跳转至
MindStudio 工具链官方文档全集
{% include-markdown "../README.md" %}
正在初始化搜索引擎
mindstudio-docs/master
MindStudio 工具链官方文档全集
mindstudio-docs/master
Home
Tools
Tools
msdebug
msdebug
docs
docs
en
en
best_practices
best_practices
Typical Cases
contributing
contributing
Contributing Guide
development_guide
development_guide
MindStudio Debugger Architecture Design Specifications
MindStudio Debugger Development Environment Setup, Build, and UT Methods
install_guide
install_guide
MindStudio Debugger Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Handling Mechanism Description
quick_start
quick_start
msDebug Quick Start
release_notes
release_notes
MindStudio Debugger Release Notes
support
support
FAQs
user_guide
user_guide
MindStudio Debugger User Guide
zh
zh
best_practices
best_practices
典型案例
contributing
contributing
贡献指南
development_guide
development_guide
MindStudio Debugger 架构设计说明书
msDebug 开发指南
install_guide
install_guide
msDebug 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
msDebug算子调试工具快速入门
release_notes
release_notes
MindStudio Debugger 版本说明
support
support
FAQ
user_guide
user_guide
MindStudio Debugger工具用户指南
msinsight
msinsight
docs
docs
en
en
developer_guide
developer_guide
Development Guide
user_guide
user_guide
Basic Operations on MindStudio Insight
FAQs
MindStudio Insight Memory Tuning
MindStudio Insight Installation Guide
MindStudio Insight Operator Tuning
Introduction
MindStudio Insight Security Statement
MindStudio Insight Serving Tuning
MindStudio Insight System Tuning
zh
zh
best_practices
best_practices
基于 MindStudio Insight 定位集群快慢卡问题
基于Linux Kernel Trace的Host Bound问题分析
jupyterlab安装指南
快捷键说明
基于 PyTorch Snapshot 数据分析内存问题案例
基于 PyTorch Snapshot 数据分析内存问题
单卡 Top-Down 算子下发瓶颈分析
Timeline泳道介绍
verl场景下Snapshot数据采集和分析案例
development_guide
development_guide
design
design
Communication部分设计文档
Compute部分设计文档
Memory部分设计文档
Operator部分设计文档
智能防抖函数（SmartDebounce）
Summary 设计文档
MindStudio Insight 支持 Device 内存分析特性设计说明书
支持内存快照展示和分析
Timeline设计文档
泳道绘制
开发指南
开发环境搭建
install_guide
install_guide
MindStudio Insight安装指南
legal
legal
MindStudio Insight安全声明
quick_start
quick_start
快速入门（算子调优篇）
快速入门（系统调优篇）
release_notes
release_notes
MindStudio Insight 版本发布说明
support
support
issue_feedback
issue_feedback
Communication相关问题汇总
崩溃类问题常见处理措施
导入相关问题汇总
安装相关问题汇总
jupterlab相关问题修复
operator页签问题汇总
Timeline 汇总issue以及解决措施
白屏问题汇总
FAQ
user_guide
user_guide
MindStudio Insight基础操作
MindStudio Insight内存调优
MindStudio Insight算子调优
简介
MindStudio Insight服务化调优
MindStudio Insight系统调优
msit
msit
docs
docs
design
design
特性设计
特性设计
en
en
common
common
Contribution Process and Specifications
contributing
contributing
Contribution Guide
legal
legal
Disclaimer
License Statement
Public Network Address
msIT Security Statement
MindStudio Vulnerability Handling Mechanism Description
quick_start
quick_start
Inference Development Toolchain Quick Start
user_guide
user_guide
msIT Tool Selection Guide
Brief Introduction
zh
zh
common
common
贡献流程和规范说明
contributing
contributing
贡献指南
legal
legal
免责声明
LICENSE声明
公网地址
msit安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
推理开发工具链快速入门
user_guide
user_guide
msIT 工具选型指南
简介
mskl
mskl
docs
docs
en
en
api_reference
api_reference
MindStudio Kernel Launcher API Reference
contributing
contributing
Contributing Guide
development_guide
development_guide
msKL Repository Architecture
MindStudio Kernel Launcher Development Guide
install_guide
install_guide
MindStudio Kernel Launcher Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Mechanism Description
quick_start
quick_start
msKL Lightweight Kernel Call Quick Start
release_notes
release_notes
MindStudio Kernel Launcher Release Notes
support
support
MindStudio Kernel Launch FAQ
user_guide
user_guide
MindStudio Kernel Launcher User Guide
zh
zh
api_reference
api_reference
MindStudio Kernel Launcher API接口文档
contributing
contributing
贡献指南
development_guide
development_guide
msKL 代码仓架构说明
msKL 开发指南
install_guide
install_guide
msKL 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
msKL 算子 Kernel 轻量化调用快速入门
release_notes
release_notes
MindStudio Kernel Launcher 版本说明
support
support
MindStudio Kernel Launch 常见问题
user_guide
user_guide
MindStudio Kernel Launcher工具用户指南
mskpp
mskpp
docs
docs
en
en
api_reference
api_reference
msKPP API Reference
contributing
contributing
Contribution
development_guide
development_guide
MindStudio Kernel Performance Prediction Architecture Design Specifications
Development Environment Setup, Compilation, and UT Methods for MindStudio Kernel Performance Prediction
install_guide
install_guide
MindStudio Kernel Performance Prediction Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Declaration
MindStudio Vulnerability Handling Mechanism Description
quick_start
quick_start
msKPP Quick Start
release_notes
release_notes
MindStudio Kernel Performance Prediction Release Notes
user_guide
user_guide
Constraints and Precautions for MindStudio Kernel Performance Prediction
MindStudio Kernel Performance Prediction User Guide
zh
zh
api_reference
api_reference
MindStudio Kernel Performance Prediction对外接口使用说明
contributing
contributing
贡献指南
development_guide
development_guide
MindStudio Kernel Performance Prediction 架构设计说明书
msKPP 开发指南
install_guide
install_guide
msKPP 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
msKPP 算子建模工具快速入门
release_notes
release_notes
MindStudio Kernel Performance Prediction 版本说明
user_guide
user_guide
MindStudio Kernel Performance Prediction 工具限制与注意事项
MindStudio Kernel Performance Prediction 用户指南
msmemscope
msmemscope
docs
docs
en
en
api_reference
api_reference
API Reference
best_practices
best_practices
Basic Cases
communication_guide
communication_guide
Communication Guide
Issue Submission and Handling Guide
design
design
memscope Component Implementation Design Specifications
development_guide
development_guide
Contributing Guide
msMemScope Development Guide
install_guide
install_guide
msMemScope Installation Guide
legal
legal
Disclaimer
License Declaration
msMemScope Security Statement
memory_issue_troubleshooting_guide
memory_issue_troubleshooting_guide
memory_debug
memory_debug
Memory Leak
quick_start
quick_start
Getting Started with msMemScope
release_notes
release_notes
Release Notes
user_guide
user_guide
Memory Analysis
Memory Collection
Output File Specifications
Introduction
zh
zh
api_reference
api_reference
API参考
best_practices
best_practices
基础案例
communication_guide
communication_guide
交流指南
Issue提交指南
design
design
memscope组件实现设计说明书
development_guide
development_guide
贡献指南
msMemScope 开发指南
install_guide
install_guide
msMemScope 安装指南
legal
legal
免责声明
License 说明
msMemScope 安全声明
memory_issue_troubleshooting_guide
memory_issue_troubleshooting_guide
memory_debug
memory_debug
显存泄漏
quick_start
quick_start
msMemScope快速入门
release_notes
release_notes
版本说明
user_guide
user_guide
内存分析
内存采集
输出文件说明
简介
msmodeling
msmodeling
docs
docs
ai-native
ai-native
AI Native 架构
AI Native 使用案例
AI Native 工作流指南
design
design
Design Document: A3 RoCE Device Profile 新增支持
特性设计：A5 系列硬件 Device Profile 新增支持
Block Sparse Attention backend 实现设计
