昇腾亲和的多模态加速系列套件,现已支持vLLM Omni,Diffusers+CacheDit等框架。
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
[Docs]补充安装指引、测试说明、模型权重获取及版本信息 Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !336 merge doc_yiyongxing into dev [Docs]补充安装指引、测试说明、模型权重获取及版本信息 Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #172 # Purpose 本次修改解决以下文档问题,改善开源社区新用户体验: 1. **README 增加安装入口**:README.md 和 README.en.md 新增「安装指南」章节,链接至 installation.md / installing_guide.md。 2. **测试指引完善**:中英文 test.md 增加「测试环境说明」章节,说明 CPU 兼容测试与 NPU 依赖测试的分类,补充 run_UT_test.sh 和 run_test.sh 三种模式(--all / --cpu_only / --npu_only)的完整使用方法。 3. **模型权重获取指引**:中英文 quick_start.md 将原来的单段「模型下载与运行」拆分为「获取推理脚本」「获取模型权重」「运行推理」三步,明确说明仓库不含权重,给出 HuggingFace 权重下载链接和期望目录结构。 4. **CANN 安装指引**:中文 installation.md 新增「CANN 安装」小节,给出官方安装指南链接和环境变量设置命令。 5. **依赖版本更新**: - CANN 版本统一更新为 9.0.0(涉及 installation.md、release_note.md、Dockerfile、docker/README.md、installing_guide.md、SKILL.md) - torch/torch_npu 版本列表增加 2.10 - 新增部署依赖 triton==3.5.0 和 triton-ascend==3.2.1(requirements.txt、installation.md、installing_guide.md、SKILL.md) # Test Plan - 文档内容为纯文本修改,无代码逻辑变更 - 建议 markdownlint 检查通过:pre-commit run markdownlint --all-files --hook-stage manual - 建议验证文档站点可正常构建: ```bash python -m pip install -r docs/requirements-docs.txt python docs/build_docs.py See merge request: Ascend/MindIE-SD!336 | 1 个月前 | |
[Docs] refresh lint and dependency config, docker, docs Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !243 merge dev into dev [Docs] refresh lint and dependency config, docker, docs Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #72 # Purpose This PR delivers five grouped changes: 1. add a 910B AArch64 docker build environment; 2. refresh repository lint/dependency inputs; 3. reorganize developer documentation and navigation; 4. align homepage, manuals, and tech report entrypoints; 5. consolidate governance skill routing and require proactive commit/PR draft output. # Test Plan 1. Review the new docker build files and README. 2. Review repository dependency and pre-commit configuration changes. 3. Review Chinese and English developer guide updates and navigation entrypoints. 4. Review homepage/manual/report entry changes. 5. Review AGENTS.md and .agents/skills/mindie-sd-community-governance/SKILL.md for routing and PR-drafting requirements. # Test Report  See merge request: Ascend/MindIE-SD!243 | 3 个月前 | |
feat: implement basic dev container capabilities Co-authored-by: hejianping<hejianping7@huawei.com> # message auto-generated for no-merge-commit merge: !235 merge build/devcontainer into dev feat: implement basic dev container capabilities Created-by: hejianping Commit-by: hejianping Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fixes [#69](https://gitcode.com/Ascend/MindIE-SD/issues/69) # Purpose 在社区提供DevContainer核心能力,让 1.环境标准化:把“我本地能跑”变成“团队都能跑” 2 多角色/多场景配置:一个仓库不必只有一种环境 3 Features:把工具安装标准化、模块化 4 Templates:把最佳实践打包成开箱模板 5 生命周期脚本:让环境初始化自动完成 6 预构建能力:把“第一次进入很慢”变成“秒级可用” 7 CLI:把 IDE 体验变成流水线能力 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request 在wsl下测试执行UT成功,信息如下: INFO:root:creating build/lib/mindiesd/ops/vendors/customize INFO:root:creating build/lib/mindiesd/ops/vendors/customize/framework INFO:root:creating build/lib/mindiesd/ops/vendors/customize/framework/onnx INFO:root:copying mindiesd/ops/vendors/customize/framework/onnx/libcust_onnx_parsers.so -> build/lib/mindiesd/ops/vendors/customize/framework/onnx INFO:root:creating build/lib/mindiesd/ops/vendors/customize/framework/tensorflow INFO:root:copying mindiesd/ops/vendors/customize/framework/tensorflow/npu_supported_ops.json -> build/lib/mindiesd/ops/vendors/customize/framework/tensorflow INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/config INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/config/ascend310p INFO:root:copying mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/config/ascend310p/aic-ascend310p-ops-info.json -> build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/config/ascend310p INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl/constants.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl/flash_attention_fwd.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl/flash_attention_tik.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl/tik_ops_utils.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/ai_core/tbe/customize_impl INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl/constants.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl/flash_attention_fwd.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl/flash_attention_tik.