| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】【feature】补充qwen3-omni校准集 Co-authored-by: anreywmh<18845895998@163.com> # message auto-generated for no-merge-commit merge: !148 merge w_data into master 【msmodelslim】【feature】补充qwen3-omni校准集 Created-by: anreywmh Commit-by: anreywmh Merged-by: ascend-robot Description: 【msmodelslim】【feature】补充qwen3-omni校准集 See merge request: Ascend/msmodelslim!148 | 7 个月前 | |
【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Co-authored-by: xiaoheng181<eudemoniaxh@163.com> # message auto-generated for no-merge-commit merge: !495 merge sd_refactor_docs_and_examples into master 【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Created-by: xiaoheng181 Commit-by: xiaoheng181 Merged-by: ascend-robot Description: 感谢您贡献的Pull Request！ 在提交之前，请务必阅读 [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md)。 Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md). ## PR描述 (What this PR does / why we need it?) - 请明确说明您提交PR的变更内容。本部分旨在概述所做的变更，以及此PR是如何解决该问题的。请尽可能地提供有助于评审人员更高效、更快速完成检视审查的实用说明。 本 PR 是多模态生成（multimodal_sd_modelslim_v1）重构的第五笔（资料），依赖 PR1～PR4 代码均已合入（或本分支已堆叠包含全部代码提交）。同步用户文档、示例与 lab 资产，使对外说明与重构后的 inference_config / dataset / 场景化 model_type 一致。 开发者指南 docs/zh/developer_guide/integrating_multimodal_generation_model.md 新增多模态生成模型接入指南（接口分区、Wan2.2 场景化结构、Hunyuan 参数桥接、双专家与校准数据约定等）。 用户指南 docs/zh/feature_guide/quick_quantization_v1/usage.md 补充/调整一键量化中 inference_config、dataset、多模态 SD 相关说明。 示例 example/multimodal_sd/Wan2_2/README.md Wan2.2 场景化 model_type 与配置示例说明。 Lab lab_calib/wan2_2_*、lab_calib/hunyuanvideo/ 分场景校准数据索引与样例图。 Lab lab_practice/wan2_2/*.yaml、lab_practice/hunyuan_video/*.yaml 与 inference_config / dataset 对齐的实践配置。 配置 config/config.ini 与 Wan2.2 场景化 model_type、loader 注册等保持一致。 - 请说明为何需要这些更改，例如具体的使用场景或bug描述。 代码 PR 单独合入后，用户若仍按旧 model_config 文档操作易配置失败；本 PR 将接入方式、YAML 样例与 lab 资产一次性对齐，降低合入后的使用成本。 依赖：请在 PR1～PR4 全部合入 master 后再合本 PR（或确认 target 分支已包含全部代码变更）。 - 关联issue号（如果有）。 - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Related issue number (if any) ## 面向用户的变更 (Does this PR introduce _any_ user-facing change)? - 请注意，这里指的是**任何**面向用户的变更，包括但不限于API、用户界面或其他使用方式上的变更。 有（文档与示例层面）。 新增/更新面向用户的 Markdown 说明与 lab YAML、校准数据索引。 无 生产 Python 逻辑变更；行为以已合入的代码 PR 为准。 - Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. ## 功能验证 (How was this patch tested?) 请确认CI已通过增量及存量的单元测试用例。 如果本次测试方式与常规单元测试不同，请详细说明您的测试步骤(最好提供完整的可复现的操作路径及关键截图)，以便Committer能够快速复现验证，也便于后续的维护。 如果未添加测试，请说明未添加的原因，以及为何难添加测试。 - [_] 功能自验 - [_] 本地自验截图(涉及个人标识符等敏感信息请注意脱敏) - [_] 新增/变更内容是否已新增/适配UT测试用例看护 CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. - [_] Self-verification of the feature. - [_] Screenshot of local self-verification (please anonymize any sensitive information such as personal identifiers) - [_] Have new or modified unit test (UT) cases been added or adapted to cover the newly added or changed content? See merge request: Ascend/msmodelslim!495 | 4 个月前 | |
