Skip to content
Robin's AI Lab
Search
⌘
Ctrl
K
Main Navigation
首页
AI 技术
🦙 大语言模型
🤗 强化微调
✌️ 推理部署
🏗️ 模型架构
🚀 大规模并行
🍭 扩散模型
🍔 多模态
🍒 强化学习
🇨🇳 Ascend生态
更多
⚒️ 效率工具
📖 深度学习百科
🤖 机器学习百科
🔗 外部链接
主题切换
菜单
回到顶部
文章目录
Mindspeed 特性汇总
类别
特性
文档链接
本地跳转
并行与分布式特性
Pipeline 并行
Pipeline Parallel
pipeline-parallel.md
pipeline-parallel.md
Virtual Pipeline Parallel
virtual-pipeline-parallel.md
virtual-pipeline-parallel.md
Unaligned Pipeline
unaligned-pipeline.md
unaligned-pipeline.md
Nanopipe Pipeline Parallel
nanopipe-pipeline-parallel.md
nanopipe-pipeline-parallel.md
Multi Parameter Pipeline
multi_parameter_pipeline.md
multi_parameter_pipeline.md
Tensor 并行
Tensor Parallel
tensor-parallel.md
tensor-parallel.md
Tensor Parallel 2D
tensor-parallel-2d.md
tensor-parallel-2d.md
Sequence 并行
Sequence Parallel
sequence-parallel.md
sequence-parallel.md
Conv3D Sequence Parallel
conv3d_sequence_paralle.md
conv3d_sequence_paralle.md
Context 并行
Ulysses Context Parallel
ulysses-context-parallel.md
ulysses-context-parallel.md
Unaligned Ulysses Context Parallel
unaligned-ulysses-context-parallel.md
unaligned-ulysses-context-parallel.md
Ring Attention Context Parallel
ring-attention-context-parallel.md
ring-attention-context-parallel.md
Hybrid Context Parallel
hybrid-context-parallel.md
hybrid-context-parallel.md
Context Parallelism KV Cache
context_parallelism_kv_cache.md
context_parallelism_kv_cache.md
Data 并行
Data Parallel
data-parallel.md
data-parallel.md
Async DDP
async-ddp.md
async-ddp.md
Async DDP Param Gather
async-ddp-param-gather.md
async-ddp-param-gather.md
内存与优化特性
内存管理
Adaptive Memory
adaptive-memory.md
adaptive-memory.md
Smart Swap
smart_swap.md
smart_swap.md
Swap Attention
swap_attention.md
swap_attention.md
Swap Optimizer
swap-optimizer.md
swap-optimizer.md
重计算
Recomputation
recomputation.md
recomputation.md
Recompute Independent Pipelining
recompute_independent_pipelining.md
recompute_independent_pipelining.md
Norm Recompute
norm-recompute.md
norm-recompute.md
Activation Function Recompute
activation-function-recompute.md
activation-function-recompute.md
优化器
Virtual Optimizer
virtual-optimizer.md
virtual-optimizer.md
Distributed Optimizer
distributed-optimizer.md
distributed-optimizer.md
Fused EMA AdamW Optimizer
fused_ema_adamw_optimizer.md
fused_ema_adamw_optimizer.md
Reuse FP32 Param
reuse-fp32-param.md
reuse-fp32-param.md
注意力与模型架构特性
注意力机制
Flash Attention
flash-attention.md
flash-attention.md
Fusion Attention V2
fusion-attn-v2.md
fusion-attn-v2.md
Generate Mask
generate-mask.md
generate-mask.md
模型组件
SwiGLU
swiglu.md
swiglu.md
RMS Norm
rms_norm.md
rms_norm.md
Rotary Embedding
rotary-embedding.md
rotary-embedding.md
ALiBi
alibi.md
alibi.md
Shared Experts
shared-experts.md
shared-experts.md
Noop Layers
noop-layers.md
noop-layers.md
通信与性能特性
通信优化
Communication Over Computation
communication-over-computation.md
communication-over-computation.md
Async Log Allreduce
async-log-allreduce.md
async-log-allreduce.md
HCCL Group Buffer Set
hccl-group-buffer-set.md
hccl-group-buffer-set.md
HCCL Replace Gloo
hccl-replace-gloo.md
hccl-replace-gloo.md
Double Ring
double-ring.md
double-ring.md
性能优化
NPU Matmul Add
npu_matmul_add.md
npu_matmul_add.md
NPU Deterministic
npu_deterministic.md
npu_deterministic.md
Ops FLOPS Calculation
ops_flops_cal.md
ops_flops_cal.md
自动化与流水线特性
自动化
Automated Pipeline
automated-pipeline.md
automated-pipeline.md
Auto Settings
auto_settings.md
auto_settings.md
Automatic Parallelism
Automatic_Parallelism.md
Automatic_Parallelism.md
流水线特性
Variable Sequence Lengths
variable_seq_lengths.md
variable_seq_lengths.md
Multi Parameter Pipeline and Variable Seq Lengths
multi_parameter_pipeline_and_variable_seq_lengths.md
multi_parameter_pipeline_and_variable_seq_lengths.md
DualPipeV
dualpipev.md
dualpipev.md
特殊功能
压缩
Compress Dense
compress-dense.md
compress-dense.md
层特定
Layer ZeRO
LayerZeRO.md
LayerZeRO.md
其他
EOD Reset
eod-reset.md
eod-reset.md
MC2
mc2.md
mc2.md
Unaligned Linear
unaligned_linear.md
unaligned_linear.md