Recent advancements in diffusion models are focused on enhancing sampling efficiency and practical deployment across various applications. Techniques such as Dynamic Gradient Weighting and Just-in-Time spatial acceleration are significantly reducing the computational burden associated with iterative sampling, achieving speedups of up to 7x while maintaining high fidelity in generated outputs. Innovations like the Longest Stable Prefix scheduler for diffusion language models are addressing inefficiencies in token processing, yielding up to 3.4x faster inference rates. Additionally, the introduction of frameworks like dLLM is streamlining the development and deployment of diffusion language models, making them more accessible for researchers and practitioners. These developments are crucial for commercial applications in areas like image synthesis, natural language processing, and style transfer, where rapid, high-quality generation is essential. As the field matures, the focus is shifting toward optimizing resource allocation and enhancing model adaptability, paving the way for broader integration into real-world systems.
Diffusion Models (DMs) have achieved state-of-the-art generative performance across multiple modalities, yet their sampling process remains prohibitively slow due to the need for hundreds of function ...
Diffusion transformers (DiTs) achieve high generative quality but lock FLOPs to image resolution, limiting principled latency-quality trade-offs, and allocate computation uniformly across input spatia...
Discrete diffusion models offer global context awareness and flexible parallel generation. However, uniform random noise schedulers in standard DLLM training overlook the highly non-uniform informatio...
Diffusion Transformers have established a new state-of-the-art in image synthesis, but the high computational cost of iterative sampling severely hampers their practical deployment. While existing acc...
Although diffusion language models (DLMs) are evolving quickly, many recent models converge on a set of shared components. These components, however, are distributed across ad-hoc research codebases o...
Diffusion Language Models (DLMs) promise highly parallel text generation, yet their practical inference speed is often bottlenecked by suboptimal decoding schedulers. Standard approaches rely on 'scat...
Diffusion models achieve state-of-the-art generative performance but are fundamentally bottlenecked by their slow iterative sampling process. While diffusion distillation techniques enable high-fideli...
Guided diffusion sampling relies on approximating often intractable likelihood scores, which introduces significant noise into the sampling dynamics. We propose using adaptive moment estimation to sta...
Recent research has shown that text-to-image diffusion models are capable of generating high-quality images guided by text prompts. But can they be used to generate or approximate real-world images fr...
We introduce the Multilevel Euler-Maruyama (ML-EM) method compute solutions of SDEs and ODEs using a range of approximators $f^1,\dots,f^k$ to the drift $f$ with increasing accuracy and computational ...