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl INFO:root:copying mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl/tik_ops_utils.py -> build/lib/mindiesd/ops/vendors/customize/op_impl/vector_core/tbe/customize_impl INFO:root:creating build/lib/mindiesd/ops/vendors/customize/op_proto INFO:root:copying mindiesd/ops/vendors/customize/op_proto/libcust_op_proto.so -> build/lib/mindiesd/ops/vendors/customize/op_proto INFO:root:copying mindiesd/plugin/libPTAExtensionOPS.so -> build/lib/mindiesd/plugin WARNING:root:warning: CustomBuildPy: byte-compiling is disabled, skipping. root@00464f0fc4a3:/workspace/MindIE-SD# bash tests/run_UT_test.sh ........2026-04-08 20:11:51.462+08:00 CST [10710] [138009548819328] [MindIE-SD] [ERROR] exception.py:26: [MIE06E000002] Config parameter err. Test config exception! .2026-04-08 20:11:51.462+08:00 CST [10710] [138009548819328] [MindIE-SD] [ERROR] exception.py:47: [MIE06E000005] Model exec err. Test model exec exception! .2026-04-08 20:11:51.462+08:00 CST [10710] [138009548819328] [MindIE-SD] [ERROR] exception.py:40: [MIE06E000004] Model init err. Test model init exception! .2026-04-08 20:11:51.463+08:00 CST [10710] [138009548819328] [MindIE-SD] [ERROR] exception.py:19: [MIE06E000001] Parameters invalid. Test parameter exception! .2026-04-08 20:11:51.463+08:00 CST [10710] [138009548819328] [MindIE-SD] [ERROR] exception.py:33: [MIE06E000003] Torch exec err. Test torch exception! ..... ---------------------------------------------------------------------- Ran 17 tests in 0.038s OK # Test Report See merge request: Ascend/MindIE-SD!235 | 3 个月前 | |
update: 更新文件 PULL_REQUEST_TEMPLATE.md Signed-off-by: limuyuan31 <limuyuan3@huawei.com> | 7 个月前 | |
[dev]同步最新代码 Co-authored-by: mazhixin00_00<mazhixin7@huawei.com> | 7 个月前 | |
[Feature][ops]Add mul_add fused operator support Co-authored-by: wangwei<daviwang2026@qq.com> # message auto-generated for no-merge-commit merge: !481 merge feature/mul_add_ops_dev into dev [Feature][ops]Add mul_add fused operator support Created-by: daviwang Commit-by: wangwei Merged-by: ascend-robot Description: # Purpose 新增 mul_add 融合算子,支持在昇腾 NPU 上执行融合乘加计算: - 算子接口:mindiesd::mul_add(a, b, c) -> y - 计算公式:y = a + b * c - 输入 shape:a/b = [batch, seq_len, hidden_size],c = [batch, 1, hidden_size] - 支持数据类型:FP16、BF16 - 支持计算平台:ascend910b、ascend910_93、ascend950 主要变更: 1. 在 csrc/ops/mul_add/ 下新增算子定义(OpDef)、shape/dtype 推导(proto)、tiling 计算及 AscendC kernel 实现。 2. 在 csrc/plugin/ 下新增 mul_add PTA plugin,并在 register_ops.cpp 中注册到 mindiesd torch 库。 3. 更新 csrc/CMakeLists.txt,将 mul_add.cpp 加入 PTAExtensionOPS 编译。 4. 更新 build/build_ops.sh 默认算子列表,确保默认构建包含 mul_add。 5. 在 tests/ops/mul_add/ 下新增单元测试,覆盖基础 shape、典型 LLM shape、小 shape 边界及 FP16 精度场景。 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request 1. 编译 AscendC 算子: bash cd build bash build_ops.sh 2. 编译 PTA plugin: bash cd build bash build_plugin.sh 3. 运行 mul_add 单元测试(需在 NPU 环境): bash cd tests/ops/mul_add python test_mul_add.py 4. 可选:通过环境变量指定 NPU 卡号: ``bash MINDIESD_TEST_NPU_ID=7 python test_mul_add.py # Test Report • 测试覆盖: • 基础小 shape:(1, 4, 128) • 典型 LLM shape:(1, 7200, 4608)、(1, 256, 4608)、(1, 7200, 3072)、(1, 256, 3072) • 多组小 shape 边界:(1, 1, 64) ~ (1, 64, 1024) • FP16 精度:(1, 256, 3072) • 参考实现:ref = a.float() + b.float() * c.float() • 精度阈值:rtol=1e-2, atol=1e-2 - 运行环境:CANN 9.1.T560, Ascend950PR NPU, Python 3.11, aarch64 - 所有新增测试用例均通过: - test_mul_add_basic PASSED - test_mul_add_typical PASSED - test_mul_add_small_shapes PASSED - test_mul_add_fp16` PASSED - 最大误差:BF16 < 1e-2,FP16 < 1e-2 See merge request: Ascend/MindIE-SD!481 | 2 天前 | |
[Feature][ops]Add mul_add fused operator support Co-authored-by: wangwei<daviwang2026@qq.com> # message auto-generated for no-merge-commit merge: !481 merge feature/mul_add_ops_dev into dev [Feature][ops]Add mul_add fused operator support Created-by: daviwang Commit-by: wangwei Merged-by: ascend-robot Description: # Purpose 新增 mul_add 融合算子,支持在昇腾 NPU 上执行融合乘加计算: - 算子接口:mindiesd::mul_add(a, b, c) -> y - 计算公式:y = a + b * c - 输入 shape:a/b = [batch, seq_len, hidden_size],c = [batch, 1, hidden_size] - 支持数据类型:FP16、BF16 - 支持计算平台:ascend910b、ascend910_93、ascend950 主要变更: 1. 在 csrc/ops/mul_add/ 下新增算子定义(OpDef)、shape/dtype 推导(proto)、tiling 计算及 AscendC kernel 实现。 2. 在 csrc/plugin/ 下新增 mul_add PTA plugin,并在 register_ops.cpp 中注册到 mindiesd torch 库。 3. 更新 csrc/CMakeLists.txt,将 mul_add.cpp 加入 PTAExtensionOPS 编译。 4. 更新 build/build_ops.sh 默认算子列表,确保默认构建包含 mul_add。 5. 在 tests/ops/mul_add/ 下新增单元测试,覆盖基础 shape、典型 LLM shape、小 shape 边界及 FP16 精度场景。 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request 1. 编译 AscendC 算子: bash cd build bash build_ops.sh 2. 编译 PTA plugin: bash cd build bash build_plugin.sh 3. 运行 mul_add 单元测试(需在 NPU 环境): bash cd tests/ops/mul_add python test_mul_add.py 4. 可选:通过环境变量指定 NPU 卡号: ``bash MINDIESD_TEST_NPU_ID=7 python test_mul_add.py # Test Report • 测试覆盖: • 基础小 shape:(1, 4, 128) • 典型 LLM shape:(1, 7200, 4608)、(1, 256, 4608)、(1, 7200, 3072)、(1, 256, 3072) • 多组小 shape 边界:(1, 1, 64) ~ (1, 64, 1024) • FP16 精度:(1, 256, 3072) • 参考实现:ref = a.float() + b.float() * c.float() • 精度阈值:rtol=1e-2, atol=1e-2 - 运行环境:CANN 9.1.T560, Ascend950PR NPU, Python 3.11, aarch64 - 所有新增测试用例均通过: - test_mul_add_basic PASSED - test_mul_add_typical PASSED - test_mul_add_small_shapes PASSED - test_mul_add_fp16` PASSED - 最大误差:BF16 < 1e-2,FP16 < 1e-2 See merge request: Ascend/MindIE-SD!481 | 2 天前 | |