特性设计：TensorCast 适配 DeepSeek-V4 模型（Flash/Pro）
Design Document: [Feature Title / 特性标题]
特性设计：TensorCast 适配 FLUX.1-dev 图像生成仿真
GLM5 TensorCast 适配设计
GLM 5.2 实测算子仿真接入设计
特性设计：TensorCast 适配 Kimi K2.5 模型
特性设计：TensorCast 适配 Kimi K3 模型
Feature Design
特性设计：TensorCast 适配 MiniMax-M3 / MiniMax-M3-VL 模型
Design Document: TensorCast New Model Adaptation (Run-Through Scope)
Design Document: modeldiagnostics 仿真结构诊断 / Simulation Structure Diagnostics
特性设计
特性设计：吞吐寻优多硬件展示与终端 ASCII Plot
特性设计：服务化参数实测寻优功能
特性设计：服务化参数实测寻优功能适配pd分离参数搜索
特性设计：[算子性能数据采集能力建设]
实测算子性能数据库生命周期设计
Design Document: 工具支持基于实测算子性能的建模
查询驱动的 Shape 网格生成设计
Feature Design
Web UI 前端架构演进与体验优化设计
Web UI 重构设计文档（Vue 3 + FastAPI）
en
en
install_guide
install_guide
msModeling Installation Guide
legal
legal
msModeling Security Statement
quick_start
quick_start
msModeling Quick Start
user_guide
user_guide
support_matrix
support_matrix
Model and Feature Support Matrix
Fine-Grained Serving Simulation User Guide
TensorCast New Model Adaptation Guide (Run-Through Scope)
Model Inference Performance Simulation User Guide
Serving Performance Simulation User Guide
Web UI User Guide
OptiX Plugin Developer Guide
Measured Serving Optimization User Guide
perf_database
perf_database
forward_pass_traces
forward_pass_traces
tutorial
tutorial
Microbench xxxrun.py 生成教程
Op Mapping 教程 v2
RFC
RFC
RFC: Support for DeepseekV32 Model
RFC: DeepseekV32模型适配支持
RFC: DeepSeek-V4 Model Adaptation Support (Flash/Pro)
RFC: DeepSeek-V4 模型适配支持（Flash/Pro）
RFC: 增加diffusers模型建模支持
RFC：增加 FLUX.1-dev 图像生成仿真支持
RFC: Add MiniMax-M3 Model Adaptation Support
RFC: MiniMax-M3 模型适配支持
Qwen3.5 Simulation Support Design Document
Qwen3.5 仿真支持设计文档
RFC: Qwen3.8 Model Adaptation Support
RFC: Qwen3.8 模型适配支持
RFC：Qwen-Image-Edit Transformer workload simulation
RFC: Support Ulysses for Diffusers Model Proposal
tensorcast/ops/matmul.py
RFC：msmodeling AI Native 工作流
RFC: Chunked Prefill Latency 查表优化方案
RFC: msmodeling Chunked Prefill 建模支持方案
RFC: CLI 与 Web UI 参数系统解耦重构
RFC: MsModeling Support for Decode Context Parallel Simulation
RFC: MsModeling 支持 Decode Context Parallel 仿真
RFC: Support for User Custom Model
RFC: 支持用户自定义模型
RFC: Support for Custom Operator Modeling
RFC: 支持用户自定义算子建模
RFC: Context Parallel Modeling for DeepSeek-Style Attention
RFC: DeepSeek 风格注意力的 Context Parallel 建模方案
RFC: 设备画像自然语言导入器（deviceconfig Skill）
RFC: Dflash 统一建模方案
RFC: Support Classifier-free Guidance Parallel for Diffusers Model Proposal
RFC: Diffusers 模型远端 Repo ID 自动加载支持
RFC: General Model and Configuration Loading Optimization Proposal
RFC: 通用模型加载与配置加载优化方案
RFC: Modeling Support for Hunyuanvideo Model
RFC: Hunyuanvideo模型建模支持
RFC：图像生成推理性能仿真架构
RFC: User-Defined Fusion Op Performance Evaluation v2 (Plugin Mode)
RFC: 用户自定义融合算子性能评估 v2 (Plugin模式)
RFC: msmodeling 公开命令行统一规范化
RFC: Current Mixed-Batch Variable-Token Modeling in Throughput Optimizer
RFC: Throughput Optimizer 当前 Mixed-Batch 变长 Token 建模实现
RFC: msmodeling-env-installer Skill 设计方案
NPU Layer Analyzer Design Document
NPU Layer Analyzer 设计文档
RFC: optix 寻优工具 Agent 模式（Agent-Driven Service Parameter Tuning）
RFC: OptiX 部署子进程环境隔离
RFC: 真机轻量化寻优部署器（optix-deploy Skill）
RFC: optix 服务参数寻优可靠性增强
RFC: 参数范围推荐器（param-recommend Skill）
RFC: OptiX 实测寻优适配 PD 分离场景
1. Overview（概述）
RFC: PD Ratio Throughput Optimization
RFC: PD配比吞吐量寻优
RFC: Standalone Performance Database Collection Tooling
RFC: 性能数据库独立采集工具链
RFC: Pipeline Parallel Simulation Support
rfc_pipeline_parallel_support_zh
RFC: Pipeline Parallel Throughput Optimizer Support
RFC: Pipeline Parallel 吞吐量寻优支持
性能回归测试框架设计文档
RFC: Support Prefix Cache Hit Rate in CLI textgenerate / throughputoptimizer
RFC: CLI textgenerate / throughputoptimizer 支持 Prefix Cache 命中率
RFC: 实测算子性能数据库异常点发现与复测闭环
RFC: Profiling-Driven Empirical Performance Model
RFC: 基于 Profiling 的经验性能模型
RFC: Profiling Operator Interpolation Module Design (Phase 1)
RFC: 实测算子接入插值模块设计（第一阶段）
RFC: Profiling Interpolation Phase2 Operator Capability Expansion
RFC: Profiling 插值 Phase2 算子能力扩展
RFC：Q2 建模技术规划
RFC: Quantization Configuration System Optimization Proposal
RFC: 量化配置系统优化方案
RFC: Support for Adapting the Qwen3-VL Multimodal Model
RFC: Qwen3-VL多模态模型适配支持
RFC: Transformer 重复层代表层优化
msmodeling 子 SIG 分工与运作规范
RFC：独立单算子 msprof 性能测试
RFC: SwiGLU Operator Fusion
RFC: SwiGLU算子融合
1. Overview (概述)
rfc_text_generate_executor_skill_zh
RFC: msmodeling 三方库非必要依赖消减
RFC: Throughput Optimizer 适配 Dflash / DSpark 支持
rfc_throughput_optimizer_executor_skill_zh
RFC: throughput-optimizer-explainer Skill 设计方案
RFC: Throughput Optimizer 多硬件工作负载复用
RFC: Adopt uv for Dependency and Environment Management
zh
zh
install_guide
install_guide
msModeling 安装指南
推荐实践：服务化实测寻优环境与部署栈
legal
legal
msModeling安全声明
quick_start
quick_start
msModeling快速入门
user_guide
user_guide
support_matrix
support_matrix
模型支持与特性支持矩阵
OptiX插件 开发指导
服务化实测寻优使用指南
服务化细粒度仿真 使用指南
TensorCast 新模型适配开发指导（跑通级）
模型推理性能仿真 使用指南
服务化性能仿真 使用指南
Web UI 使用说明
msmodelslim
msmodelslim
docs
docs
en
en
best_practices
best_practices
Typical Cases
msModelSlim Quantized Weights Conversion Guide for AutoAWQ and AutoGPTQ
Quantization Accuracy Tuning Guide
Use Cases for W8A8 Quantized Weights in the Acceleration Library Scenario
Qwen3-32B W8A8 Accuracy Tuning Case
Sparse Quantization Accuracy Tuning Case
W8A16 Accuracy Tuning Strategy
W8A8 Quantization Accuracy Tuning Strategy
common
common
Coding Standards
Developer Testing Guide
contributing
contributing
Contributing to MindStudio ModelSlim
development_guide
development_guide
design
design
Design_of_Automatic_Optimization_Acceleration_Feature
msModelSlim Rich Quantized Model Support Feature Design Specification
Model Structure Extraction Feature Design Specification
msModelSlim Sensitive Layer Analysis Module Refactoring Feature Design Specification
msModelSlim Architecture
Quantization Format Integration Guide
LLM Model Integration Guide
Multimodal Generation Model Integration Guide
Multimodal Understanding Model Integration Guide
Multi-card Quantization Adaptation Guide
install_guide
install_guide
msModelSlim Tool Installation Guide
legal
legal
To Users of msModelSlim
LICENSE Notice
msModelSlim Security Statement
python_api_v0
python_api_v0
distillation_apis
distillation_apis
KnowledgeDistillConfig
KnowledgeDistillConfig
addcustomlossfunc
addintersoftlabel
addoutputsoftlabel
sethardlabel
setteachertrain
getdistillmodel
foundation_model_compression_apis
foundation_model_compression_apis
foundation_model_quantization_apis
foundation_model_quantization_apis
AntiOutlier
AntiOutlierConfig
CalibrationData
FakeQuantizeCalibrator
LayerSelector
AntiOutlierAdapter
CalibratorAdapter
ModelAdapter
process()
Calibrator
QuantConfig
run()
save()
process()
Calibrator
FAQuantizer
quant()
QuantConfig
run()
save()
foundation_model_sparsification_apis
foundation_model_sparsification_apis
compress()
Compressor
SparseConfig
long_sequence_compression_apis
long_sequence_compression_apis
getalibiwindows
getcompressheads
RACompressConfig
RACompressor
RARopeCompressConfig
RARopeCompressor
weight_compress_apis
weight_compress_apis
CompressConfig
Compressor
export()
exportsafetensors()
run()
