Graph

知识图谱元数据

可视化是第二阶段。MVP 先展示由内容 frontmatter 生成的节点和边, 用来验证底层图谱数据模型。

节点

类型名称
概念Action Rate Multimodality
概念adaptive normalization
概念adaRMS
概念Advantage
概念Attention Mask
概念Behavior Cloning
概念Block-Causal Attention
概念Brownian motion
概念conditional action distribution
概念DAgger
概念Diffusion model
概念Distribution Shift
概念Flow Matching
概念LayerNorm
概念Offline RL
概念OOD Action
概念Padding Mask
概念Policy Improvement
概念Post-Norm
概念Pre-Norm
概念Prefix-LM
概念residual connection
概念Residual Stream
概念RMSNorm
概念Stochastic Differential Equation
概念timestep conditioning
概念Value Function
概念π₀
概念π₀.₅
概念π₀.₅ Action Expert
博客从 Behavior Cloning 到 DAgger,再到 Offline RL:监督、分布与价值
博客从 Diffusion SDE 到 π₀.₅ Flow Matching Action Head
博客从 Flow Matching 原理出发的机器人数据采集指导
博客从 make_attn_mask 理解 Attention 的完整机制
博客从 π₀.₅ 的 adaRMS 理解 Transformer 归一化
学习线程从 Diffusion SDE 到 π₀.₅ Flow Matching Action Head
学习线程理解 pi0.5 中的 adaRMS

