VLA models exhibit a compute-bound VLM phase followed by a memory-bound action phase on edge hardware; DP-Cache and V-AEFusion reduce redundancy and enable pipeline parallelism for up to 6x speedup on NPUs with marginal task degradation.
hub
Timestep embedding tells: It’s time to cache for video diffusion model
22 Pith papers cite this work, alongside 2 external citations. Polarity classification is still indexing.
22
Pith papers citing it
2
external citations · Pith
abstract
As a fundamental backbone for video generation, diffusion models are challenged by low inference speed due to the sequential nature of denoising. Previous methods speed up the models by caching and reusing model outputs at uniformly selected timesteps. However, such a strategy neglects the fact that differences among model outputs are not uniform across timesteps, which hinders selecting the appropriate model outputs to cache, leading to a poor balance between inference efficiency and visual quality. In this study, we introduce Timestep Embedding Aware Cache (TeaCache), a training-free caching approach that estimates and leverages the fluctuating differences among model outputs across timesteps. Rather than directly using the time-consuming model outputs, TeaCache focuses on model inputs, which have a strong correlation with the modeloutputs while incurring negligible computational cost. TeaCache first modulates the noisy inputs using the timestep embeddings to ensure their differences better approximating those of model outputs. TeaCache then introduces a rescaling strategy to refine the estimated differences and utilizes them to indicate output caching. Experiments show that TeaCache achieves up to 4.41x acceleration over Open-Sora-Plan with negligible (-0.07% Vbench score) degradation of visual quality.
hub tools
citation-role summary
background 2
method 2
citation-polarity summary
representative citing papers
LayerCache enables per-layer-group caching in flow matching models via adaptive JVP span selection and greedy 3D scheduling, delivering 1.37x speedup with PSNR 37.46 dB, SSIM 0.9834, and LPIPS 0.0178 on Qwen-Image.
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
Dream-Tac unifies visual and tactile signals in a world action model using contact-gated fusion and attention bias, reporting 31.7% average action accuracy gains on six manipulation tasks.
Light Interaction accelerates interactive video world models up to 2.59x via adaptive context management, denoising cache acceleration, and 3D block sparse attention without retraining.
AtlasVid proposes a decoupled global-local diffusion framework that trains at low resolution with LoRA and generalizes to ultra-high-resolution long video synthesis via semantic proxy guidance and locality-preserving attention.
CoCoDiff achieves 3.6x average and 8.4x peak speedup for distributed DiT inference on up to 96 GPU tiles via tile-aware all-to-all, V-first scheduling, and selective V communication.
LIPAR prunes redundant inter-frame latent patches in video generation and recovers attention to deliver 1.53x speedup at 19.3 FPS with no quality drop or extra training.
DreamZero uses a 14B video diffusion model as a World Action Model to achieve over 2x better zero-shot generalization on real robots than state-of-the-art VLAs, real-time 7Hz closed-loop control, and cross-embodiment transfer with 10-30 minutes of data.
BAC accelerates transformer-based Diffusion Policy up to 3x by block-level adaptive feature caching using an Adaptive Caching Scheduler and Bubbling Union Algorithm to control error propagation.
SkyReels-V2 produces infinite-length film videos via MLLM-based captioning, progressive pretraining, motion RL, and diffusion forcing with non-decreasing noise schedules.
OTCache uses optimal transport to interpolate caching schedules between a graph-based reference and an Optuna-optimized anchor, delivering 3.66x-4.7x speedups on FLUX.1, Qwen-Image and HunyuanVideo with improved fidelity.
Sol Video Inference Engine uses parallel skill agents to optimize cache, sparse attention, token pruning, quantization, and kernel fusion, delivering over 2x end-to-end acceleration with near-lossless quality on three video models.
DVG dynamically selects content-aware spatio-temporal acceleration strategies for diffusion-based video generation, delivering up to 7x speedup with near-lossless quality on models like HunyuanVideo.
Focused Forcing is a training-free per-frame KV selection method that combines attention scores with diversity metrics and head-importance estimation to accelerate autoregressive video diffusion up to 1.48x while improving quality.
INT8 W8A8 post-training quantization of Ideogram 4.0 preserves FP8 quality on a 200-prompt benchmark while outperforming NF4 on CLIP score and offering a favorable quality-memory trade-off via GGUF Q4_K.
EchoTorrent combines multi-teacher distillation, adaptive CFG calibration, hybrid long-tail forcing, and VAE decoder refinement to enable few-pass autoregressive streaming video generation with improved temporal consistency and audio-lip sync.
A survey of large-model inference optimization, organized as a four-layer 'token-operations' taxonomy: multi-model fusion, model optimization, compute-model fusion, and compute-network-model fusion.
Coherence-first rendering with 15 FPS anchors plus FSR4 upsampling to 30 FPS preserves scene geometry and identity longer than native 30 FPS generation across tested forest, sword, desert, and snow scenes, with LPIPS favoring the coherence branch.
citing papers explorer
-
Characterizing Vision-Language-Action Models across XPUs: Constraints and Acceleration for On-Robot Deployment
VLA models exhibit a compute-bound VLM phase followed by a memory-bound action phase on edge hardware; DP-Cache and V-AEFusion reduce redundancy and enable pipeline parallelism for up to 6x speedup on NPUs with marginal task degradation.