low_rank_decompression_apis
low_rank_decompression_apis
Decompose
Decompose
init
decomposenetwork
fromdict
fromfile
fromfixed
fromratio
fromvbmf
countparameters
multimodal_inference_apis
multimodal_inference_apis
DitCache
DitCache
DitCacheAdaptor: DiT Model Cache Adapter
DitCacheSearchConfig
sampling_optimization_apis
sampling_optimization_apis
ReStepAdaptor
ReStepSearchConfig
pruning_apis
pruning_apis
PruneConfig
PruneConfig
addblocksparams
setsteps
PruneTorch
PruneTorch
init
analysis
prune
prunebydesc
setimportanceevaluationfunction
setnodereservedratio
prunemodelweight
quantization_apis
quantization_apis
[mindspore]post_training_quantization
[mindspore]post_training_quantization
createquantconfig
quantizemodel
savemodel
[onnx]post_training_quantization
[onnx]post_training_quantization
exportquantonnx
OnnxCalibrator
preprocessfunccoco
preprocessfuncimagenet
QuantConfig
QuantConfig
run()
runquantize
[pytorch]post_training_quantization
[pytorch]post_training_quantization
Calibrator
exportparam
exportquantonnx
exportquantsafetensor()
getquantparams
QuantConfig
quantization_aware_training
quantization_aware_training
QatConfig
qsinqat
saveqsinqatmodel
sparse_acceleration_training_apis
sparse_acceleration_training_apis
sparsemodeldepth
sparsemodelwidth
unified_multimodal_generation_apis
unified_multimodal_generation_apis
quantmodel()
Quantization Session Configuration
settimestepidx()
Public Interface
quick_start
quick_start
msModelSlim Quick Start
release_notes
release_notes
Release Notes
support
support
FAQ
user_guide
user_guide
feature_guide
feature_guide
auto_precision_tuning
auto_precision_tuning
Automatic Tuning Configuration Protocol
Automatic Tuning Usage Guide
quick_quantization_v1
quick_quantization_v1
convert
convert
Weight Conversion Guide
Debug Mode User Guide
Group Processor
Quick Quantization Results
Quick Quantization Guide
sensitive_layer_analysis
sensitive_layer_analysis
Quantization-Sensitive Layer Analysis Tool Usage Guide
traditional_quantization_v0
traditional_quantization_v0
Fake-Quantization Accuracy Test Tool (Precision Tool)
Compression and Structure Optimization (Mainly for Foundation Models)
Foundation Model Quantization and Calibration
Inference Optimization for Multimodal Generative Models
MindSpeed Adapter
Training Acceleration and Model Reconstruction
Quantization Code Samples
msModelSlim Quantization Weight Format
Sparse Training Acceleration
Traditional Model Quantization and Calibration
Quantization and Calibration
model_support
model_support
Foundation Model Support Matrix
quantization_algorithms
quantization_algorithms
auto_tuning_strategies
auto_tuning_strategies
Binary Fallback Tuning Algorithm
Standing High Tuning Algorithm
Standing High With Experience Tuning Algorithm
outlier_suppression_algorithms
outlier_suppression_algorithms
Adapt Rotation: Adaptive Rotation Optimization Algorithm
AWQ: Activation-aware Weight Quantization Algorithm
Flex AWQ SSZ: Flexible Activation-Aware Weight Quantization Smoothing Algorithm
Flex Smooth Quant: Flexible Smooth Quantization Algorithm
Iterative Smooth: Outlier Suppression Algorithm
KVSmooth: Outlier Suppression Algorithm for KVCache Quantization
QuaRot: Rotation-based Outlier Suppression Algorithm
SmoothQuant: Outlier Suppression Algorithm
quantization_algorithms
quantization_algorithms
AutoRound: Low-bit Quantization Algorithm
CeilX: Adaptive Divisor MXFP4 Weight Quantization Algorithm
DualScale: w4a4 Quantization Scheme Description
FA3 Quantization: Flash Attention 3 Activation Quantization Algorithm
Floating-Point Sparsity: ADMM-based Model Sparsification Algorithm
Four Over Six: Adaptive Block Scaling Weight Quantization Algorithm
GPTQ: Weight Quantization Algorithm
Histogram: Activation Quantization Algorithm
KVCache Quantization: Cache Quantization Algorithm
LAOS: W4A4 Quantization Scheme
Linear Quantization Algorithm
MinMax Quantization Algorithm
MSERound: Per-Block MSE Optimal MXFP8 Weight Quantization Algorithm
PDMIX: Phase-wise Mixed Activation Quantization Algorithm
SSZ: Weight Quantization Algorithm
SVDQuant: Post-Training Quantization Algorithm for Diffusion Models Based on Low-Rank Residual Reconstruction
sensitive_layer_analysis
sensitive_layer_analysis
Attention MSE (mse): Sensitive Layer Analysis Algorithm
Kurtosis: Sensitive Layer Analysis Algorithm
Layer-wise MSE (mselayerwise): Sensitive Layer Analysis Algorithm
Model-wise MSE (msemodelwise): Sensitive Layer Analysis Algorithm
Quantile: Sensitive Layer Analysis Algorithm
Std: Sensitive Layer Analysis Algorithm
quantization_formats
quantization_formats
AscendV1 Format Description
compressed-tensors Format Specification
MindStudio ModelSlim User Guide
Project Directory Structure Overview
zh
zh
api_reference
api_reference
cli
cli
msmodelslim analyze 命令行 API 文档
msmodelslim quant 命令行 API 文档
msmodelslim tune 命令行 API 文档
config
config
format
format
ascendv1saver 配置说明
compressedtensors 配置说明
mindieformatsaver 配置说明
processor
processor
adaptrotation 配置说明
autoroundquant 配置说明
awq 配置说明
binaryanalysis 配置说明
binaryoperatorlayerwise 配置说明
binaryoperatormodelwise 配置说明
dynamiccache 配置说明
fa3quant 配置说明
flatquant 配置说明
flexawqssz 配置说明
flexsmoothquant 配置说明
floatsparse 配置说明
group 配置说明
itersmooth 配置说明
kvsmooth 配置说明
linearquant 配置说明
load 配置说明
oasq 配置说明
onlinequarot 配置说明
quarot 配置说明
saver 配置说明
smoothquant 配置说明
svdres 配置说明
trainablelinearquant 配置说明
unaryanalysis 配置说明
task
task
modelslimconvert 配置说明
modelslimv1 配置说明
multimodalsdmodelslimv1 配置说明
multimodalvlmmodelslimv1 配置说明
PracticeConfig 配置说明
tuning
tuning
evaluationserviceoriented 配置说明
binaryfallback 配置说明
standinghigh 配置说明
standinghighwithexperience 配置说明
tuningplan 配置说明
自动调优配置协议说明
1. 定义静态量化模板 (Anchor)
python_api_v0
python_api_v0
distillation_apis
distillation_apis
KnowledgeDistillConfig
KnowledgeDistillConfig
addcustomlossfunc
addintersoftlabel
addoutputsoftlabel
sethardlabel
setteachertrain
getdistillmodel
foundation_model_compression_apis
foundation_model_compression_apis
foundation_model_quantization_apis
foundation_model_quantization_apis
AntiOutlier
AntiOutlierConfig
CalibrationData
FakeQuantizeCalibrator
LayerSelector
AntiOutlierAdapter
CalibratorAdapter
ModelAdapter
process()
Calibrator
QuantConfig
run()
save()
process()
Calibrator
FAQuantizer
quant()
QuantConfig
run()
save()
foundation_model_sparsification_apis
foundation_model_sparsification_apis
compress()
Compressor
SparseConfig
long_sequence_compression_apis
long_sequence_compression_apis
getalibiwindows
getcompressheads
RACompressConfig
RACompressor
RARopeCompressConfig
RARopeCompressor
weight_compress_apis
weight_compress_apis
CompressConfig
Compressor
export()
exportsafetensors()
run()
low_rank_decompression_apis
low_rank_decompression_apis
Decompose
Decompose
init
decomposenetwork
fromdict
fromfile
fromfixed
fromratio
fromvbmf
countparameters
multimodal_inference_apis
multimodal_inference_apis
DitCache
DitCache
DitCacheAdaptor: DiT模型缓存适配器
DitCacheSearchConfig
sampling_optimization_apis
sampling_optimization_apis
ReStepAdaptor
ReStepSearchConfig
