputs | 2 个月前 | |
DSV4 CSA/HCA DFunction | 2 个月前 | |
feat: support Qwen3.5 and Qwen3-VL-MoE training Add dense, MoE, and vision-language model integration with checkpoint conversion, data loading, and unified trainer execution. Correct TP, CP, EP, PP, FSDP, and HSDP synchronization semantics so supported BF16 single-card and hybrid-parallel configurations follow the same training behavior. | 3 个月前 | |
fix: clean up static code scan alerts for assigned owners Resolve H2305/C0411/W1805/H2205 findings in examples, dmodule, fully_shard, fast-tuner, and config modules; suppress C0413 where HYPER_PARALLEL_PLATFORM must be set before hyper_parallel imports. | 3 个月前 | |
Update generate checks Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> | 3 个月前 | |
| 3 个月前 | ||
feat: add fused GDN backend for linear attention CP | 2 个月前 | |
!1134 merge fix_pack_to_cpu into master fix_pack_to_cpu Created-by: silkage_jiajia Commit-by: 宋佳琪 Merged-by: liuchongming74 Description: <!-- Thanks for sending a pull request! Here are some tips for you: 1) If this is your first time, please read our contributor guidelines: https://gitee.com/mindspore/mindspore/blob/master/CONTRIBUTING.md 2) If you want to contribute your code but don't know who will review and merge, please add label mindspore-assistant to the pull request, we will find and do it as soon as possible. --> **What type of PR is this?** <!-- 选择下面一种标签替换下方 /kind <label>，可选标签类型有： - /kind bug - /kind task - /kind feature - /kind refactor - /kind clean_code 如PR描述不符合规范，修改PR描述后需要/check-pr重新检查PR规范。 --> /kind bug ---- **What does this PR do / why do we need it**: 激活换出过程中，AsyncSaveOnCpu.pack_to_cpu 原先直接返回输入张量。即使 unpack_from_cpu 清空 swap storage，hook 保存的对象仍可能持有原张量，导致原张量无法及时释放并增加训练过程中的内存占用。 本次修改让 pack_to_cpu 返回与原张量对象分离、但共享底层数据的对象： - hyper_parallel/platform/torch/activation_checkpoint/activation_swap.py 使用 tensor.detach() 作为 packed 数据。 - hyper_parallel/platform/mindspore/activation_checkpoint/activation_swap.py 使用 tensor.data 作为 packed 数据。 - 覆盖无需换出、策略要求保存、当前未注册 swap group 以及正常换出等返回路径。 - 不改变 swap storage 管理、SwapTensor 注册逻辑、对外 API 和调用方式。 - 新增两个平台的回归测试，验证清理 swap storage 后原张量可以被回收，packed 数据仍然有效。 本次共修改 4 个文件，新增 71 行、删除 11 行。 ---- **Which issue(s) this PR fixes**: <!-- 注意：在下方 #号 后仅输入issue编号或SR/AR编号，粘贴issue链接无效。 如PR描述不符合规范，修改PR描述后需要/check-pr重新检查PR规范。 --> Fixes #https://gitcode.com/mindspore/org-issues/issues/43145 ---- **Test Plan and Test result：What scenarios were tested, and what were the verification results（Function, performance, reliability, etc.）**： 执行： python -m pytest -q tests/ut/platform/torch/activation_checkpoint/test_activation_swap.py tests/ut/platform/mindspore/activation_checkpoint/test_activation_swap.py 结果：15 passed, 1 skipped in 0.25s。 验证内容包括： - packed tensor 与原始张量不是同一对象。 - 清理 swap storage 后原始张量可以被垃圾回收。 - packed 数据仍保持有效，并能够正常解包。 - 两个平台的激活换出相关回归测试通过。 ---- **Self-checklist**：（**请自检，在[ ]内打上x，我们将检视你的完成情况，否则会导致pr无法合入**） - [x] **设计**：PR对应的方案已经经过Maintainer评审，方案检视意见均已答复并完成方案修改 - [x] **测试**：PR中的代码已有UT测试用例进行覆盖，新增测试用例已随本PR一并提交 - [x] **验证**：PR描述已包含本次Bugfix预期目标的详细验证结果 - [ ] **接口**：是否涉及对外接口变更，相应变更已得到接口评审组织的通过，API对应的注释信息已经刷新正确 - [ ] **文档**：是否涉及官网文档修改，如果涉及请及时提交资料到Doc仓 <!-- **Special notes for your reviewers**: --> <!-- - [ ] 是否导致无法前向兼容 --> <!-- - [ ] 是否涉及依赖的三方库变更 --> See merge request: mindspore/hyper-parallel!1134 | 2 个月前 | |
fix: improve memory visualizer with large file support and UI polish | 2 个月前 | |
feat: add fused GDN backend for linear attention CP | 2 个月前 | |
feat(dtensor): support ragged shard end to end | 2 个月前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 2 个月前 | ||
| 3 个月前 | ||
| 4 个月前 | ||
| 2 个月前 | ||
| 2 个月前 | ||
| 3 个月前 | ||
| 3 个月前 | ||
| 3 个月前 | ||
| 3 个月前 | ||
| 2 个月前 | ||
| 2 个月前 | ||
| 2 个月前 | ||
| 2 个月前 | ||
| 2 个月前 |