文件最后提交记录最后更新时间
docs(pytorch): docs updated Co-authored-by: z60119525<zhaotao68@h-partners.com> # message auto-generated for no-merge-commit merge: !2434 merge 26.0.0 into 26.0.0 docs(pytorch): docs updated Created-by: kali00 Commit-by: z60119525 Merged-by: ascend-robot Description: ## What this PR does / why we need it? https://gitcode.com/Ascend/MindSpeed-MM/issues/314 1、修改CANN安装引导 ## Does this PR introduce any user-facing change? 修改安装指导中关于CANN社区页面的引用链接 ## How was this patch tested? Not need to be tested for doc. See merge request: Ascend/MindSpeed-MM!24341 个月前
docs(mindspore):fix model_migration_mindspore.md and readme.md Co-authored-by: weixin_47897441<wuyouqi1@h-partners.com> # message auto-generated for no-merge-commit merge: !2387 merge 26.0.0 into 26.0.0 docs(mindspore):fix model_migration_mindspore.md and readme.md Created-by: weixin_47897441 Commit-by: weixin_47897441 Merged-by: ascend-robot Description: ## What this PR does / why we need it? Fix typos in documentation: correct model names (e.g., Qwen2.5, Qwen3VL) and remove redundant characters in the README table and migration guide. ## Does this PR introduce any user-facing change? No. ## How was this patch tested? Documentation only (N/A). See merge request: Ascend/MindSpeed-MM!23871 个月前
[Docs]splitting install_guide and rename quickstart Co-authored-by: z60119525<zhaotao68@h-partners.com> # message auto-generated for no-merge-commit merge: !2383 merge 26.0.0 into 26.0.0 [Docs]splitting install_guide and rename quickstart Created-by: kali00 Commit-by: z60119525 Merged-by: ascend-robot Description: ## What this PR does / why we need it? Please describe the background and detailed changes of the PR. If it is a bugfix, please attach the related issue. ## Does this PR introduce any user-facing change? Please describe whether the PR will result in any user-facing usage changes. If there is related documentation, please specify its path. ## How was this patch tested? Please explain how to verify the correctness and effectiveness of this feature, as well as its usage constraints and limitations. See merge request: Ascend/MindSpeed-MM!23831 个月前
[Docs]splitting install_guide and rename quickstart Co-authored-by: z60119525<zhaotao68@h-partners.com> # message auto-generated for no-merge-commit merge: !2383 merge 26.0.0 into 26.0.0 [Docs]splitting install_guide and rename quickstart Created-by: kali00 Commit-by: z60119525 Merged-by: ascend-robot Description: ## What this PR does / why we need it? Please describe the background and detailed changes of the PR. If it is a bugfix, please attach the related issue. ## Does this PR introduce any user-facing change? Please describe whether the PR will result in any user-facing usage changes. If there is related documentation, please specify its path. ## How was this patch tested? Please explain how to verify the correctness and effectiveness of this feature, as well as its usage constraints and limitations. See merge request: Ascend/MindSpeed-MM!23831 个月前
readme.md

MindSpore后端支持说明

MindSpeed MM已支持接入华为自研AI套件MindSpore,旨在提供华为全栈易用的端到端的多模态模型训练解决方案,以此获得更极致的性能体验。MindSpore后端提供了一套对标PyTorch的API,用户无需进行额外代码适配即可无缝切换。


🔥🔥🔥Latest News

🚀🚀🚀 MindSpore后端已支持CogVideoXQwen2.5VLQwen25OmniQwen3VL模型。

版本配套

MindSpeed MM + MindSpore后端的依赖配套如下表,安装步骤参考基础安装指导

依赖软件
昇腾NPU驱动固件 在研版本
昇腾 CANN 在研版本
MindSpore 2.7.2
Python >=3.10

环境部署

具体部署步骤请查看部署文档

快速上手

快速上手操作请查看快速上手文档


模型/特性介绍

已支持特性概览

模型 \ 特性 TP TP-SP VPP PP CP EP Distributed Optimizer Recomputation LoRA
CogVideoX系列-T2V CP (Ulysses)
CogVideoX系列-I2V CP (Ulysses)
Qwen2.5VL-7B
Qwen2.5VL-72B
Qwen2.5Omni-7B
Qwen3VL-8B
Qwen3VL-30B

备注:

支持模型

MindSpeed MM (MindSpore后端)模型支持列表
模型任务 模型 参数量 任务 集群 精度格式 NPU性能 参考性能 平均序列长度 支持情况 支持版本
多模态生成
CogVideoX-T2V 5B 预训练 1x8 BF16 0.46 (SPS) / 2.3.0
CogVideoX-I2V 5B 预训练 1x8 BF16 0.46 (SPS) / 2.3.0
CogVideoX 1.5-T2V 5B 预训练 1x8 BF16 2.09 (SPS) / 2.3.0
5B Lora微调 1x8 BF16 3.03 (SPS) / 支持中
CogVideoX 1.5-I2V 5B 预训练 1x8 BF16 2.01 (SPS) / 2.3.0
5B Lora微调 1x8 BF16 3.92 (SPS) / 支持中
Wan2.1-I2V 1.3B 预训练 1x8 BF16 0.61 (SPS) / 2.3.0
Wan2.1-T2V 1.3B 预训练 1x8 BF16 0.73 (SPS) / 2.3.0
1.3B Lora微调 1x8 BF16 / / 支持中
多模态理解 Qwen2.5-VL 3B 微调 1x8 BF16 21.79 (SPS) 563 支持中
7B 微调 1x8 BF16 12.67 (SPS) 563 2.3.0
32B 微调 2x8 BF16 / 563 支持中
72B 微调 8x8 BF16 256.28 (TPS) 563 2.3.0
Qwen25Omni 7B 微调 1x8 BF16 / / 2.3.0
Qwen3VL 8B 微调 1x8 BF16 / / 2.3.0
30B 微调 1x16 BF16 / / 2.3.0

特性规划

工具使用

昇腾Profiling采集工具

MindSpeed MM集成了昇腾profiling采集工具,以提供对模型运行情况的分析。该工具能够依照配置采集模型的算子、显存等关键信息,同时支持动静态两种采集方式,协助开发者分析模型瓶颈,并可根据实际场景需求选择使用。

具体方法见 README 的profiling章节。

MindStudio Insight性能分析工具

针对大模型集群场景的性能调优,这里推荐一款优秀的可视化调优工具MindStudio Insight。 MindStudio Insight提供了包括Timeline视图、通信分析、计算耗时等的可视化呈现,以便用户分析潜在的性能瓶颈,并指导如何采取措施消除或减少这些瓶颈。

具体使用方法见《MindStudio Insight操作指南》,下载地址《MindStudio Insight》