pruning_apis
pruning_apis
PruneConfig
PruneConfig
addblocksparams
setsteps
PruneTorch
PruneTorch
init
analysis
prune
prunebydesc
setimportanceevaluationfunction
setnodereservedratio
prunemodelweight
quantization_apis
quantization_apis
[mindspore]post_training_quantization
[mindspore]post_training_quantization
createquantconfig
quantizemodel
savemodel
[onnx]post_training_quantization
[onnx]post_training_quantization
exportquantonnx
OnnxCalibrator
preprocessfunccoco
preprocessfuncimagenet
QuantConfig
QuantConfig
run()
runquantize
[pytorch]post_training_quantization
[pytorch]post_training_quantization
Calibrator
exportparam
exportquantonnx
exportquantsafetensor()
getquantparams
QuantConfig
quantization_aware_training
quantization_aware_training
QatConfig
qsinqat
saveqsinqatmodel
sparse_acceleration_training_apis
sparse_acceleration_training_apis
sparsemodeldepth
sparsemodelwidth
unified_multimodal_generation_apis
unified_multimodal_generation_apis
quantmodel()
量化会话配置
settimestepidx()
公共接口
best_practices
best_practices
DeepSeek-V4-Flash W8A8 一键量化案例
DeepSeek-V4-Pro 新模型 W4A8 量化案例
Kimi-K3 新模型 W4A8 量化案例
msModelSlim量化权重转AutoAWQ&AutoGPTQ使用指南
Qwen3-32B W8A8精度调优案例
Wan2.2 新模型 W4A4F4 量化案例
contributing
contributing
design
design
接口文档自动生成设计
msModelSlim 丰富量化模型支持特性设计说明书
msModelSlim 敏感层分析模块重构特性设计说明书
模型结构提取特性设计说明书
msModelSlim 自动调优加速特性设计说明书
development_guide
development_guide
docs_standards
docs_standards
公共校验清单
量化术语百科校验清单
{{ termname }}
流程指南校验清单
{{workflowname}}流程指南
通用案例校验清单
{{modelname}}-{{scenariotype}}案例
量化配置文档校验清单
{{ configname }} 配置说明
命令行 API 文档校验清单
{{ commandname }} 命令行 API 文档
msModelSlim 架构
编码规范
项目目录结构总览
开发者测试指南
Contributing to Mindstudio ModelSlim
install_guide
install_guide
msModelSlim 安装指南
knowledge_base
knowledge_base
model
model
LLM大模型接入指南
多模态生成模型接入指南
多模态理解模型接入指南
parallel
parallel
data_parallelism
data_parallelism
数据并行使用指南
数据并行
distributed_task_scheduler
distributed_task_scheduler
分布式任务调度器使用指南
分布式任务调度器
expert_parallelism
expert_parallelism
专家并行使用指南
专家并行
ptq
ptq
convert
convert
权重转换量化术语百科词条
权重转换使用指南
dit
dit
DiT 模型接入量化流程指南
Diffusion Transformer（DiT）量化术语百科词条
Diffusion Transformer（DiT）量化使用指南
llm
llm
LLM 模型接入量化流程指南
大语言模型（LLM）量化术语百科词条
大语言模型（LLM）量化使用指南
vlm
vlm
VLM 模型接入量化流程指南
视觉语言模型（VLM）量化术语百科词条
视觉语言模型（VLM）量化使用指南
训练后量化（PTQ）术语百科词条
quantization_algorithms
quantization_algorithms
adapt_rotation
adapt_rotation
term_adapt_rotation
Adapt Rotation 参数配置流程指南
attention_mse
attention_mse
term_attention_mse
Attention MSE 分析配置流程指南
autoround
autoround
term_autoround
AutoRound 参数配置流程指南
awq_smooth
awq_smooth
term_awq_smooth
AWQ 参数配置流程指南
ceil_x
ceil_x
term_ceil_x
usage_ceil_x
daos
daos
DAOS 低比特量化方案 量化术语百科词条
DAOS 参数配置流程指南
dual_scale
dual_scale
term_dual_scale
usage_dual_scale
fa3_quant
fa3_quant
term_fa3_quant
FA3 Quant 参数配置流程指南
flex_awq_ssz
flex_awq_ssz
term_flex_awq_ssz
usage_flex_awq_ssz
flex_smooth_quant
flex_smooth_quant
term_flex_smooth_quant
Flex Smooth Quant 参数配置流程指南
float_sparse
float_sparse
term_float_sparse
浮点稀疏 参数配置流程指南
fouroversix
fouroversix
term_fouroversix
usage_fouroversix
gptq
gptq
term_gptq
usage_gptq
histogram_activation_quantization
histogram_activation_quantization
term_histogram_activation_quantization
usage_histogram_activation_quantization
iterative_smooth
iterative_smooth
term_iterative_smooth
Iterative Smooth 参数配置流程指南
kurtosis
kurtosis
term_kurtosis
usage_kurtosis
kv_smooth
kv_smooth
term_kv_smooth
KV Smooth 参数配置流程指南
kvcache_quant
kvcache_quant
term_kvcache_quant
usage_kvcache_quant
laos
laos
term_laos
usage_laos
linear_quant
linear_quant
term_linear_quant
线性量化参数配置流程指南
minmax
minmax
term_minmax
usage_minmax
mse_layer_wise
mse_layer_wise
term_mse_layer_wise
usage_mse_layer_wise
mse_model_wise
mse_model_wise
term_mse_model_wise
usage_mse_model_wise
mse_round
mse_round
term_mse_round
usage_mse_round
oasq
oasq
OASQ 离群感知平滑量化算法 量化术语百科词条
OASQ 参数配置流程指南
pdmix
pdmix
term_pdmix
usage_pdmix
quantile
quantile
term_quantile
usage_quantile
quarot
quarot
term_quarot
QuaRot 参数配置流程指南
ra_compress
ra_compress
RA Compress 长序列压缩算法词条
RA Compress 使用指南
smooth_quant
smooth_quant
term_smooth_quant
SmoothQuant 参数配置流程指南
ssz
ssz
term_ssz
usage_ssz
std
std
term_std
usage_std
svdquant
svdquant
term_svdquant
usage_svdquant
trainable_linear_quant
trainable_linear_quant
Trainable Linear Quant 可训练线性量化算法 量化术语百科词条
Trainable Linear Quant 参数配置流程指南
quantization_basic
quantization_basic
数据类型：FP16 / BF16
数据类型：FP8（E4M3）
数据类型：INT4
数据类型：INT8
数据类型：MXFP8 / MXFP4
量化与反量化
quantization_format
quantization_format
ascendv1
ascendv1
AscendV1 使用指南
AscendV1 量化格式 量化术语百科词条
compressed_tensors
compressed_tensors
compressed-tensors 使用指南
compressed-tensors 量化格式 量化术语百科词条
mindie_sd
mindie_sd
MindIE-SD 使用指南
MindIE-SD 量化格式 量化术语百科词条
量化格式接入指南
quantization_mode
quantization_mode
fa_quantization
fa_quantization
FA FP8 动态量化
FA INT8 动态量化
FA INT8 PerHead 量化
FA MXFP4 动态量化
FA Q-FP8 动态 K/V-FP8 静态量化
FA Q-INT8 动态 K/V-INT8 静态量化
FA QK-MXFP8 动态 / V-MXFP8 PerChannel 静态量化
kv_cache_quantization
kv_cache_quantization
KVCache-PerChannel 量化
linear_layer_quantization
linear_layer_quantization
浮点稀疏量化
SVDQuant 量化
W4A4 动态量化
W4A4 MX 双 Scale 量化
W4A4 MX 动态量化
W4A8 动态量化
W4A8 MX 动态量化
W8A16 静态量化
W8A8 动态量化
W8A8 FP8 动态量化
W8A8 MX 动态量化
W8A8 PD-Mix 量化
W8A8 稀疏量化
W8A8 静态量化
GEMM
tuning_strategies
tuning_strategies
binary_fallback
binary_fallback
binary_fallback
standing_high
standing_high
standing_high
standing_high_with_experience
standing_high_with_experience
standing_high_with_experience
legal
legal
致msModelSlim使用者
LICENSE声明
msModelSlim安全声明
quick_start
quick_start
msModelSlim 快速入门
release_notes
release_notes
版本说明
support
support
FAQ
部署与推理 FAQ
硬件兼容 FAQ
安装与依赖 FAQ
精度与调优 FAQ
运行与资源 FAQ
告警与提示 FAQ
user_guide
user_guide
traditional_quantization_v0
traditional_quantization_v0
伪量化精度测试工具（Precision Tool）
压缩与结构优化（大模型为主）
大模型量化与校准
Install torchnpu and decord
mindspeed_adapter
训练加速与模型改造
量化代码示例
msmodelslim量化权重格式
稀疏加速训练
传统模型量化与校准
量化与校准
主流模型量化部署流程指南
新模型量化调优流程
量化推理精度异常定位流程指南
process_quantization_precision_tuning
自动调优使用说明
调试模式使用指南
一键量化使用指南
一键量化完整指南
Attention 敏感层分析使用指南
Attention Head 筛选分析使用指南
层级敏感层分析使用指南
线性层敏感层分析使用指南
权重量化使用指南
msmonitor
msmonitor
docs
docs
zh
zh
advanced_features
advanced_features
mindstudiomonitor模块接口参考
Monitor 特性介绍
design
design
MindStudio Monitor特性分析与设计说明书
developer_guide
developer_guide
开发指南
getting_started
getting_started
msMonitor工具安装指南
msMonitor工具快速入门
legal
legal
MindStudio漏洞机制说明
公网地址
msMonitor安全声明
user_guide
user_guide
dyno使用说明
dynolog使用说明
MindSpore框架下msMonitor的使用方法
npu-monitor使用说明
nputrace使用说明
项目目录
FAQ
总体介绍
版本说明
msopcom
msopcom
docs
docs
en
en
contributing
contributing
Contribution Guide
development_guide
development_guide
MindStudio Ops Common Architecture Design
MindStudio Ops Common Development Environment Setup, Compilation, and Unit Testing
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Mechanism
zh
zh
contributing
contributing
贡献指南
development_guide
development_guide
MindStudio Ops Common 架构设计说明
MindStudio Ops Common 开发环境搭建及编译和UT方法
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
msopgen
msopgen
docs
docs
en
en
best_practices
best_practices
Typical Cases of MindStudio Ops Generator