[Docs]:修改文档资料 Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !451 merge docker_0713 into dev [Docs]:修改文档资料 Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: <!-- PR描述模板更新日期:20251224 --> # Which issue(s) this PR fixes or accomplishes > **如问题已解决,按照下方示例附上ISSUE单号 / Apply bug or request issue as follow if the solved**\ > Fixes #ISSUE ID\ > **Fixes关键字会自动关闭issue,如问题部分解决请不要使用Fixes,可以用下方标签替代\ > Fixes will automatically close issue, please use the following tag if only part of the issue is solved**\ > Fix part of #ISSUE ID # Purpose 1. 按照镜像发布规范修改tag命名 2. 修复CANN安装指导和链接中内容章节对不上的问题 3. 修复triton-ascend 3.2.1不在pypi源中导致的安装问题 4. 修复TTFHW扫出的缺少编译构建时的依赖包的问题 5. 统一文档内的链接在打开时是新建浏览器标签打开,不会覆盖原链接打开 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request # Test Report See merge request: Ascend/MindIE-SD!451 | 15 天前 | |
[Docs]: 新增联系我们章节并补充微信交流群二维码 Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !478 merge dev into dev [Docs]: 新增联系我们章节并补充微信交流群二维码 Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > <!-- TODO: 若有关联 issue 请补充编号,例如 Fixes #xxx --> # Purpose 在 README 的「快速入门」与「开发文档」之间新增「联系我们」章节,方便用户反馈与交流: - 引导通过 GitCode Issues 参与讨论(https://gitcode.com/Ascend/MindIE-SD/issues) - 提供微信交流群二维码(docs/figures/wx_sd.jpg) - 说明群二维码失效时可通过个人微信(docs/figures/wx_lwl.jpg)邀请入群 # Test Plan 1. 在 GitCode 仓库页面预览 README,确认「联系我们」章节位置正确(位于「快速入门」之后、「开发文档」之前)。 2. 点击两个二维码图片,确认能正常显示且路径正确(docs/figures/wx_sd.jpg、docs/figures/wx_lwl.jpg)。 3. 点击 GitCode Issues 链接,确认跳转有效。 4. 执行 markdownlint 专项检查。 # Test Report - markdownlint(manual stage)对 README.md 检查:本次新增内容(42–54 行)无违规。 - 注:README.md:66–67 报 MD004(「生态支持」使用 + 列表),为既有问题,不在本次改动范围,未改动。 - 两张二维码图片均已存在于 docs/figures/。 - pre-commit:trim trailing whitespace / fix end of files / codespell / typos 均 Passed;Gitleaks 因本地缺少 ./gitleaks 二进制未跑(本改动无任何密钥)。 # Purpose # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request # Test Report See merge request: Ascend/MindIE-SD!478 | 6 天前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
[Bugfix] Resolve 4 UT errors under torch 2.10 environment Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !456 merge hitest_0716 into dev [Bugfix] Resolve 4 UT errors under torch 2.10 environment Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fixes #<229> # Purpose 修复 torch 2.10.x 环境下全量 UT 的 4 个 error(Ran 590 tests ... FAILED errors=4)。 1. mindiesd/layers/register_ops.py:114 将 torch.__version__.startswith("2.1") 改为 _get_torch_major_minor(torch.__version__) == "2.1"。 原写法在 torch 2.10 下因 "2.10".startswith("2.1") 为 True 而误走 2.1 分支, _native_register_fake 未定义,导致 test_compatible_register_fake_{decorator,wrapper} patch 失败。 复用同文件已有的 _get_torch_major_minor(与 SUPPORTED_TORCH_PLUGIN_VARIANTS["2.10"] 解析一致)。 2. tests/quantization/test_quantize.py:404/426 dtype 由 torch.float32 改为 torch.bfloat16,对齐 QuantConfig 仅允许 fp16/bf16 的约束 (config.py:138-139),与同文件 line 380 既有写法一致。 修复 test_online_quantize_with_mm_and_fa、test_online_quantize_fa_layers_no_match。 # Test Plan - 定向回归:bash run_test.sh layers/test_register_ops.py quantization/test_quantize.py - 全量回归(可选):bash run_test.sh,确认 errors=0 - 建议在 torch 2.1.x 与 torch 2.10.x 各跑一次,验证版本分支选择正确 - 提交前:pre-commit run --all-files(本 PR 仅改 .py,无 .md,无需 markdownlint) # Test Report 环境:torch 2.10.x / python 3.12.13 / CANN 9.0.0 定向回归(修复后): - layers/test_register_ops.py:Ran 6 tests in 0.008s — OK(修复前 2 errors) - quantization/test_quantize.py:Ran 43 tests in 0.214s — OK (skipped=1)(修复前 2 errors) See merge request: Ascend/MindIE-SD!456 | 14 天前 | |
[Feature]新增gitleaks敏感信息检测 Co-authored-by: zhangdiago<zhangdi5@huawei.com> # message auto-generated for no-merge-commit merge: !458 merge gitleaks into dev [Feature]新增gitleaks敏感信息检测 Created-by: zhangdiago Commit-by: zhangdiago Merged-by: ascend-robot Description: <!-- PR描述模板更新日期:20251224 --> # Which issue(s) this PR fixes or accomplishes > **如问题已解决,按照下方示例附上ISSUE单号 / Apply bug or request issue as follow if the solved**\ > Fixes #ISSUE ID\ > **Fixes关键字会自动关闭issue,如问题部分解决请不要使用Fixes,可以用下方标签替代\ > Fixes will automatically close issue, please use the following tag if only part of the issue is solved**\ > Fix part of #ISSUE ID N/A-新增CICDgitleaks二进制离线扫描工具,非问题修复 # Purpose 引入gitleaks二进制离线扫描工具 新增pre-commit/.gitleaks.toml配置,继承官方全部检测规则 配置pre-commit钩子,提交前自动扫描密钥硬编码风险。 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request PR代码提交前自动扫描密钥硬编码等风险 # Test Report See merge request: Ascend/MindIE-SD!458 | 13 天前 | |
