Per-position AR prediction entropy jointly drives speculative AR decoding and an anti-symmetric single-step drift decoder, yielding 3.8–5.5× faster hybrid visual generation without distillation.
Laptop-diff: Layer pruning and normalized dis- tillation for compressing diffusion models
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
representative citing papers
LIFT decomposes distillation into coarse linear alignment then fine refinement while PLACE adds error-based local adaptation, allowing stable training of 1.3M-parameter students (1.6% teacher size) to FID 15.73 across diffusion and flow models.
2ndMatch finetunes pruned diffusion models via second-order Jacobian matching inspired by Finite-Time Lyapunov Exponents to reduce the quality gap with dense models on image generation tasks.
WeiT applies Kronecker-based constraints during pre-training to disentangle size-agnostic knowledge into reusable weight templates and size-specific lightweight scalers for efficient initialization of models with varying depths and widths.
CR-Diff applies block-wise pruning followed by output amplification to diffusion models, improving consistency and fidelity at unseen resolutions while retaining default-resolution performance.
citing papers explorer
-
Drift-AR: Single-Step Visual Autoregressive Generation via Anti-Symmetric Drifting
Per-position AR prediction entropy jointly drives speculative AR decoding and an anti-symmetric single-step drift decoder, yielding 3.8–5.5× faster hybrid visual generation without distillation.
-
LIFT and PLACE: A Simple, Stable, and Effective Knowledge Distillation Framework for Lightweight Diffusion Models
LIFT decomposes distillation into coarse linear alignment then fine refinement while PLACE adds error-based local adaptation, allowing stable training of 1.3M-parameter students (1.6% teacher size) to FID 15.73 across diffusion and flow models.
-
2ndMatch: Finetuning Pruned Diffusion Models via Second-Order Jacobian Matching
2ndMatch finetunes pruned diffusion models via second-order Jacobian matching inspired by Finite-Time Lyapunov Exponents to reduce the quality gap with dense models on image generation tasks.
-
Constraint-based Pre-training: From Structured Constraints to Scalable Model Initialization
WeiT applies Kronecker-based constraints during pre-training to disentangle size-agnostic knowledge into reusable weight templates and size-specific lightweight scalers for efficient initialization of models with varying depths and widths.
-
Cross-Resolution Diffusion Models via Network Pruning
CR-Diff applies block-wise pruning followed by output amplification to diffusion models, improving consistency and fidelity at unseen resolutions while retaining default-resolution performance.