contributing
contributing
Contribution Guide
development_guide
development_guide
msOpGen Architecture Design Specifications
MindStudio msOpGen Development Environment Setup and UT Methods
install_guide
install_guide
MindStudio Ops Generator Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Handling Mechanism Description
overview
overview
Overview
quick_start
quick_start
MindStudio Ops Generator Quick Start
MindStudio Ops System Test Quick Start
release_notes
release_notes
MindStudio Ops Generator Release Notes
support
support
msOpGen FAQ
user_guide
user_guide
MindStudio Ops Generator User Guide
zh
zh
best_practices
best_practices
MindStudio Ops Generator典型案例
contributing
contributing
贡献指南
development_guide
development_guide
msOpGen 架构设计说明书
MindStudio msOpGen 开发环境搭建和UT方法
install_guide
install_guide
msOpGen 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
overview
overview
简介
quick_start
quick_start
msOpGen算子调试工具快速入门
MindStudio Ops System Test快速入门
release_notes
release_notes
MindStudio Ops Generator 版本说明
support
support
msOpGen 常见问题
user_guide
user_guide
MindStudio Ops Generator工具用户指南
msopmodeling
msopmodeling
docs
docs
zh
zh
contributing
contributing
贡献指南
development_guide
development_guide
算子性能建模工具开发指南
install_guide
install_guide
算子性能建模工具安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
算子性能建模工具快速入门
user_guide
user_guide
算子性能建模工具使用指南
msopprof
msopprof
docs
docs
en
en
best_practices
best_practices
Typical Cases
contributing
contributing
Contributing Guide
development_guide
development_guide
MindStudio Ops Profiler Architecture Design Specifications
MindStudio Ops Profiler Development Environment Setup, Compilation, and UT Methods
install_guide
install_guide
MindStudio Ops Profiler Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Declaration
MindStudio Vulnerability Handling Mechanism Description
quick_start
quick_start
MindStudio Ops Profiler Quick Start
release_notes
release_notes
MindStudio Ops Profiler Release Notes
user_guide
user_guide
Constraints and Precautions for MindStudio Ops Profiler
Extended Functions
msOpProf Profile Data
msOpProf Simulator Mode Performance Data
msOpProf Simulator Mode User Guide
msOpProf Usage Scenarios
msOpProf Mode User Guide
zh
zh
best_practices
best_practices
典型案例
contributing
contributing
贡献指南
development_guide
development_guide
MindStudio Ops Profiler 架构设计说明书
msOpProf 开发指南
install_guide
install_guide
msOpProf 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
msOpProf 算子性能调优工具快速入门
release_notes
release_notes
MindStudio Ops Profiler 版本说明
user_guide
user_guide
MindStudio Ops Profiler 工具限制与注意事项
扩展功能
msopprof模式性能数据
msOpProf simulator模式性能数据
msopprof simulator模式用户指南
msOpProf使用场景
msOpProf模式用户指南
msoptuner
msoptuner
docs
docs
en
en
install_guide
install_guide
MindStudio Ops Tuner Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Mechanism
zh
zh
install_guide
install_guide
MindStudio Ops Tuner安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
msot
msot
docs
docs
design
design
26.0.0
26.0.0
检测工具功能设计文档
检测工具功能设计文档
算子检测工具功能设计文档
调优工具功能设计文档
调优工具功能设计文档
MindStudio-Ops-Profiler特性设计说明书
调试工具功能设计文档
调试工具功能设计文档
算子调试工具功能设计文档
26.1.0
26.1.0
检测工具功能设计文档
检测工具功能设计文档
MindStudio Sanitizer 功能设计文档
调优工具功能设计文档
调优工具功能设计文档
MindStudio Ops Profiler 功能设计说明书
调试工具功能设计文档
调试工具功能设计文档
MindStudio Debugger 功能设计说明书
26.2.0
26.2.0
检测工具功能设计文档
检测工具功能设计文档
MindStudio Sanitizer 功能设计文档
调优工具功能设计文档
调优工具功能设计文档
MindStudio Ops Profiler 功能设计说明书
调试工具功能设计文档
调试工具功能设计文档
MindStudio Debugger 功能设计说明书
en
en
common
common
Contribution Workflow and Code Style Guide
MindStudio Tool Development Environment Setup Guide
MindStudio Unified Build Image Creation Guide
contributing
contributing
Contributing Guide
install_guide
install_guide
MindStudio Operator Tools Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Mechanism Description
overview
overview
Introduction
quick_start
quick_start
Ascend AI Operator Development Toolchain Learning Environment Installation Guide
Operator Development Toolchain Quick Start
release_notes
release_notes
MindStudio Operator Tools Release Notes
user_guide
user_guide
msOT Tool Selection Guide
zh
zh
common
common
贡献流程和规范说明
MindStudio 工具开发环境安装指导
MindStudio 统一构建镜像制作指南
contributing
contributing
贡献指南
install_guide
install_guide
msOT 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
overview
overview
简介
quick_start
quick_start
昇腾 AI 算子开发工具链学习环境安装指南
算子开发工具链快速入门
release_notes
release_notes
MindStudio Operator Tools 版本说明
user_guide
user_guide
msOT 工具选型指南
msprobe
msprobe
docs
docs
en
en
baseline
baseline
Precision Pre-check Baseline in MindSpore
Precision Data Collection Baseline in MindSpore
Performance Baseline Report of the Training Status Monitoring Tool
Precision Data Collection Baseline in PyTorch
best_practices
best_practices
Enabling Tools for Common Frameworks
Foundation Model Inference Accuracy Debugging Guide
Foundation Model Training Accuracy Debugging Guide
contributing
contributing
Contributing to MindStudio Probe
development_guide
development_guide
Developer Guide
MindStudio 26.0.0 Precision Debugging Feature Analysis and Design Specification
MindStudio 26.1.0 Precision Debugging Feature Analysis and Design Specification
examples
examples
Configuring Layer Mapping for Hierarchical Model Visualization
Data Collection and Automatic Comparison in MindSpeed and LLamaFactory
install_guide
install_guide
msProbe Installation Guide
legal
legal
Disclaimer
License Description
MindStudio Vulnerability Handling Mechanism Description
Public Network Addresses
msProbe Security Statement
quick_start
quick_start
Quick Start of msProbe in the MindSpore Scenario
msProbe PyTorch Quick Start
support
support
FAQs
user_guide
user_guide
accuracy_checker
accuracy_checker
Offline Precision Pre-check for MindSpore Dynamic Graphs
Offline Precision Pre-check in PyTorch
accuracy_compare
accuracy_compare
ATB Data Precision Comparison
One-Click Precision Comparison for Offline Inference Models
Comparison Result Description
Precision Comparison in MindSpore
Graph Comparison in Hierarchical Visualization (MindSpore)
Comparison of Dump Data Precision for Offline Models
Precision Comparison in PyTorch
Compilation Accuracy Comparison in PyTorch
Graph Comparison in Hierarchical Visualization (PyTorch)
Network-wide Operator Precision Comparison in TorchAir Mode
Trend Visualization
dump
dump
aclgraphdump User Guide
Precision Data Collection in ATB
Configuration File Introduction
Data Conversion
Single-Point Saving Tool
Inference Offline Model Data Collection
Precision Data Collection in MindSpore
Precision Data Collection in MindSpore Dynamic Graph Mode
Precision Data Collection in MSAdapter
Precision Data Collection in PyTorch
Kernel-Level Precision Data Collection in PyTorch
Precision Data Collection in SGLang (SGLang Version < 0.5.11)
Precision Data Collection in SGLang (SGLang Version >= 0.5.11)
Inference in Torch Graph Mode (TorchAir)
Collecting Data for Asynchronous verl Training-Inference Consistency Comparison
Collecting Data for Verifying Data Consistency Between verl Training and Inference Based on FSDP