[Feature][ops]Add mul_add fused operator support Co-authored-by: wangwei<daviwang2026@qq.com> # message auto-generated for no-merge-commit merge: !481 merge feature/mul_add_ops_dev into dev [Feature][ops]Add mul_add fused operator support Created-by: daviwang Commit-by: wangwei Merged-by: ascend-robot Description: # Purpose 新增 mul_add 融合算子,支持在昇腾 NPU 上执行融合乘加计算: - 算子接口:mindiesd::mul_add(a, b, c) -> y - 计算公式:y = a + b * c - 输入 shape:a/b = [batch, seq_len, hidden_size],c = [batch, 1, hidden_size] - 支持数据类型:FP16、BF16 - 支持计算平台:ascend910b、ascend910_93、ascend950 主要变更: 1. 在 csrc/ops/mul_add/ 下新增算子定义(OpDef)、shape/dtype 推导(proto)、tiling 计算及 AscendC kernel 实现。 2. 在 csrc/plugin/ 下新增 mul_add PTA plugin,并在 register_ops.cpp 中注册到 mindiesd torch 库。 3. 更新 csrc/CMakeLists.txt,将 mul_add.cpp 加入 PTAExtensionOPS 编译。 4. 更新 build/build_ops.sh 默认算子列表,确保默认构建包含 mul_add。 5. 在 tests/ops/mul_add/ 下新增单元测试,覆盖基础 shape、典型 LLM shape、小 shape 边界及 FP16 精度场景。 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request 1. 编译 AscendC 算子: bash cd build bash build_ops.sh 2. 编译 PTA plugin: bash cd build bash build_plugin.sh 3. 运行 mul_add 单元测试(需在 NPU 环境): bash cd tests/ops/mul_add python test_mul_add.py 4. 可选:通过环境变量指定 NPU 卡号: ``bash MINDIESD_TEST_NPU_ID=7 python test_mul_add.py # Test Report • 测试覆盖: • 基础小 shape:(1, 4, 128) • 典型 LLM shape:(1, 7200, 4608)、(1, 256, 4608)、(1, 7200, 3072)、(1, 256, 3072) • 多组小 shape 边界:(1, 1, 64) ~ (1, 64, 1024) • FP16 精度:(1, 256, 3072) • 参考实现:ref = a.float() + b.float() * c.float() • 精度阈值:rtol=1e-2, atol=1e-2 - 运行环境:CANN 9.1.T560, Ascend950PR NPU, Python 3.11, aarch64 - 所有新增测试用例均通过: - test_mul_add_basic PASSED - test_mul_add_typical PASSED - test_mul_add_small_shapes PASSED - test_mul_add_fp16` PASSED - 最大误差:BF16 < 1e-2,FP16 < 1e-2 See merge request: Ascend/MindIE-SD!481 | 2 天前 | |
[openlibing.ci] auto fixes by pre-commit | 2 个月前 | |
[CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !462 merge clang-tidy into dev [CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix #233 # Purpose 为仓库接入 clang-tidy 静态检查,作为 pre-commit 质量门禁的一部分,并消除编译选项导致的误报: - 新增 .clang-tidy 配置(检查规则集、WarningsAsErrors、header-filter 等)。 - 在 .pre-commit-config.yaml 增加 local clang-tidy hook: - 检查范围限定 csrc/ 下 .cpp/.cc/.cxx/.c; - 修正原先指向不存在的 mx_driving/csrc/ 路径; - exclude: ^csrc/ops/,暂不扫描 ops 子树(AscendC kernel 头缺失,后续单独处理)。 - 在 setup.py 的 merge_compile_commands 之后新增 strip_clang_tidy_unsupported_flags(), 自动剥离 clang 前端无法识别的 GCC 选项(如 -fabi-version=11),避免 clang-tidy 因 "unknown argument" 失败。 # Test Plan 1. 安装 lint 依赖:python -m pip install -r requirements-lint.txt && pre-commit install 2. 构建生成 compile_commands.json(自动剥离 -fabi-version):python setup.py bdist_wheel 3. 运行 clang-tidy hook 验证范围与无 unknown argument 报错:pre-commit run clang-tidy --all-files 4. 预期:csrc/ops 不被扫描;csrc/plugin 下不再出现 -fabi-version=11 报错。 # Test Report <!-- 待补充:粘贴实际运行结果摘要 --> - pre-commit run clang-tidy --all-files 结果:(待补充) - compile_commands.json 中 -fabi-version 出现次数:0(已验证) - csrc/ops 在扫描日志中出现次数:0(已验证跳过生效) See merge request: Ascend/MindIE-SD!462 | 7 天前 | |
[CI][build]Adopt PEP 517 build flow and refresh UT coverage Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !247 merge dev into dev [CI][build]Adopt PEP 517 build flow and refresh UT coverage Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #72 # Purpose Adopt PEP 517 build flow and refresh UT coverage # Test Plan CI同步修改构建入口和覆盖率拦截,需要通过CIE配置自定义流水线验证 https://gitcode.com/Ascend/MindIE-CI/pull/62 # Test Report   See merge request: Ascend/MindIE-SD!247 | 3 个月前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
[Docs][developer-guide]Optimize developer guide structure and content Co-authored-by: blian6<bin.lian@outlook.com> # message auto-generated for no-merge-commit merge: !306 merge dev into dev [Docs][developer-guide]Optimize developer guide structure and content Created-by: blian Commit-by: blian6 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes N/A # Purpose 对 docs/zh/developer_guide/ 下的开发者指南进行系统性优化: **文件重组** - tooling.md → dev_setup.md(zh + en 同步改名) - build_guide.md:删除 5 节无关内容(镜像安装、容器安装、升级、卸载) - test.md:删除「常见异常说明」节(推理期运行时错误,与测试无关) - pattern_dev_guide.md:删除 Mismatch 调试和 Copy 消减两节 - benchmark_and_profiling.md:精简为纯性能分析文档 **新增文件(4 个)** - repo_structure.md:仓库目录结构与模块说明 - contribution_guide.md:贡献全流程(编码规范表格化、PR 流程、新手上路) - pattern_dev_guide.md:Pattern 开发 5 阶段指南 - benchmark_and_profiling.md:性能分析(Profiling 采集 + 数据分析 + 特性说明) **内容扩充** - test.md:新增「测试编写规范」节(组织原则 + 性能规范) - contribution_guide.md:编码规范从 1 段扩展为 Ruff 规则表 + pre-commit 钩子表 + Python 约定 - 新增编码注意事项(新功能支持准则、多分支默认路径优先) **文档导航** - docs/zh/index.md toctree 同步更新 **Skill 修正** - profiling-collection/SKILL.md:warmup 步数修正、移除不存在的 collect_profile.py 引用 # Test Plan 1. 