[Feature] Qwen3.5/3.6模型FA混合量化：QK-MXFP8 动态 + V-MXFP8 PerChannel 静态 Co-authored-by: joejoezhou<zhourongchen1@huawei.com> # message auto-generated for no-merge-commit merge: !934 merge feature/qwen35-397b-fa-mxfp8 into master [Feature] Qwen3.5/3.6模型FA混合量化：QK-MXFP8 动态 + V-MXFP8 PerChannel 静态 Created-by: joejoezhou Commit-by: joejoezhou Merged-by: ascend-robot Description: ## 1. 影响面评估 **接口变更（按需）：** 无。 **输出件变更（按需）：** AscendV1 格式新增 FA 混合量化量化 QK_MXFP8_DYNAMIC_V_MXFP8_PER_CHANNEL。 **非兼容变更（按需）：** 无 **SIG 评审结论（按需）：** 无 ## 2. 修改描述 **修改背景（可选）：** Qwen3.5-397B 全量化（线性层 W4A4 / W8A8）需要注意力部分配套低比特方案。Q/K 参与得分计算、随输入分布变化，宜动态免校准；V 承接缓存写入，若沿用 MX per-block 按 32 元素分块，head_dim 不是 32 整数倍时行尾会留下不足 32 元素的尾块，V 缓存随 decode 逐 token 追加，反复 padding 代价高。因此 Q/K 走 MXFP8 per-block 动态，V 改走 per-channel 静态，将 head 并入通道维跨整条序列共享 E8M0 指数，从根上消除尾块，量化参数离线固化并与 KVCache 量化的通道语义对齐。 **修改目的：** 为 Qwen3.5-397B 提供 Q/K MXFP8 per-block 动态 + V MXFP8 per-channel 静态的注意力混合量化能力，打通 IR、校准、AscendV1 落盘与模型适配全链路，并补齐对应量化模式资料。 **修改内容：** - 量化模式：新增 MXFP8FakeQuantActivationPerChannel（FA V 分支 MXFP8 per-channel 静态伪量化层） - 量化格式：AscendV1 支持 per-channel 静态分支的类型串拼装与 uint8 scale 落盘，多卡 merge 阶段跨 rank 聚合分支状态统一拼串 - FA3 量化：支持 per-channel 静态分支（离线校准取参）；动态分支免校准，校准阶段不再空转 per-head 统计 - 数据格式：mxfp per-block 量化削减整份 fp32 临时，降低 FA3 校准显存峰值 - 模型适配：qwen3_5_moe 注入 full_attention 层 FA3 量化、跳过 linear_attention 层 - 校准集：新增 qwen3_cot_vlm_text 校准集，超长样本按段落截断至 8k，避免校准 OOM - docs：新增《FA QK-MXFP8 动态 / V-MXFP8 PerChannel 静态量化》词条，同步 FA 量化模式索引、量化模式总览表及 AscendV1 quant_type 取值规则与落盘说明 ## 3. 功能验证 - [x] 功能自验：本地单测通过（190 个用例） - [ ] 本地自验用例截图（请勿包含个人信息；可附复现命令） ## 4. 自检（请逐项确认，不适用标 N/A） **典型安全编码问题** - [x] 是否已校验外部数据 - [x] 是否未采集或打印敏感信息 - [x] 是否已正确设置文件权限 - [x] 是否充分考虑浮点运算溢出、除零等异常场景 - [x] 是否已对正则表达式做 ReDos 检查 **DT** - [x] 是否具备 UT 测试用例看护：test_cases/ir/test_mxfp8_activation_static.py、test_cases/ir/api/impl/test_mxfp8_per_channel_quantization.py、test_ascendv1.py、test_ascendv1_distributed.py、test_fa3_processor.py 等 - [ ] 是否需要添加冒烟：否 See merge request: Ascend/msmodelslim!934 | 1 个月前 | |
【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Co-authored-by: xiaoheng181<eudemoniaxh@163.com> # message auto-generated for no-merge-commit merge: !495 merge sd_refactor_docs_and_examples into master 【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Created-by: xiaoheng181 Commit-by: xiaoheng181 Merged-by: ascend-robot Description: 感谢您贡献的Pull Request！ 在提交之前，请务必阅读 [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md)。 Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md). ## PR描述 (What this PR does / why we need it?) - 请明确说明您提交PR的变更内容。本部分旨在概述所做的变更，以及此PR是如何解决该问题的。请尽可能地提供有助于评审人员更高效、更快速完成检视审查的实用说明。 本 PR 是多模态生成（multimodal_sd_modelslim_v1）重构的第五笔（资料），依赖 PR1～PR4 代码均已合入（或本分支已堆叠包含全部代码提交）。