Collecting Data for Verifying Data Consistency Between verl Training and Inference Based on Megatron
Training-Inference Consistency Monitoring: Token-Level probsdiff Monitoring
Precision Data Collection in the vLLM Scenario
overflow_check
overflow_check
First Network Overflow/Underflow Node Analysis
Checkpoint Comparison
Configuration Check Before Training
msProbe: Functional Modules, Scenarios, and Limitations
Lightweight Training Status Monitoring Tool
Lightweight Training Status Monitoring Tool
Response Anomaly
verl Training-Inference Cross-Instrumentation
verl Hyperparameter Comparison and Key Hyperparameter Verification
zh
zh
baseline
baseline
MindSpore 场景的精度预检基线
MindSpore 场景的精度数据采集基线
训练状态监测工具标准性能基线报告
PyTorch 场景的精度数据采集基线
best_practices
best_practices
基于昇腾的80卡大模型训练确定性排查
基于昇腾的复杂dump数据分析GradNorm NaN定位实践
基于昇腾的DeepSeek-V4 GPQA精度优化
基于昇腾的Gemma模型重复训练Loss差异定位
基于昇腾的GLM5适配与训推一致性优化
基于昇腾的Kimi2.5训推一致性优化
基于昇腾的mmdetection训练NaN问题定位
基于昇腾的MOVA模型NaN问题定位分析
基于昇腾的多模态理解训练NaN问题定位
基于昇腾的Qwen2.5-omni模型确定性排查
基于昇腾的Qwen2.5-omni框架迁移Loss差异定位
基于昇腾的Qwen-Image-Edit模型TP8花图问题定位
强化学习训推一致性排查
基于昇腾的tora模型推理差异定位
基于昇腾的VL模型迁移vLLM精度问题
常见框架dump工具使能
大模型推理精度定位指南
Qwen2.5模型推理回复复读问题
CANN版本升级后的推理请求回复乱码问题
大模型训练精度定位指南
contributing
contributing
为MindStudio Probe贡献
development_guide
development_guide
msProbe 开发指南
MindStudio26.0.0精度调试特性分析与设计说明书
MindStudio26.1.0精度调试特性分析与设计说明书
MindStudio26.2.0精度调试特性分析与设计说明书
examples
examples
模型分级可视化如何配置layer mapping映射文件
MindSpeed&LLamaFactory数据采集和自动比对
install_guide
install_guide
msProbe 安装指南
legal
legal
免责声明
License说明
MindStudio漏洞机制说明
公网地址
msProbe安全声明
quick_start
quick_start
MindSpore场景精度调试工具快速入门
msProbe PyTorch 场景快速入门
support
support
FAQ
user_guide
user_guide
accuracy_checker
accuracy_checker
MindSpore动态图场景离线精度预检
PyTorch场景离线精度预检
accuracy_compare
accuracy_compare
ATB场景精度比对
推理离线模型一键式精度比对
对比结果说明
MindSpore场景精度比对
MindSpore场景分级可视化构图比对
离线模型dump数据精度比对
PyTorch场景精度比对
PyTorch场景编译精度比对
PyTorch场景的分级可视化构图比对
基于torch图模式（torchair）整网算子精度比对
趋势可视化
dump
dump
aclgraphdump 使用指南
ATB场景精度数据采集
配置文件介绍
数据转换功能
单点保存工具
推理离线模型数据采集
MindSpore场景精度数据采集
MindSpore动态图精度数据采集快速入门
MSAdapter场景精度数据采集
PyTorch场景精度数据采集
PyTorch场景精度数据覆盖加载
PyTorch场景kernel级精度数据采集
SGLang精度数据采集（SGLang版本<0.5.11）
SGLang精度数据采集（SGLang版本>=0.5.11）
slime框架训推一致性预处理与数据采集
slime框架训推精度数据采集
基于torch图模式（torchair）推理场景
异步架构verl训推一致性比对数据采集
fsdp训练后端verl训推一致性比对数据采集
megatron训练后端verl训推一致性比对数据采集
训推一致性监控：逐Token级别的probsdiff监控
verl V1 Trainer训推一致性比对数据采集
vLLM采集
overflow_check
overflow_check
整网首个溢出节点分析
Checkpoint比对
训练前配置检查
msProbe 工具功能模块简介、适用场景和当前版本局限性
Monitor训练状态轻量化监测工具
训练状态轻量化监测工具
推理异常检测
训推权重一致性验证定位指南
slime超参比对与关键超参校验
VeRL 训推交叉打桩功能指南
verl超参比对与关键超参校验
msprof
msprof
docs
docs
en
en
best_practices
best_practices
Troubleshooting Guide: Asynchronous Scheduling Does Not Take Effect
msProf Use Case: Performance Analysis of a ResNet50 Inference Model
Request Length Imbalance Within the Same Batch
1. Background
Performance Degradation Caused by Misaligned Communication Addresses
CPU Cache Miss Resource Contention and Constraints
Frequent CPU Thread Switching
1. Background
DP Load Imbalance
EP Load Imbalance
Troubleshooting Guide: Framework Scheduling and Dispatch Synchronization
GIL Contention Analysis
IR Interrupt Preemption Analysis
Frequent Kernel Thread Switching Analysis
Insufficient KV Blocks
Impact of KV Cache Transfer on Model Performance
Memory Fragmentation Analysis
Memory Leak Analysis
Model Performance Degradation Causing SLO Degradation
Analysis of Excessive Preprocessing and Postprocessing Time in Model Inference
Multi-Instance Load Imbalance
1. Background
PrefixCache Misses
Pthread Lock Wait
High Python GC Overhead
Troubleshooting Guide: Long Wait Time for Model Inference Requests
Analysis of Excessive Sampler Execution Time
Excessive Scheduler Execution Time
Optimization Guide for Large Performance Gaps Between Serving and Pure-Model Inference
1. Background
High-Latency System Calls
contributing
contributing
Contributing to MindStudio Profiler
development_guide
development_guide
Developer Guide
Background
MindStudio Profiler Feature Analysis and Design Specifications
install_guide
install_guide
MindStudio Profiler Installation Guide
legal
legal
Disclaimer
License
MindStudio Profiler Security Statement
quick_start
quick_start
msProf Quick Start
support
support
FAQ
user_guide
user_guide
Extended Functions
msProf Parsing Tool
Profile Data File Reference
Profile Data File Reference (DB)
zh
zh
best_practices
best_practices
ResNet50推理模型性能分析
代码高耗时函数段优化
通信地址不对齐导致性能下降
CPU Cache Miss资源冲突与受限
CPU 线程频繁切换
下发性能不及预期
GIL锁抢占问题分析
IRQ中断打断问题分析
内核进程频繁切换
内存碎片
内存泄漏
算子编译高耗时
Pthread线程锁等待
Python GC回收高耗时
同步接口频繁调用
系统调用高耗时函数
contributing
contributing
为MindStudio Profiler贡献
development_guide
development_guide
msProf 开发指南
背景描述（Background）
MindStudio Profiler 性能调优特性分析与设计说明书
install_guide
install_guide
msProf 安装指南
legal
legal
免责声明
License
msProf安全声明
quick_start
quick_start
msProf 快速入门
support
support
FAQ
user_guide
user_guide
扩展功能
msProf解析工具
profile_data_file_references
DB格式性能数据文件参考
msprof-analyze
msprof-analyze
docs
docs
en
en
advanced_features
advanced_features
calibratenpugpu
Cluster Operator Duration Analysis
clustertimecomparesummary
clustertimesummary
communicationbottleneck
computationalopmasking
Custom Analysis Rule Development Guide
exportsummary
freeanalysis
modulestatistic
operatormfu
ppchart
Table Structures of Parsing Results and clusteranalysis.db Deliverables
development_guide
development_guide
MindStudio Profiler Analyze Feature Analysis and Design Specifications
Developer Guide
install_guide
install_guide
msprof-analyze Installation Guide
legal
legal
Disclaimer
License Notice
MindStudio Profiler Analyze Security Statement
quick_start
quick_start
msprof-analyze Quick Start
user_guide
user_guide
advisor
clusteranalyse
compare
AICPU Operator Replacement Examples
Examples for Fused Operator API Replacement During Migration to Ascend
zh
zh
advanced_features
advanced_features
NPU 和 GPU 性能数据拆解比对
集群算子耗时分析
集群性能数据细粒度比对
集群性能数据细粒度拆解
通信瓶颈分析
集群算子掩盖线性度分析
自定义分析规则开发指导
集群算子信息导出
空闲时间原因分析
性能数据模型结构拆解
算子 MFU 分析
pp流水图数据分析
解析结果和clusteranalysis.db交付件表结构说明
development_guide
development_guide
MindStudio Profiler Analyze特性分析与设计说明书
msprof-analyze 开发指南
install_guide
install_guide
msprof-analyze 安装指南
legal
legal
免责声明
License 声明
msprof-analyze安全声明
quick_start
quick_start
msprof-analyze 快速入门
user_guide
user_guide
专家建议
cluster_analyse_instruct
性能比对
AICPU算子替换样例
昇腾迁移融合算子API替换样例
mspti
mspti
docs
docs
en
en
api_reference
api_reference
c_api
c_api
context
context
msptiActivity
msptiActivityApi
msptiActivityCommunication
msptiActivityDisable
msptiActivityDisableMarkerDomain
msptiActivityEnable
msptiActivityEnableMarkerDomain
msptiActivityExternalCorrelation
msptiActivityFlag
msptiActivityFlushAll
msptiActivityFlushPeriod
msptiActivityGetNextRecord
msptiActivityHccl
msptiActivityIsEnabled
msptiActivityKernel
msptiActivityKind
msptiActivityMarker
msptiActivityMemcpy
msptiActivityMemcpyKind
msptiActivityMemory
msptiActivityMemoryKind
msptiActivityMemoryOperationType
msptiActivityMemset