基础 pre-commit 检查: pre-commit run --all-files 2. Markdown lint 检查:pre-commit run markdownlint --all-files --hook-stage manual 3. 文档站点编译:python docs/build_docs.py # Test Report | 检查项 | 结果 | |--------|------| | trailing-whitespace | ✅ Passed | | end-of-file-fixer | ✅ Passed | | codespell | ✅ Passed | | typos | ✅ Passed | | markdownlint | ✅ Passed(MD032/MD034 已修复) | | 中文 Sphinx build(25 源文件) | ✅ build succeeded,无 error/warning | | 英文 Sphinx build(21 源文件) | ✅ build succeeded,无 error/warning | See merge request: Ascend/MindIE-SD!306 | 2 个月前 | |
[CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !462 merge clang-tidy into dev [CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix #233 # Purpose 为仓库接入 clang-tidy 静态检查,作为 pre-commit 质量门禁的一部分,并消除编译选项导致的误报: - 新增 .clang-tidy 配置(检查规则集、WarningsAsErrors、header-filter 等)。 - 在 .pre-commit-config.yaml 增加 local clang-tidy hook: - 检查范围限定 csrc/ 下 .cpp/.cc/.cxx/.c; - 修正原先指向不存在的 mx_driving/csrc/ 路径; - exclude: ^csrc/ops/,暂不扫描 ops 子树(AscendC kernel 头缺失,后续单独处理)。 - 在 setup.py 的 merge_compile_commands 之后新增 strip_clang_tidy_unsupported_flags(), 自动剥离 clang 前端无法识别的 GCC 选项(如 -fabi-version=11),避免 clang-tidy 因 "unknown argument" 失败。 # Test Plan 1. 安装 lint 依赖:python -m pip install -r requirements-lint.txt && pre-commit install 2. 构建生成 compile_commands.json(自动剥离 -fabi-version):python setup.py bdist_wheel 3. 运行 clang-tidy hook 验证范围与无 unknown argument 报错:pre-commit run clang-tidy --all-files 4. 预期:csrc/ops 不被扫描;csrc/plugin 下不再出现 -fabi-version=11 报错。 # Test Report <!-- 待补充:粘贴实际运行结果摘要 --> - pre-commit run clang-tidy --all-files 结果:(待补充) - compile_commands.json 中 -fabi-version 出现次数:0(已验证) - csrc/ops 在扫描日志中出现次数:0(已验证跳过生效) See merge request: Ascend/MindIE-SD!462 | 7 天前 | |
[Docs][RTD]Fix ReadTheDocs build to use zh/en source directories Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !293 merge rtd_0513 into dev [Docs][RTD]Fix ReadTheDocs build to use zh/en source directories Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes 无关issue,仅修正文档服务 # Purpose - 修复 ReadTheDocs 因python版本和Sphinx版本不匹配导致无法构建的问题。 修改内容: 1. .readthedocs.yaml:改用 python3.11 - 修复 ReadTheDocs 构建的文档首页显示英文目录而非中文目录的问题。 根因:ReadTheDocs 默认使用 docs/ 作为 Sphinx 源目录,首页渲染的是 docs/index.md,其 toctree caption 为英文("Getting Started"、"User Guide" 等)。而中文首页 docs/zh/index.md 的 toctree caption 为中文("快速开始"、"加速特性" 等),但 ReadTheDocs 未以其为构建源目录。 修改内容: 1. .readthedocs.yaml:改用 build.commands 自定义构建,分别用 docs/zh/ 和 docs/en/ 作为源目录构建中文和英文文档,根入口重定向到中文首页 2. docs/conf.py:根据 READTHEDOCS 环境变量动态设置语言切换器 URL,适配 ReadTheDocs 与本地构建的不同输出路径 3. docs/_templates/language-switcher.html 及其组件:改进 JS 路径解析,同时支持本地构建的 /html/ 路径和 ReadTheDocs 的 /zh/、/en/ 路径 # Test Plan 1. 合并后触发 ReadTheDocs 构建,观察构建日志是否成功 2. 访问 https://mindie-sd-docs.readthedocs.io/zh-cn/latest/ 确认自动跳转到中文首页 3. 检查中文首页左侧目录是否为中文标题("快速开始"、"加速特性" 等) 4. 检查英文首页左侧目录是否为英文标题("Getting Started"、"Feature Guides" 等) 5. 验证语言切换器在中英文页面间正确跳转 # Test Report https://mindie-sd-docs.readthedocs.io/zh-cn/latest/zh/index.html See merge request: Ascend/MindIE-SD!293 | 2 个月前 | |
[Docs] refresh lint and dependency config, docker, docs Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !243 merge dev into dev [Docs] refresh lint and dependency config, docker, docs Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #72 # Purpose This PR delivers five grouped changes: 1. add a 910B AArch64 docker build environment; 2. refresh repository lint/dependency inputs; 3. reorganize developer documentation and navigation; 4. align homepage, manuals, and tech report entrypoints; 5. consolidate governance skill routing and require proactive commit/PR draft output. # Test Plan 1. Review the new docker build files and README. 2. Review repository dependency and pre-commit configuration changes. 3. Review Chinese and English developer guide updates and navigation entrypoints. 4. Review homepage/manual/report entry changes. 5. Review AGENTS.md and .agents/skills/mindie-sd-community-governance/SKILL.md for routing and PR-drafting requirements. # Test Report  See merge request: Ascend/MindIE-SD!243 | 3 个月前 | |