同步用户文档、示例与 lab 资产，使对外说明与重构后的 inference_config / dataset / 场景化 model_type 一致。 开发者指南 docs/zh/developer_guide/integrating_multimodal_generation_model.md 新增多模态生成模型接入指南（接口分区、Wan2.2 场景化结构、Hunyuan 参数桥接、双专家与校准数据约定等）。 用户指南 docs/zh/feature_guide/quick_quantization_v1/usage.md 补充/调整一键量化中 inference_config、dataset、多模态 SD 相关说明。 示例 example/multimodal_sd/Wan2_2/README.md Wan2.2 场景化 model_type 与配置示例说明。 Lab lab_calib/wan2_2_*、lab_calib/hunyuanvideo/ 分场景校准数据索引与样例图。 Lab lab_practice/wan2_2/*.yaml、lab_practice/hunyuan_video/*.yaml 与 inference_config / dataset 对齐的实践配置。 配置 config/config.ini 与 Wan2.2 场景化 model_type、loader 注册等保持一致。 - 请说明为何需要这些更改，例如具体的使用场景或bug描述。 代码 PR 单独合入后，用户若仍按旧 model_config 文档操作易配置失败；本 PR 将接入方式、YAML 样例与 lab 资产一次性对齐，降低合入后的使用成本。 依赖：请在 PR1～PR4 全部合入 master 后再合本 PR（或确认 target 分支已包含全部代码变更）。 - 关联issue号（如果有）。 - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Related issue number (if any) ## 面向用户的变更 (Does this PR introduce _any_ user-facing change)? - 请注意，这里指的是**任何**面向用户的变更，包括但不限于API、用户界面或其他使用方式上的变更。 有（文档与示例层面）。 新增/更新面向用户的 Markdown 说明与 lab YAML、校准数据索引。 无 生产 Python 逻辑变更；行为以已合入的代码 PR 为准。 - Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. ## 功能验证 (How was this patch tested?) 请确认CI已通过增量及存量的单元测试用例。 如果本次测试方式与常规单元测试不同，请详细说明您的测试步骤(最好提供完整的可复现的操作路径及关键截图)，以便Committer能够快速复现验证，也便于后续的维护。 如果未添加测试，请说明未添加的原因，以及为何难添加测试。 - [_] 功能自验 - [_] 本地自验截图(涉及个人标识符等敏感信息请注意脱敏) - [_] 新增/变更内容是否已新增/适配UT测试用例看护 CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. - [_] Self-verification of the feature. - [_] Screenshot of local self-verification (please anonymize any sensitive information such as personal identifiers) - [_] Have new or modified unit test (UT) cases been added or adapted to cover the newly added or changed content? See merge request: Ascend/msmodelslim!495 | 4 个月前 | |
【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Co-authored-by: xiaoheng181<eudemoniaxh@163.com> # message auto-generated for no-merge-commit merge: !495 merge sd_refactor_docs_and_examples into master 【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Created-by: xiaoheng181 Commit-by: xiaoheng181 Merged-by: ascend-robot Description: 感谢您贡献的Pull Request！ 在提交之前，请务必阅读 [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md)。 Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md). ## PR描述 (What this PR does / why we need it?) - 请明确说明您提交PR的变更内容。本部分旨在概述所做的变更，以及此PR是如何解决该问题的。请尽可能地提供有助于评审人员更高效、更快速完成检视审查的实用说明。 本 PR 是多模态生成（multimodal_sd_modelslim_v1）重构的第五笔（资料），依赖 PR1～PR4 代码均已合入（或本分支已堆叠包含全部代码提交）。同步用户文档、示例与 lab 资产，使对外说明与重构后的 inference_config / dataset / 场景化 model_type 一致。 开发者指南 docs/zh/developer_guide/integrating_multimodal_generation_model.md 新增多模态生成模型接入指南（接口分区、Wan2.2 场景化结构、Hunyuan 参数桥接、双专家与校准数据约定等）。 用户指南 docs/zh/feature_guide/quick_quantization_v1/usage.md 补充/调整一键量化中 inference_config、dataset、多模态 SD 相关说明。 示例 example/multimodal_sd/Wan2_2/README.md Wan2.2 场景化 model_type 与配置示例说明。 