msptiActivityPopExternalCorrelationId
msptiActivityPushExternalCorrelationId
msptiActivityRegisterCallbacks
msptiActivitySourceKind
msptiApiCallbackSite
msptiBuffersCallbackCompleteFunc
msptiBuffersCallbackRequestFunc
msptiCallbackData
msptiCallbackDomain
msptiCallbackFunc
msptiCallbackId
msptiCallbackIdHccl
msptiCallbackIdRuntime
msptiCommunicationDataType
msptiEnableCallback
msptiEnableDomain
msptiExternalCorrelationKind
msptiObjectId
msptiResult
msptiSubscribe
msptiSubscriberHandle
msptiUnsubscribe
python_api
python_api
context
context
CommunicationData
CommunicationMonitor.flushall
CommunicationMonitor.setbuffersize
CommunicationMonitor.start
CommunicationMonitor.stop
HcclData
HcclMonitor.flush\all
HcclMonitor.set\buffer\size
HcclMonitor.start
HcclMonitor.stop
KernelData
KernelMonitor.flush\all
KernelMonitor.set\buffer\size
KernelMonitor.start
KernelMonitor.stop
MarkerData
MsptiActivityFlag
MsptiActivityKind
MsptiActivitySourceKind
MsptiCommunicationDataType
MsptiResult
MstxMonitor.disable\domain
MstxMonitor.enable\domain
MstxMonitor.flush\all
MstxMonitor.set\buffer\size
MstxMonitor.start
MstxMonitor.stop
RangeMarkerData
best_practices
best_practices
msPTI Best Practices and Typical Use Cases
contributing
contributing
Contributing to the MindStudio Profiler Tools Interface
development_guide
development_guide
Development Guide
msPTI Feature Design Specifications
install_guide
install_guide
msPTI Installation Guide
legal
legal
Disclaimer
License
msPTI Security Statement
quick_start
quick_start
msPTI Quick Start
support
support
msPTI FAQ
user_guide
user_guide
Activity API Usage Guide
Callback API User Guide
msPTI User Guide
Python API Usage Guide
msPTI Tool Sample Guide
zh
zh
api_reference
api_reference
c_api
c_api
context
context
msptiActivity
msptiActivityApi
msptiActivityAttribute
msptiActivityCommunication
msptiActivityDisable
msptiActivityDisableMarkerDomain
msptiActivityEnable
msptiActivityEnableMarkerDomain
msptiActivityExternalCorrelation
msptiActivityFlag
msptiActivityFlushAll
msptiActivityFlushPeriod
msptiActivityGetAttribute
msptiActivityGetEnabledKinds
msptiActivityGetNextRecord
msptiActivityGetNumDroppedRecords
msptiActivityGetStructSize
msptiActivityHccl
msptiActivityIsEnabled
msptiActivityKernel
msptiActivityKind
msptiActivityMarker
msptiActivityMemcpy
msptiActivityMemcpyKind
msptiActivityMemory
msptiActivityMemoryKind
msptiActivityMemoryOperationType
msptiActivityMemset
msptiActivityObjectKind
msptiActivityOverhead
msptiActivityOverheadKind
msptiActivityPopExternalCorrelationId
msptiActivityPushExternalCorrelationId
msptiActivityRegisterCallbacks
msptiActivityRegisterTimestampCallback
msptiActivitySetAttribute
msptiActivitySourceKind
msptiApiCallbackSite
msptiBuffersCallbackCompleteFunc
msptiBuffersCallbackRequestFunc
msptiCallbackData
msptiCallbackDomain
msptiCallbackFunc
msptiCallbackId
msptiCallbackIdHccl
msptiCallbackIdRuntime
msptiCommunicationDataType
msptiDomainTable
msptiEnableAllDomains
msptiEnableCallback
msptiEnableDomain
msptiExternalCorrelationKind
msptiGetCallbackName
msptiGetCallbackState
msptiGetEnabledCallbacks
msptiGetTimestamp
msptiGetVersion
msptiIsTracingSessionRunning
msptiObjectId
msptiResult
msptiSubscribe
msptiSubscriberHandle
msptiSupportedDomains
msptiTimestampCallbackFunc
msptiUnsubscribe
python_api
python_api
context
context
CommunicationData
CommunicationMonitor.flush\all
CommunicationMonitor.set\buffer\size
CommunicationMonitor.start
CommunicationMonitor.stop
HcclData
HcclMonitor.flush\all
HcclMonitor.set\buffer\size
HcclMonitor.start
HcclMonitor.stop
KernelData
KernelMonitor.flush\all
KernelMonitor.set\buffer\size
KernelMonitor.start
KernelMonitor.stop
MarkerData
MsptiActivityFlag
MsptiActivityKind
MsptiActivitySourceKind
MsptiCommunicationDataType
MsptiResult
MstxMonitor.disable\domain
MstxMonitor.enable\domain
MstxMonitor.flush\all
MstxMonitor.set\buffer\size
MstxMonitor.start
MstxMonitor.stop
RangeMarkerData
best_practices
best_practices
msPTI 最佳实践与典型案例
contributing
contributing
为 MindStudio Profiler Tools Interface 贡献
development_guide
development_guide
msPTI 开发指南
msPTI特性设计说明书
install_guide
install_guide
msPTI 安装指南
legal
legal
免责声明
License
msPTI安全声明
quick_start
quick_start
msPTI 快速入门
support
support
msPTI 常见问题（FAQ）
user_guide
user_guide
Activity API 使用指南
Callback API 使用指南
msPTI 用户指南
Python API 使用指南
msPTI工具样例指南
mssanitizer
mssanitizer
docs
docs
en
en
api_reference
api_reference
MindStudio Sanitizer API Reference
best_practices
best_practices
Basic Cases
contributing
contributing
Contribution Guide
development_guide
development_guide
msSanitizer Architecture Design Specifications
MindStudio Sanitizer Development Environment Setup, Build, and UT Methods
install_guide
install_guide
MindStudio Sanitizer Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Handling Mechanism Description
quick_start
quick_start
msSanitizer Quick Start
release_notes
release_notes
MindStudio Sanitizer Release Notes
support
support
MindStudio Sanitizer FAQs
user_guide
user_guide
Enabling Full Check
MindStudio Sanitizer User Guide
zh
zh
api_reference
api_reference
MindStudio Sanitizer 对外接口使用说明
best_practices
best_practices
基础案例
contributing
contributing
贡献指南
development_guide
development_guide
msSanitizer 架构设计说明书
msSanitizer 开发指南
install_guide
install_guide
msSanitizer 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
quick_start
quick_start
msSanitizer 算子检测工具快速入门
release_notes
release_notes
MindStudio Sanitizer 版本说明
support
support
MindStudio Sanitizer常见问题
user_guide
user_guide
开启全量检测
MindStudio Sanitizer 使用指南
msserviceprofiler
msserviceprofiler
docs
docs
community_extension_docs
community_extension_docs
社区扩展与插件列表
社区拓展指南：如何提交扩展
design
design
MindStudio Service Profiler 特性设计说明书
【B050】LLM推理监控平台设计文档
ms-service-metric 可扩展指标架构重构设计文档
三元关系详设
vLLM Hook Tracing 特性详细设计
en
en
cpp_api
cpp_api
serving_tuning
serving_tuning
ArrayAttr
ArrayResource
Attr Functions
Domain
Event
GetMsg
IsEnable
Launch
Link
Macro Definition
Metric
MetricInc
MetricScope
MetricScopeAsGlobal
MetricScopeAsReqID
NumArrayAttr
Resource
SpanEnd
SpanStart
trace_data_monitoring
trace_data_monitoring
Activate
addResAttribute
Attach
End
ExtractAndAttach
GetCurrent
GetTraceCtx
IsEnable
SetAttribute
SetStatus
Span
Span Class
StartSpanAsActive
TraceContext Class
Tracer Class
Unattach
developer_guide
developer_guide
msServiceProfiler Development Guide
legal
legal
Disclaimer
License
python_api
python_api
context
context
\\enter\\/\\exit\\
attr
domain
event
get\msg
init
launch
link
metric
metric\inc
metric\scope
metric\scope\as\req\id
res
span\end
span\start
Project Directory
msServiceProfiler Compare Tool
msServiceProfiler Installation Guide
msServiceProfiler Multi Analyze
msServiceProfiler
msServiceProfiler Trace for Data Monitoring
Overview
Public Network Address
Quick Start Guide for msServiceProfiler
Version Description
msServiceProfiler Security Statement
Service Performance Split Tool
Service Profiling Advisor
Serviceparam Optimizer
Custom Plugin Developer Guide
SGLang Service Profiler User Guide