[Docs] refresh lint and dependency config, docker, docs Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !243 merge dev into dev [Docs] refresh lint and dependency config, docker, docs Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #72 # Purpose This PR delivers five grouped changes: 1. add a 910B AArch64 docker build environment; 2. refresh repository lint/dependency inputs; 3. reorganize developer documentation and navigation; 4. align homepage, manuals, and tech report entrypoints; 5. consolidate governance skill routing and require proactive commit/PR draft output. # Test Plan 1. Review the new docker build files and README. 2. Review repository dependency and pre-commit configuration changes. 3. Review Chinese and English developer guide updates and navigation entrypoints. 4. Review homepage/manual/report entry changes. 5. Review AGENTS.md and .agents/skills/mindie-sd-community-governance/SKILL.md for routing and PR-drafting requirements. # Test Report  See merge request: Ascend/MindIE-SD!243 | 3 个月前 | |
docs: 更新社区治理和文档 Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !232 merge dev into dev docs: 更新社区治理和文档 Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes 更新社区治理和文档 # Purpose 更新相关文档要求 # Test Plan 文档更新 # Test Report See merge request: Ascend/MindIE-SD!232 | 3 个月前 | |
[Docs] refresh lint and dependency config, docker, docs Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !243 merge dev into dev [Docs] refresh lint and dependency config, docker, docs Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > Fix part of #72 # Purpose This PR delivers five grouped changes: 1. add a 910B AArch64 docker build environment; 2. refresh repository lint/dependency inputs; 3. reorganize developer documentation and navigation; 4. align homepage, manuals, and tech report entrypoints; 5. consolidate governance skill routing and require proactive commit/PR draft output. # Test Plan 1. Review the new docker build files and README. 2. Review repository dependency and pre-commit configuration changes. 3. Review Chinese and English developer guide updates and navigation entrypoints. 4. Review homepage/manual/report entry changes. 5. Review AGENTS.md and .agents/skills/mindie-sd-community-governance/SKILL.md for routing and PR-drafting requirements. # Test Report  See merge request: Ascend/MindIE-SD!243 | 3 个月前 | |
【docs】文档修改-增加API参考&加速API Co-authored-by: xiao-qing123<xiaoqing14@h-partners.com> # message auto-generated for no-merge-commit merge: !263 merge dev into dev 【docs】文档修改-增加API参考&加速API Created-by: xiao-qing123 Commit-by: xiao-qing123 Merged-by: ascend-robot Description: fixes [#86](https://gitcode.com/Ascend/MindIE-SD/issues/86) 1、新增API参考(社区API接口) 2、新增加速API(原社区layer层) 3、删除readme中的快速入门和单多卡并行示例内容(有单独的quick_start承载) 4、算子融合单独拆分出来,在特性章节独立存在 5、删除特性章节目录名称中的“加速特性” 6、黄区大模型检测问题修改 See merge request: Ascend/MindIE-SD!263 | 3 个月前 | |
恢复run包构建 Co-authored-by: taoyuan_guo<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !233 merge run_package into dev 恢复run包构建 Created-by: changetheway Commit-by: changetheway;taoyuan_guo Merged-by: ascend-robot Description: 修改方案: 1)将setup.py和原本的build.sh解耦,原来的算子编译和plugin编译在build.sh中执行,setup.py需要调用build.sh 2)build.sh改成两步:运行setup.py生成whl包、构建run包所需的tar.gz包 这样,既可以使用现在的python setup.py bdist_wheel构建whl包,也可以执行bash build.sh恢复run包构建 See merge request: Ascend/MindIE-SD!233 | 3 个月前 | |
[dev]同步最新代码 Co-authored-by: mazhixin00_00<mazhixin7@huawei.com> | 7 个月前 | |
[Docs] Corrected the link error and deleted expired QR codes Co-authored-by: herizhen<1270637059@qq.com> # message auto-generated for no-merge-commit merge: !370 merge dev into dev [Docs] Corrected the link error and deleted expired QR codes Created-by: herizhen Commit-by: herizhen Merged-by: ascend-robot Description: <!-- PR描述模板更新日期:20251224 --> # Which issue(s) this PR fixes or accomplishes Corrected the link error and deleted expired QR codes Fixes #177 # Purpose # Test Plan Documentation-only changes, no testing required. # Test Report See merge request: Ascend/MindIE-SD!370 | 1 个月前 | |
[Docs]: 新增联系我们章节并补充微信交流群二维码 Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !478 merge dev into dev [Docs]: 新增联系我们章节并补充微信交流群二维码 Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes > <!-- TODO: 若有关联 issue 请补充编号,例如 Fixes #xxx --> # Purpose 在 README 的「快速入门」与「开发文档」之间新增「联系我们」章节,方便用户反馈与交流: - 引导通过 GitCode Issues 参与讨论(https://gitcode.com/Ascend/MindIE-SD/issues) - 提供微信交流群二维码(docs/figures/wx_sd.jpg) - 说明群二维码失效时可通过个人微信(docs/figures/wx_lwl.jpg)邀请入群 # Test Plan 1. 在 GitCode 仓库页面预览 README,确认「联系我们」章节位置正确(位于「快速入门」之后、「开发文档」之前)。 2. 点击两个二维码图片,确认能正常显示且路径正确(docs/figures/wx_sd.jpg、docs/figures/wx_lwl.jpg)。 3. 点击 GitCode Issues 链接,确认跳转有效。 4. 执行 markdownlint 专项检查。 # Test Report - markdownlint(manual stage)对 README.md 检查:本次新增内容(42–54 行)无违规。 - 注:README.md:66–67 报 MD004(「生态支持」使用 + 列表),为既有问题,不在本次改动范围,未改动。 - 两张二维码图片均已存在于 docs/figures/。 - pre-commit:trim trailing whitespace / fix end of files / codespell / typos 均 Passed;Gitleaks 因本地缺少 ./gitleaks 二进制未跑(本改动无任何密钥)。 # Purpose # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request # Test Report See merge request: Ascend/MindIE-SD!478 | 6 天前 | |