Lab lab_calib/wan2_2_*、lab_calib/hunyuanvideo/ 分场景校准数据索引与样例图。 Lab lab_practice/wan2_2/*.yaml、lab_practice/hunyuan_video/*.yaml 与 inference_config / dataset 对齐的实践配置。 配置 config/config.ini 与 Wan2.2 场景化 model_type、loader 注册等保持一致。 - 请说明为何需要这些更改，例如具体的使用场景或bug描述。 代码 PR 单独合入后，用户若仍按旧 model_config 文档操作易配置失败；本 PR 将接入方式、YAML 样例与 lab 资产一次性对齐，降低合入后的使用成本。 依赖：请在 PR1～PR4 全部合入 master 后再合本 PR（或确认 target 分支已包含全部代码变更）。 - 关联issue号（如果有）。 - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Related issue number (if any) ## 面向用户的变更 (Does this PR introduce _any_ user-facing change)? - 请注意，这里指的是**任何**面向用户的变更，包括但不限于API、用户界面或其他使用方式上的变更。 有（文档与示例层面）。 新增/更新面向用户的 Markdown 说明与 lab YAML、校准数据索引。 无 生产 Python 逻辑变更；行为以已合入的代码 PR 为准。 - Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. ## 功能验证 (How was this patch tested?) 请确认CI已通过增量及存量的单元测试用例。 如果本次测试方式与常规单元测试不同，请详细说明您的测试步骤(最好提供完整的可复现的操作路径及关键截图)，以便Committer能够快速复现验证，也便于后续的维护。 如果未添加测试，请说明未添加的原因，以及为何难添加测试。 - [_] 功能自验 - [_] 本地自验截图(涉及个人标识符等敏感信息请注意脱敏) - [_] 新增/变更内容是否已新增/适配UT测试用例看护 CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. - [_] Self-verification of the feature. - [_] Screenshot of local self-verification (please anonymize any sensitive information such as personal identifiers) - [_] Have new or modified unit test (UT) cases been added or adapted to cover the newly added or changed content? See merge request: Ascend/msmodelslim!495 | 4 个月前 | |
【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Co-authored-by: xiaoheng181<eudemoniaxh@163.com> # message auto-generated for no-merge-commit merge: !495 merge sd_refactor_docs_and_examples into master 【重构】多模态生成模型量化服务重构wan2.2/hunyuanvideo【增加资料和配置】 Created-by: xiaoheng181 Commit-by: xiaoheng181 Merged-by: ascend-robot Description: 感谢您贡献的Pull Request！ 在提交之前，请务必阅读 [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md)。 Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md). ## PR描述 (What this PR does / why we need it?) - 请明确说明您提交PR的变更内容。本部分旨在概述所做的变更，以及此PR是如何解决该问题的。请尽可能地提供有助于评审人员更高效、更快速完成检视审查的实用说明。 本 PR 是多模态生成（multimodal_sd_modelslim_v1）重构的第五笔（资料），依赖 PR1～PR4 代码均已合入（或本分支已堆叠包含全部代码提交）。同步用户文档、示例与 lab 资产，使对外说明与重构后的 inference_config / dataset / 场景化 model_type 一致。 开发者指南 docs/zh/developer_guide/integrating_multimodal_generation_model.md 新增多模态生成模型接入指南（接口分区、Wan2.2 场景化结构、Hunyuan 参数桥接、双专家与校准数据约定等）。 用户指南 docs/zh/feature_guide/quick_quantization_v1/usage.md 补充/调整一键量化中 inference_config、dataset、多模态 SD 相关说明。 示例 example/multimodal_sd/Wan2_2/README.md Wan2.2 场景化 model_type 与配置示例说明。 Lab lab_calib/wan2_2_*、lab_calib/hunyuanvideo/ 分场景校准数据索引与样例图。 Lab lab_practice/wan2_2/*.yaml、lab_practice/hunyuan_video/*.yaml 与 inference_config / dataset 对齐的实践配置。 配置 config/config.ini 与 Wan2.2 场景化 model_type、loader 注册等保持一致。 - 请说明为何需要这些更改，例如具体的使用场景或bug描述。 