边

类型来源目标
相关concept:action-rate-multimodalityconcept:conditional-action-distribution
相关concept:action-rate-multimodalityconcept:flow-matching
相关concept:action-rate-multimodalityconcept:distribution-shift
相关concept:adaptive-normalizationconcept:adarms
相关concept:adaptive-normalizationconcept:timestep-conditioning
相关concept:adaptive-normalizationconcept:rmsnorm
相关concept:adarmsconcept:rmsnorm
相关concept:adarmsconcept:adaptive-normalization
相关concept:adarmsconcept:timestep-conditioning
相关concept:adarmsconcept:flow-matching
相关concept:adarmsconcept:pre-norm
相关concept:adarmsconcept:residual-stream
相关concept:advantageconcept:value-function
相关concept:advantageconcept:offline-rl
相关concept:advantageconcept:policy-improvement
相关concept:attention-maskconcept:block-causal-attention
相关concept:attention-maskconcept:padding-mask
相关concept:attention-maskconcept:prefix-lm
相关concept:behavior-cloningconcept:distribution-shift
相关concept:behavior-cloningconcept:dagger
相关concept:behavior-cloningconcept:offline-rl
相关concept:block-causal-attentionconcept:attention-mask
相关concept:block-causal-attentionconcept:prefix-lm
相关concept:brownian-motionconcept:stochastic-differential-equation
相关concept:brownian-motionconcept:diffusion-model
相关concept:conditional-action-distributionconcept:flow-matching
相关concept:conditional-action-distributionconcept:pi05-action-expert
相关concept:daggerconcept:distribution-shift
相关concept:daggerconcept:behavior-cloning
相关concept:daggerconcept:offline-rl
相关concept:diffusion-modelconcept:stochastic-differential-equation
相关concept:diffusion-modelconcept:brownian-motion
相关concept:diffusion-modelconcept:flow-matching
相关concept:distribution-shiftconcept:behavior-cloning
相关concept:distribution-shiftconcept:dagger
相关concept:flow-matchingconcept:timestep-conditioning
相关concept:flow-matchingconcept:adarms
相关concept:layernormconcept:rmsnorm
相关concept:layernormconcept:pre-norm
相关concept:layernormconcept:post-norm
相关concept:offline-rlconcept:value-function
相关concept:offline-rlconcept:advantage
相关concept:offline-rlconcept:ood-action
相关concept:offline-rlconcept:policy-improvement
相关concept:offline-rlconcept:behavior-cloning
相关concept:ood-actionconcept:offline-rl
相关concept:ood-actionconcept:value-function
相关concept:ood-actionconcept:distribution-shift
相关concept:padding-maskconcept:attention-mask
相关concept:padding-maskconcept:block-causal-attention
相关concept:policy-improvementconcept:value-function
相关concept:policy-improvementconcept:advantage
相关concept:policy-improvementconcept:offline-rl
相关concept:post-normconcept:pre-norm
相关concept:post-normconcept:residual-connection
相关concept:post-normconcept:residual-stream
相关concept:pre-normconcept:residual-stream
相关concept:pre-normconcept:residual-connection
相关concept:pre-normconcept:post-norm
相关concept:pre-normconcept:rmsnorm
相关concept:pre-normconcept:layernorm
相关concept:prefix-lmconcept:block-causal-attention
相关concept:prefix-lmconcept:attention-mask
相关concept:residual-connectionconcept:residual-stream
相关concept:residual-connectionconcept:pre-norm
相关concept:residual-connectionconcept:post-norm
相关concept:residual-streamconcept:residual-connection
相关concept:residual-streamconcept:pre-norm
相关concept:residual-streamconcept:post-norm
相关concept:residual-streamconcept:rmsnorm
相关concept:residual-streamconcept:adarms
相关concept:rmsnormconcept:layernorm
相关concept:rmsnormconcept:adarms
相关concept:rmsnormconcept:adaptive-normalization
相关concept:rmsnormconcept:pre-norm
相关concept:stochastic-differential-equationconcept:brownian-motion
相关concept:stochastic-differential-equationconcept:diffusion-model
相关concept:stochastic-differential-equationconcept:flow-matching
相关concept:timestep-conditioningconcept:flow-matching
相关concept:timestep-conditioningconcept:adaptive-normalization
相关concept:timestep-conditioningconcept:adarms
相关concept:value-functionconcept:advantage
相关concept:value-functionconcept:offline-rl
相关concept:value-functionconcept:policy-improvement
相关concept:pi0concept:pi05
相关concept:pi0concept:flow-matching
相关concept:pi0concept:timestep-conditioning
相关concept:pi05concept:pi0
相关concept:pi05concept:pi05-action-expert
相关concept:pi05concept:flow-matching
相关concept:pi05concept:timestep-conditioning
相关concept:pi05concept:adarms
相关concept:pi05-action-expertconcept:pi05
相关concept:pi05-action-expertconcept:adarms
相关concept:pi05-action-expertconcept:timestep-conditioning
相关concept:pi05-action-expertconcept:residual-stream
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:behavior-cloning
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:dagger
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:distribution-shift
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:value-function
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:advantage
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:offline-rl
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:ood-action
引用blog:from-behavior-cloning-to-dagger-to-offline-rlconcept:policy-improvement
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:diffusion-model
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:brownian-motion
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:stochastic-differential-equation
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:flow-matching
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:pi05
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:pi05-action-expert
引用blog:diffusion-sde-to-pi05-flow-matching-action-headconcept:conditional-action-distribution
引用blog:flow-matching-data-collection-guideconcept:flow-matching
引用blog:flow-matching-data-collection-guideconcept:action-rate-multimodality
引用blog:flow-matching-data-collection-guideconcept:distribution-shift
引用blog:from-attention-mask-to-information-flowconcept:attention-mask
引用blog:from-attention-mask-to-information-flowconcept:block-causal-attention
引用blog:from-attention-mask-to-information-flowconcept:padding-mask
引用blog:from-attention-mask-to-information-flowconcept:prefix-lm
引用blog:from-adarms-to-transformer-normalizationconcept:pi05
引用blog:from-adarms-to-transformer-normalizationconcept:adarms
引用blog:from-adarms-to-transformer-normalizationconcept:rmsnorm
引用blog:from-adarms-to-transformer-normalizationconcept:layernorm
引用blog:from-adarms-to-transformer-normalizationconcept:pre-norm
引用blog:from-adarms-to-transformer-normalizationconcept:post-norm
引用blog:from-adarms-to-transformer-normalizationconcept:residual-stream
引用blog:from-adarms-to-transformer-normalizationconcept:timestep-conditioning
引用blog:from-adarms-to-transformer-normalizationconcept:flow-matching
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:diffusion-model
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:brownian-motion
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:stochastic-differential-equation
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:flow-matching
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:pi05
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:pi05-action-expert
引用thread:diffusion-sde-to-pi05-flow-matchingconcept:conditional-action-distribution
引用thread:understanding-adarms-in-pi05concept:pi05
引用thread:understanding-adarms-in-pi05concept:adarms
引用thread:understanding-adarms-in-pi05concept:rmsnorm
引用thread:understanding-adarms-in-pi05concept:layernorm
引用thread:understanding-adarms-in-pi05concept:pre-norm
引用thread:understanding-adarms-in-pi05concept:post-norm
引用thread:understanding-adarms-in-pi05concept:residual-stream
引用thread:understanding-adarms-in-pi05concept:timestep-conditioning
引用thread:understanding-adarms-in-pi05concept:flow-matching