【docs】文档修改-增加API参考&加速API Co-authored-by: xiao-qing123<xiaoqing14@h-partners.com> # message auto-generated for no-merge-commit merge: !263 merge dev into dev 【docs】文档修改-增加API参考&加速API Created-by: xiao-qing123 Commit-by: xiao-qing123 Merged-by: ascend-robot Description: fixes [#86](https://gitcode.com/Ascend/MindIE-SD/issues/86) 1、新增API参考(社区API接口) 2、新增加速API(原社区layer层) 3、删除readme中的快速入门和单多卡并行示例内容(有单独的quick_start承载) 4、算子融合单独拆分出来,在特性章节独立存在 5、删除特性章节目录名称中的“加速特性” 6、黄区大模型检测问题修改 See merge request: Ascend/MindIE-SD!263 | 3 个月前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
docs: 更新社区治理和文档 Co-authored-by: weixin_44144262<lijinxi2@huawei.com> # message auto-generated for no-merge-commit merge: !232 merge dev into dev docs: 更新社区治理和文档 Created-by: weixin_44144262 Commit-by: weixin_44144262 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes 更新社区治理和文档 # Purpose 更新相关文档要求 # Test Plan 文档更新 # Test Report See merge request: Ascend/MindIE-SD!232 | 3 个月前 | |
[Feature][build]支持多 torch 版本 wheel 打包 Co-authored-by: guowenna1<guowenna1@huawei.com> # message auto-generated for no-merge-commit merge: !342 merge pypi into dev [Feature][build]支持多 torch 版本 wheel 打包 Created-by: guowenna1 Commit-by: guowenna1 Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes https://gitcode.com/Ascend/MindIE-SD/issues/180 # Purpose 本 PR 支持构建一个可适配多个 torch 版本的 MindIE-SD wheel 包,同时保留原有固定 torch 版本构建方式。 主要变更: - 新增 MINDIESD_WHEEL_MODE=multi_torch 打包模式。 - 默认仍保持原有固定 torch 版本构建方式不变。 - 运行时根据当前 torch.__version__ 自动选择对应的 libPTAExtensionOPS.so。 - 支持在同一个 wheel 中包含 torch 2.6、2.7、2.8、2.9、2.10 对应的 plugin 动态库。 - 新增 build/build_multi_torch_wheel.sh,用于从已准备好的 torch 容器中编译并收集多版本 plugin。 - 更新 package data 规则,支持打包 plugin/torchXX/*.so。 - 将 distribution 标记为二进制包,便于后续通过 auditwheel 修复为 PyPI 可接受的 manylinux wheel。 - 支持通过 ASCEND_OP_NAME 和 ASCEND_COMPUTE_UNIT 配置 Ascend ops 构建范围。 # Test Plan 可按以下方式验证: 1. 检查 Python 和 shell 语法: ``bash python3 -m py_compile setup.py mindiesd/layers/register_ops.py bash -n build/build_ops.sh bash -n build/build_multi_torch_wheel.sh 构建多 torch 版本 wheel: bash bash build/build_multi_torch_wheel.sh 检查 wheel 中是否包含多版本 plugin: bash python3 - <<'PY' import zipfile from pathlib import Path wheel = Path("dist/mindiesd-3.0.0-cp311-cp311-linux_aarch64.whl") with zipfile.ZipFile(wheel) as zf: libs = sorted(name for name in zf.namelist() if name.endswith("libPTAExtensionOPS.so")) print("\n".join(libs)) PY 修复为 PyPI 可接受的 manylinux wheel: bash python3 -m auditwheel repair \ --plat manylinux_2_34_aarch64 \ -w dist/repaired \ dist/mindiesd-3.0.0-cp311-cp311-linux_aarch64.whl 检查上传包元数据: bash python3 -m twine check dist/repaired/*.whl Test Report 已执行以下检查: bash python3 -m py_compile setup.py mindiesd/layers/register_ops.py bash -n build/build_ops.sh bash -n build/build_multi_torch_wheel.sh 结果:通过。 已在以下容器中验证 multi-torch wheel 构建流程: lala_torch26 lala_torch27 lala_torch28 lala_torch29 lala_torch210 生成的 wheel 包含以下 plugin: mindiesd/plugin/torch26/libPTAExtensionOPS.so mindiesd/plugin/torch27/libPTAExtensionOPS.so mindiesd/plugin/torch28/libPTAExtensionOPS.so mindiesd/plugin/torch29/libPTAExtensionOPS.so mindiesd/plugin/torch210/libPTAExtensionOPS.so PyPI 上传包检查: bash python3 -m twine check dist/repaired/mindiesd-3.0.0-cp311-cp311-manylinux_2_34_aarch64.whl 结果:通过。 补充说明: 1. **变更类别** 功能增强 / 构建与发布 2. **影响文件** setup.py、pyproject.toml、mindiesd/layers/register_ops.py、build/build_ops.sh、build/build_multi_torch_wheel.sh`、quant flash attention 相关 opdef 文件。 See merge request: Ascend/MindIE-SD!342 | 1 个月前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
[Chore][deps]Reorganize requirements into layers and loosen version constraints Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !398 merge requirment into dev [Chore][deps]Reorganize requirements into layers and loosen version constraints Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix part of #92 # Purpose 按 #92「外部依赖治理」的阶段一(依赖声明收敛)重构依赖分层并放宽版本约束,让安装边界更清晰、降低版本漂移: - 核心运行依赖收敛到 requirements.txt:仅 torch/torch_npu 钉 2.9.0,其余用 >=/不锁;新增此前 import mindiesd 缺失的 safetensors。 - 服务化依赖就近放到 examples/service/requirements.txt(唯一真相源);模型生态间接依赖交由外部模型仓自管。 - examples/dummy_run/requirements.txt 补 torch_npu、tensorboard(示例实际需要)。 - 放宽测试/Lint 工具版本:coverage、pre-commit 改 >=;timm==0.9.12(1.x 会 ImportError)、mypy==1.13.0(保类型检查可复现)按需保留 pin。 - 文档(installation/quick_start/service.md,中英文同步)补充分层安装说明与 triton-ascend 3.2.1 的 GitCode 安装来源。 - THIRD_PARTY_LICENSES.txt 去版本号并对齐当前依赖集(补 torch_npu/triton/triton-ascend/safetensors,删过期条目)。 # Test Plan 1. 安装核心运行依赖:pip install -r requirements.txt(应非破坏性,不动 torch/torch_npu)。 2. 源码构建 wheel:python setup.py bdist_wheel。 3. 验证 import mindiesd 及调用接口(如构造 QuantConfig / OnlineQuantConfig)。 4. (SparseLinearAttention)从 GitCode 发布页装 triton-ascend 3.2.1,确认无禁用告警。 5. 核对 installation/quick_start 文档安装路径与 THIRD_PARTY_LICENSES 清单。 # Test Report - pip install -r requirements.txt:非破坏性(装入 triton 3.5.0 / triton-ascend 3.2.1 / strenum,torch 与 torch_npu 不变)。 - python setup.py bdist_wheel:成功,产出 dist/mindiesd-3.0.0-*.whl 与 mindiesd/plugin/libPTAExtensionOPS.so。 - import mindiesd:通过,16 个 __all__ 符号全部可达。 See merge request: Ascend/MindIE-SD!398 | 30 天前 | |