-
LayerCache: Exploiting Layer-wise Velocity Heterogeneity for Efficient Flow Matching Inference
LayerCache enables per-layer-group caching in flow matching models via adaptive JVP span selection and greedy 3D scheduling, delivering 1.37x speedup with PSNR 37.46 dB, SSIM 0.9834, and LPIPS 0.0178 on Qwen-Image.
-
Dynamic-in-Few-Step: Unifying Dynamic Computation and Few-Step Distillation for Efficient Video Generation
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
-
Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation
Dream-Tac unifies visual and tactile signals in a world action model using contact-gated fusion and attention bias, reporting 31.7% average action accuracy gains on six manipulation tasks.
-
Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models
Light Interaction accelerates interactive video world models up to 2.59x via adaptive context management, denoising cache acceleration, and 3D block sparse attention without retraining.
-
AtlasVid: Efficient Ultra-High-Resolution Long Video Generation via Decoupled Global-Local Modeling
AtlasVid proposes a decoupled global-local diffusion framework that trains at low resolution with LoRA and generalizes to ultra-high-resolution long video synthesis via semantic proxy guidance and locality-preserving attention.
-
CoCoDiff: Optimizing Collective Communications for Distributed Diffusion Transformer Inference Under Ulysses Sequence Parallelism
CoCoDiff achieves 3.6x average and 8.4x peak speedup for distributed DiT inference on up to 96 GPU tiles via tile-aware all-to-all, V-first scheduling, and selective V communication.
-
Video Compression Meets Video Generation: Latent Inter-Frame Pruning with Attention Recovery
LIPAR prunes redundant inter-frame latent patches in video generation and recovers attention to deliver 1.53x speedup at 19.3 FPS with no quality drop or extra training.
-
World Action Models are Zero-shot Policies
DreamZero uses a 14B video diffusion model as a World Action Model to achieve over 2x better zero-shot generalization on real robots than state-of-the-art VLAs, real-time 7Hz closed-loop control, and cross-embodiment transfer with 10-30 minutes of data.
-
Block-wise Adaptive Caching for Accelerating Diffusion Policy
BAC accelerates transformer-based Diffusion Policy up to 3x by block-level adaptive feature caching using an Adaptive Caching Scheduler and Bubbling Union Algorithm to control error propagation.
-
SkyReels-V2: Infinite-length Film Generative Model
SkyReels-V2 produces infinite-length film videos via MLLM-based captioning, progressive pretraining, motion RL, and diffusion forcing with non-decreasing noise schedules.
-
OTCache: Optimal Transport for Geometry-Aware Caching in Diffusion Models
OTCache uses optimal transport to interpolate caching schedules between a graph-based reference and an Optuna-optimized anchor, delivering 3.66x-4.7x speedups on FLUX.1, Qwen-Image and HunyuanVideo with improved fidelity.
-
Sol Video Inference Engine: Agent-Native Full-Stack Acceleration Framework for Efficient Video Generation
Sol Video Inference Engine uses parallel skill agents to optimize cache, sparse attention, token pruning, quantization, and kernel fusion, delivering over 2x end-to-end acceleration with near-lossless quality on three video models.
-
Dynamic Video Generation: Shaping Video Generation Across Time and Space
DVG dynamically selects content-aware spatio-temporal acceleration strategies for diffusion-based video generation, delivering up to 7x speedup with near-lossless quality on models like HunyuanVideo.
-
Focused Forcing: Content-Aware Per-Frame KV Selection for Efficient Autoregressive Video Diffusion
Focused Forcing is a training-free per-frame KV selection method that combines attention scores with diversity metrics and head-importance estimation to accelerate autoregressive video diffusion up to 1.48x while improving quality.
-
Holding the FP8 Quality Ceiling at 8-Bit Weights and Activations: INT8 and GGUF Post-Training Quantization of Ideogram 4.0 for Consumer GPUs
INT8 W8A8 post-training quantization of Ideogram 4.0 preserves FP8 quality on a 200-prompt benchmark while outperforming NF4 on CLIP score and offering a favorable quality-memory trade-off via GGUF Q4_K.
-
EchoTorrent: Towards Swift, Sustained, and Streaming Multi-Modal Video Generation
EchoTorrent combines multi-teacher distillation, adaptive CFG calibration, hybrid long-tail forcing, and VAE decoder refinement to enable few-pass autoregressive streaming video generation with improved temporal consistency and audio-lip sync.
-
Token-Operations-Oriented Inference Optimization Techniques for Large Models
A survey of large-model inference optimization, organized as a four-layer 'token-operations' taxonomy: multi-model fusion, model optimization, compute-model fusion, and compute-network-model fusion.
-
Fewer, Better Frames: A Compute-Normalized Proof of Concept for Coherence-First World-Model Rendering with Model-Guided FSR4 Frame Generation
Coherence-first rendering with 15 FPS anchors plus FSR4 upsampling to 30 FPS preserves scene geometry and identity longer than native 30 FPS generation across tested forest, sword, desert, and snow scenes, with LPIPS favoring the coherence branch.
- RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion Models
- Spectral Progressive Diffusion for Efficient Image and Video Generation
- AdaCorrection: Adaptive Offset Cache Correction for Accurate Diffusion Transformers