vLLM Serving Prometheus Metric Monitoring Tool User Guide
vLLM Service Profiler User Guide
MindStudio Vulnerability Handling Mechanism Description
zh
zh
best_practices
best_practices
异步双发未生效
同一Batch内请求长度不均
DP负载不均
EP负载不均
框架调度下发不同步
新测评数据集触发 KVCache 容量瓶颈
KVCache传输影响模型性能
模型性能劣化导致SLO劣化
模型前后处理耗时过长
多实例负载不均
PrefixCache未命中
模型推理请求等待时间过长
sampler执行耗时过长
scheduler耗时过长
服务化与纯模型性能差异过大
动态 EPLB 开启后投机推理接受率下降
cpp_api
cpp_api
serving_tuning
serving_tuning
ArrayAttr
ArrayResource
Attr系列
Domain
Event
GetMsg
IsEnable
Launch
Link
宏定义
Metric
MetricInc
MetricScope
MetricScopeAsGlobal
MetricScopeAsReqID
NumArrayAttr
Resource
SpanEnd
SpanStart
trace_data_monitoring
trace_data_monitoring
Activate
addResAttribute
Attach
End
ExtractAndAttach
GetCurrent
GetTraceCtx
IsEnable
SetAttribute
SetStatus
Span
Span类
StartSpanAsActive
TraceContext类
Tracer类
Unattach
developer_guide
developer_guide
msServiceProfiler 开发指南
legal
legal
免责声明
License
python_api
python_api
context
context
\\enter\\/\\exit\\
attr
domain
event
get\msg
init
launch
link
metric
metric\inc
metric\scope
metric\scope\as\req\id
res
span\end
span\start
项目目录
服务化性能数据比对工具
msServiceProfiler 安装指南
服务化多维度解析工具
服务化调优工具
msServiceProfiler Trace数据监测
简介
公网地址
服务化性能调优工具快速入门
版本说明
msServiceProfiler安全声明
服务化拆解工具
服务化专家建议工具
服务化自动寻优工具
自定义插件开发指导
SGLang 服务化性能采集工具使用指南
vLLM-Ascend 可观测性性能诊断指南
vLLM Hook Tracing 使用指南
vLLM 服务化 Prometheus 数据监测工具使用指南
vLLM 服务化性能采集工具使用指南
MindStudio 漏洞机制说明
{% include-markdown "../CONTRIBUTING.md" %}
mstt
mstt
docs
docs
en
en
common
common
Contribution Workflow and Guidelines
contributing
contributing
Contribution Guide
legal
legal
Disclaimer
LICENSE Statement
MindStudio Vulnerability Mechanism Description
Public Network Addresses
msTT Security Statement
quick_start
quick_start
Training Development Toolchain Quick Start
user_guide
user_guide
msTT Tool Selection Guide
zh
zh
common
common
贡献流程和规范说明
contributing
contributing
贡献指南
legal
legal
免责声明
LICENSE声明
MindStudio漏洞机制说明
公网地址
msTT安全声明
quick_start
quick_start
训练开发工具链快速入门
user_guide
user_guide
msTT 工具选型指南
mstx
mstx
docs
docs
en
en
api_reference
api_reference
Common
Common
mstxDomainCreateA
mstxDomainDestroy
mstxDomainMarkA
mstxDomainRangeEnd
mstxDomainRangeStartA
mstxGetToolId
mstxMarkA
mstxRangeEnd
mstxRangeStartA
Mem
Mem
mstxMemHeapRegister
mstxMemHeapUnregister
mstxMemPermissionsAssign
mstxMemRegionsRegister
mstxMemRegionsUnregister
contributing
contributing
Contribution Guide
development_guide
development_guide
MindStudio Tools Extension Library Development Guide
install_guide
install_guide
MindStudio Tools Extension Library Installation Guide
legal
legal
Disclaimer
LICENSE Statement
Security Statement
MindStudio Vulnerability Mechanism
release_notes
release_notes
MindStudio Tools Extension Library Release Notes
zh
zh
api_reference
api_reference
Common
Common
mstxDomainCreateA
mstxDomainDestroy
mstxDomainMarkA
mstxDomainRangeEnd
mstxDomainRangeStartA
mstxGetToolId
mstxMarkA
mstxRangeEnd
mstxRangeStartA
Mem
Mem
mstxMemHeapRegister
mstxMemHeapUnregister
mstxMemPermissionsAssign
mstxMemRegionsRegister
mstxMemRegionsUnregister
contributing
contributing
贡献指南
development_guide
development_guide
MindStudio Tools Extension Library 开发指南
install_guide
install_guide
msTX 安装指南
legal
legal
免责声明
LICENSE声明
安全声明
MindStudio 漏洞机制说明
release_notes
release_notes
MindStudio Tools Extension Library 版本说明
msagent
msagent
docs
docs
zh
zh
agent_guide
agent_guide
Accuracy 精度调试
Minos 文档辅助
Modeling 仿真建模
Operator 算子调优
Profiler 性能调优
Quantizer 模型量化调优
SpecTrainer 投机解码 on-policy 重采样
best_practices
best_practices
torchnpu.profiler 接口适配最佳实践
精度问题——NaN溢出问题最佳实践
模型融合算子识别与替换最佳实践
算子级 MFU 实测分析最佳实践
模型量化自动调优最佳实践
精度问题——梯度尖刺（Gradient Spike）最佳实践
design
design
msAgent 设计文档
msprof-mcp 设计文档
NPU平台内存快照分析设计文档
developer_guide
developer_guide
添加自定义 Agent
1. 概述
编译与打包
贡献指南
ReadTheDocs 本地验证说明
版本与兼容性
example
example
Accuracy 使用示例
getting_started
getting_started
msAgent安装指南
msAgent快速入门
legal
legal
免责声明
LICENSE1
LICENSE2
LICENSE_intro
安全声明
user_guide
user_guide
Agent YAML Tool/Skills 过滤规则说明
配置与扩展
上下文压缩使用指南
📝 document-ux-review 使用说明
FAQ
接入 msAgent 调试调优能力
Retry Middleware 使用指南（Deepagents/LangChain）
msAgent使用指南
msboost
msboost
docs
docs
zh
zh
install_guide
install_guide
MindStudio Boost 安装指南
release_notes
release_notes
MindStudio Boost 版本发布说明
user_guide
user_guide
MindStudio Boost 使用指南
msmonitor-dyno
msmonitor-dyno
docs
docs
en
en
advanced_features
advanced_features
mindstudiomonitor Interfaces
Monitor Features
best_practices
best_practices
msMonitor Use Cases
contributing
contributing
Contributing to MindStudio Monitor
development_guide
development_guide
Development Guide
MindStudio Monitor Feature Analysis and Design Specification
install_guide
install_guide
msMonitor Tool Installation Guide
legal
legal
Disclaimer
License
MindStudio Vulnerability Handling Mechanism
Public Network Address
msMonitor Security Statement
quick_start
quick_start
msMonitor Quick Start
support
support
FAQs
user_guide
user_guide
dyno Usage Guide
dynolog Usage Guide
Using msMonitor with the MindSpore Framework
npu-monitor Usage Guide
nputrace Usage Guide
Overview
zh
zh
advanced_features
advanced_features
mindstudiomonitor模块接口参考
Monitor 特性介绍
best_practices
best_practices
msMonitor 使用案例
contributing
contributing
为MindStudio Monitor贡献
design
design
msmonitor DCMI 采集 + Chrome Trace 分层可视化 技术方案设计（RFC）
development_guide
development_guide
msMonitor 开发指南
MindStudio Monitor特性分析与设计说明书
install_guide
install_guide
msMonitor 安装指南
legal
legal
免责声明
License
MindStudio漏洞机制说明
公网地址
msMonitor安全声明
quick_start
quick_start
msMonitor工具快速入门
support
support
FAQ
user_guide
user_guide
dyno使用说明
dynolog使用说明
MindSpore框架下msMonitor的使用方法
npu-monitor使用说明
nputrace使用说明
总体介绍
msmonitor-whl
msmonitor-whl
docs
docs
en
en
advanced_features
advanced_features
mindstudiomonitor Interfaces
Monitor Features
best_practices
best_practices
msMonitor Use Cases
contributing
contributing
Contributing to MindStudio Monitor
development_guide
development_guide
Development Guide
MindStudio Monitor Feature Analysis and Design Specification
install_guide
install_guide
msMonitor Tool Installation Guide
legal
legal
Disclaimer
License
MindStudio Vulnerability Handling Mechanism
Public Network Address
msMonitor Security Statement
quick_start
quick_start
msMonitor Quick Start
support
support
FAQs
user_guide
user_guide
dyno Usage Guide
dynolog Usage Guide
Using msMonitor with the MindSpore Framework
npu-monitor Usage Guide
nputrace Usage Guide
Overview
zh
zh
advanced_features
advanced_features
mindstudiomonitor模块接口参考
Monitor 特性介绍
best_practices
best_practices
msMonitor 使用案例
contributing
contributing
为MindStudio Monitor贡献
design
design
msmonitor DCMI 采集 + Chrome Trace 分层可视化 技术方案设计（RFC）
development_guide
development_guide
msMonitor 开发指南
MindStudio Monitor特性分析与设计说明书
install_guide
install_guide
msMonitor 安装指南
legal
legal
免责声明
License
MindStudio漏洞机制说明
公网地址
msMonitor安全声明
quick_start
quick_start
msMonitor工具快速入门
support
support
FAQ
user_guide
user_guide
dyno使用说明
dynolog使用说明
MindSpore框架下msMonitor的使用方法
npu-monitor使用说明
nputrace使用说明
总体介绍
{% include-markdown "../README.md" %}
¶
回到页面顶部