[Docs]:修改文档资料 Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !451 merge docker_0713 into dev [Docs]:修改文档资料 Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: <!-- PR描述模板更新日期:20251224 --> # Which issue(s) this PR fixes or accomplishes > **如问题已解决,按照下方示例附上ISSUE单号 / Apply bug or request issue as follow if the solved**\ > Fixes #ISSUE ID\ > **Fixes关键字会自动关闭issue,如问题部分解决请不要使用Fixes,可以用下方标签替代\ > Fixes will automatically close issue, please use the following tag if only part of the issue is solved**\ > Fix part of #ISSUE ID # Purpose 1. 按照镜像发布规范修改tag命名 2. 修复CANN安装指导和链接中内容章节对不上的问题 3. 修复triton-ascend 3.2.1不在pypi源中导致的安装问题 4. 修复TTFHW扫出的缺少编译构建时的依赖包的问题 5. 统一文档内的链接在打开时是新建浏览器标签打开,不会覆盖原链接打开 # Test Plan > 设计了哪些测试内容,指导他人如何对你的PR进行测试\ > Apply information to show others your test design and how to test your Pull Request # Test Report See merge request: Ascend/MindIE-SD!451 | 15 天前 | |
[CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Co-authored-by: changetheway<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !462 merge clang-tidy into dev [CI][lint]Integrate clang-tidy static analysis and auto-strip incompatible compile flags Created-by: changetheway Commit-by: changetheway Merged-by: ascend-robot Description: # Which issue(s) this PR fixes or accomplishes Fix #233 # Purpose 为仓库接入 clang-tidy 静态检查,作为 pre-commit 质量门禁的一部分,并消除编译选项导致的误报: - 新增 .clang-tidy 配置(检查规则集、WarningsAsErrors、header-filter 等)。 - 在 .pre-commit-config.yaml 增加 local clang-tidy hook: - 检查范围限定 csrc/ 下 .cpp/.cc/.cxx/.c; - 修正原先指向不存在的 mx_driving/csrc/ 路径; - exclude: ^csrc/ops/,暂不扫描 ops 子树(AscendC kernel 头缺失,后续单独处理)。 - 在 setup.py 的 merge_compile_commands 之后新增 strip_clang_tidy_unsupported_flags(), 自动剥离 clang 前端无法识别的 GCC 选项(如 -fabi-version=11),避免 clang-tidy 因 "unknown argument" 失败。 # Test Plan 1. 安装 lint 依赖:python -m pip install -r requirements-lint.txt && pre-commit install 2. 构建生成 compile_commands.json(自动剥离 -fabi-version):python setup.py bdist_wheel 3. 运行 clang-tidy hook 验证范围与无 unknown argument 报错:pre-commit run clang-tidy --all-files 4. 预期:csrc/ops 不被扫描;csrc/plugin 下不再出现 -fabi-version=11 报错。 # Test Report <!-- 待补充:粘贴实际运行结果摘要 --> - pre-commit run clang-tidy --all-files 结果:(待补充) - compile_commands.json 中 -fabi-version 出现次数:0(已验证) - csrc/ops 在扫描日志中出现次数:0(已验证跳过生效) See merge request: Ascend/MindIE-SD!462 | 7 天前 | |
恢复run包构建 Co-authored-by: taoyuan_guo<guotaoyuan1@h-partners.com> # message auto-generated for no-merge-commit merge: !233 merge run_package into dev 恢复run包构建 Created-by: changetheway Commit-by: changetheway;taoyuan_guo Merged-by: ascend-robot Description: 修改方案: 1)将setup.py和原本的build.sh解耦,原来的算子编译和plugin编译在build.sh中执行,setup.py需要调用build.sh 2)build.sh改成两步:运行setup.py生成whl包、构建run包所需的tar.gz包 这样,既可以使用现在的python setup.py bdist_wheel构建whl包,也可以执行bash build.sh恢复run包构建 See merge request: Ascend/MindIE-SD!233 | 3 个月前 |
MindIE SD
📢 Latest News
- 12/31/2025: MindIE SD提供稀疏Attention计算能力
- 12/25/2025:vLLM Omni x MindIE SD 实现 Qwen-Image-Edit-2511 / Qwen-Image-Layered 昇腾原生高性能推理
- 11/30/2025:MindIE SD 正式宣布开源并面向公众开放!会议日历
🚀 简介
MindIE SD(Mind Inference Engine Stable Diffusion)是 MindIE 的视图生成推理模型套件,它的目标是为稳定扩散(Stable Diffusion, SD)系列大模型提供在昇腾硬件及其软件栈上的端到端推理解决方案。该软件系统内部集成了各功能模块,并对外提供统一的编程接口。
以下是两个 MindIE-SD 代码仓库智能体,只需点击 "Ask AI" 徽章,即可进入其专属页面,有效缓解源码阅读的困难,开启智能代码学习与问答体验!它们将帮助您更深入地理解 MindIE-SD 的运行原理,并协助解决使用过程中遇到的问题与错误。
📦 安装指南
快速开始:
pip install mindiesd
其他安装方式和环依赖详见安装指导。
🚀 架构介绍
MindIE SD 架构和关键特性详见架构介绍。

⚡️ 快速入门
快速开始: 以wan2.1为例,介绍使用mindiesd后的整体加速效果。
💬 联系我们
欢迎通过 GitCode Issues 直接参与讨论与反馈。
为了更便捷地交流并获得更快的响应,我们建立了 MindIE-SD 微信交流群。您可以扫描左侧的微信群二维码加入交流群;若群二维码已失效,请扫描右侧的个人微信二维码,添加好友后由其邀请您加入交流群。
| 微信交流群 | 个人微信(群二维码失效时添加) |
|---|---|
![]() |
![]() |
📚 开发文档
贡献指南:说明如何提交 Issue、Pull Request 以及代码规范。 AI Agent支持: 介绍仓内默认使用的一些skills及使用agent辅助开发流程 测试验证:介绍单元测试样例运行流程
🌐 生态支持
以下项目已集成MindIE SD作为多模态推理后端:
📝 Paper Citations
@misc{RainFusion2.0@2025,
title = {RainFusion2.0: Temporal-Spatial Awareness and Hardware-Efficient Block-wise Sparse Attention},
url = {https://gitcode.com/Ascend/MindIE-SD.git},
note = {Open-source software available at https://gitcode.com/Ascend/MindIE-SD.git},
author = {Aiyue Chen and others},
year = {2025}
}