代码 PR 单独合入后，用户若仍按旧 model_config 文档操作易配置失败；本 PR 将接入方式、YAML 样例与 lab 资产一次性对齐，降低合入后的使用成本。 依赖：请在 PR1～PR4 全部合入 master 后再合本 PR（或确认 target 分支已包含全部代码变更）。 - 关联issue号（如果有）。 - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Related issue number (if any) ## 面向用户的变更 (Does this PR introduce _any_ user-facing change)? - 请注意，这里指的是**任何**面向用户的变更，包括但不限于API、用户界面或其他使用方式上的变更。 有（文档与示例层面）。 新增/更新面向用户的 Markdown 说明与 lab YAML、校准数据索引。 无 生产 Python 逻辑变更；行为以已合入的代码 PR 为准。 - Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. ## 功能验证 (How was this patch tested?) 请确认CI已通过增量及存量的单元测试用例。 如果本次测试方式与常规单元测试不同，请详细说明您的测试步骤(最好提供完整的可复现的操作路径及关键截图)，以便Committer能够快速复现验证，也便于后续的维护。 如果未添加测试，请说明未添加的原因，以及为何难添加测试。 - [_] 功能自验 - [_] 本地自验截图(涉及个人标识符等敏感信息请注意脱敏) - [_] 新增/变更内容是否已新增/适配UT测试用例看护 CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. - [_] Self-verification of the feature. - [_] Screenshot of local self-verification (please anonymize any sensitive information such as personal identifiers) - [_] Have new or modified unit test (UT) cases been added or adapted to cover the newly added or changed content? See merge request: Ascend/msmodelslim!495 | 4 个月前 | |
【docs】：大模型检查低错修改 Co-authored-by: zzm30<zhengzhimin1@h-partners.com> # message auto-generated for no-merge-commit merge: !333 merge master into master 【docs】：大模型检查低错修改 Created-by: zzm30 Commit-by: zzm30 Merged-by: ascend-robot Description: 感谢您贡献的Pull Request！ 在提交之前，请务必阅读 [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md)。 Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ [CONTRIBUTING.md](https://gitcode.com/Ascend/msmodelslim/blob/master/CONTRIBUTING.md). ## PR描述 (What this PR does / why we need it?) - 请明确说明您提交PR的变更内容。本部分旨在概述所做的变更，以及此PR是如何解决该问题的。请尽可能地提供有助于评审人员更高效、更快速完成检视审查的实用说明。 - 请说明为何需要这些更改，例如具体的使用场景或bug描述。 - 关联issue号（如果有）。 - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Related issue number (if any) ## 面向用户的变更 (Does this PR introduce _any_ user-facing change)? - 请注意，这里指的是**任何**面向用户的变更，包括但不限于API、用户界面或其他使用方式上的变更。 - Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. ## 功能验证 (How was this patch tested?) 请确认CI已通过增量及存量的单元测试用例。 如果本次测试方式与常规单元测试不同，请详细说明您的测试步骤(最好提供完整的可复现的操作路径及关键截图)，以便Committer能够快速复现验证，也便于后续的维护。 如果未添加测试，请说明未添加的原因，以及为何难添加测试。 - [_] 功能自验 - [_] 本地自验截图(涉及个人标识符等敏感信息请注意脱敏) - [_] 新增/变更内容是否已新增/适配UT测试用例看护 CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. - [_] Self-verification of the feature. - [_] Screenshot of local self-verification (please anonymize any sensitive information such as personal identifiers) - [_] Have new or modified unit test (UT) cases been added or adapted to cover the newly added or changed content? See merge request: Ascend/msmodelslim!333 | 5 个月前 | |
【msmodelslim】刷新文件头部声明和版权 Co-authored-by: caishengcheng<caishengcheng@huawei.com> # message auto-generated for no-merge-commit merge: !28 merge license into master 【msmodelslim】刷新文件头部声明和版权 Created-by: caishengcheng Commit-by: caishengcheng Merged-by: ascend-robot Description: 【msmodelslim】刷新文件头部声明和版权 See merge request: Ascend/msmodelslim!28 | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】【feature】Qwen3-Coder-480B-A35B w4a8 量化 Co-authored-by: ylzzz<yelinzhong@huawei.com> # message auto-generated for no-merge-commit merge: !58 merge ylzzz_qwen3_code into master 【msmodelslim】【feature】Qwen3-Coder-480B-A35B w4a8 量化 Created-by: ylzzz Commit-by: ylzzz Merged-by: ascend-robot Description: qwen3 code  See merge request: Ascend/msmodelslim!58 | 8 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
[Feature] introduce ra_compress long-sequence compression algorithm Co-authored-by: zwling<906711886@qq.com> # message auto-generated for no-merge-commit merge: !875 merge feature/ra-compress-reintroduce into master [Feature] introduce ra_compress long-sequence compression algorithm Created-by: zwling Commit-by: zwling Merged-by: ascend-robot Description: ## 背景 上游 master 中 ra_compress（长序列压缩算法）在 9441c0a 被 revert 删除。本 PR 在最新 master 基础上重新引入该算法（含 package layout 调整），并修复 attn_head CLI 的两个问题。 ## 内容 1. 重新引入 ra_compress 长序列压缩算法（attn_head 分析指标） - 代码：msmodelslim/processor/analysis/unary_operator/metrics/ra_compress/ - CLI：analyze attn_head 子命令 - 文档：ra_compress 词条/使用指南、attn_head 分析使用指南 - 配置/测试/校准数据同步引入 2. package layout 与 attention_mse 对齐（impl.py + __init__.py re-export） 3. bugfix： - _normalize_analyze_argv 的 scope 列表加入 attn_head，避免 msmodelslim analyze attn_head ... 被误注入默认 linear scope - 移除 attn_head 的 set_defaults(calib_dataset='calib_dummy.jsonl')， 避免静默回退到不满足 ≥10000 token 要求的 mix_calib.jsonl - 文档所有 attn_head CLI 示例显式指定 --calib_dataset calib_dummy.jsonl ## 测试 - 基于 upstream/master，cherry-pick 全程无冲突 - msmodelslim analyze attn_head -h 正常 See merge request: Ascend/msmodelslim!875 | 1 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【feature】add qwen3-32b w4a4 config and dataset Co-authored-by: xyxin_006<xyxin_hit@163.com> # message auto-generated for no-merge-commit merge: !233 merge w4a4_config_dataset into master 【feature】add qwen3-32b w4a4 config and dataset Created-by: xyxin_006 Commit-by: xyxin_006 Merged-by: ascend-robot Description: 1、补充Qwen3-32B w4a4量化配置以及校准数据集 2、同步修改LAOS量化方案文档 3、增加quant_type支持w4a4格式，文档中增加Qwen3-32B-w4a4拉起量化指令 See merge request: Ascend/msmodelslim!233 | 6 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 | |
【msmodelslim】cleancode Co-authored-by: caishengcheng<caishengcheng@huawei.com> | 9 个月前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 9 个月前 | ||
| 7 个月前 | ||
| 4 个月前 | ||
| 1 个月前 | ||
| 4 个月前 | ||
| 4 个月前 | ||
| 4 个月前 | ||
| 5 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 8 个月前 | ||
| 9 个月前 | ||
| 1 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 6 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 | ||
| 9 个月前 |