One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation

Diffusion models, praised for their success in generative tasks, are increasingly being applied to robotics, demonstrating exceptional performance in behavior cloning. However, their slow generation process stemming from iterative denoising steps poses a challenge for real-time applications in resource-constrained robotics setups and dynamically changing environments. In this paper, we introduce the One-Step Diffusion Policy (OneDP), a novel approach that distills knowledge from pre-trained diffusion policies into a single-step action generator, significantly accelerating response times for robotic control tasks. We ensure the distilled generator closely aligns with the original policy distribution by minimizing the Kullback-Leibler (KL) divergence along the diffusion chain, requiring only $2\%$-$10\%$ additional pre-training cost for convergence. We evaluated OneDP on 6 challenging simulation tasks as well as 4 self-designed real-world tasks using the Franka robot. The results demonstrate that OneDP not only achieves state-of-the-art success rates but also delivers an order-of-magnitude improvement in inference speed, boosting action prediction frequency from 1.5 Hz to 62 Hz, establishing its potential for dynamic and computationally constrained robotic applications. We share the project page at https://research.nvidia.com/labs/dir/onedp/.

Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step

Score identity Distillation (SiD) is a data-free method that has achieved state-of-the-art performance in image generation by leveraging only a pretrained diffusion model, without requiring any training data. However, the ultimate performance of SiD is constrained by the accuracy with which the pretrained model captures the true data scores at different stages of the diffusion process. In this paper, we introduce SiDA (SiD with Adversarial Loss), which not only enhances generation quality but also improves distillation efficiency by incorporating real images and adversarial loss. SiDA utilizes the encoder from the generator's score network as a discriminator, boosting its ability to distinguish between real images and those generated by SiD. The adversarial loss is batch-normalized within each GPU and then combined with the original SiD loss. This integration effectively incorporates the average "fakeness" per GPU batch into the pixel-based SiD loss, enabling SiDA to distill a single-step generator either from scratch or by fine-tuning an existing one. SiDA converges significantly faster than its predecessor when trained from scratch, and swiftly improves upon the original model's performance after an initial warmup period during fine-tuning from a pre-distilled SiD generator. This one-step adversarial distillation method has set new benchmarks for generation performance when distilling EDM diffusion models pretrained on CIFAR-10 (32x32) and ImageNet (64x64), achieving FID scores of $\mathbf{1.499}$ on CIFAR-10 unconditional, $\mathbf{1.396}$ on CIFAR-10 conditional, and $\mathbf{1.110}$ on ImageNet 64x64. Our open-source code will be integrated into the SiD codebase on GitHub.

Diffusion-RPO: Aligning Diffusion Models through Relative Preference Optimization

Aligning large language models with human preferences has emerged as a critical focus in language modeling research. Yet, integrating preference learning into Text-to-Image (T2I) generative models is still relatively uncharted territory. The Diffusion-DPO technique made initial strides by employing pairwise preference learning in diffusion models tailored for specific text prompts. We introduce Diffusion-RPO, a new method designed to align diffusion-based T2I models with human preferences more effectively. This approach leverages both prompt-image pairs with identical prompts and those with semantically related content across various modalities. Furthermore, we have developed a new evaluation metric, style alignment, aimed at overcoming the challenges of high costs, low reproducibility, and limited interpretability prevalent in current evaluations of human preference alignment. Our findings demonstrate that Diffusion-RPO outperforms established methods such as Supervised Fine-Tuning and Diffusion-DPO in tuning Stable Diffusion versions 1.5 and XL-1.0, achieving superior results in both automated evaluations of human preferences and style alignment. Our code is available at https://github.com/yigu1008/Diffusion-RPO

Long and Short Guidance in Score identity Distillation for One-Step Text-to-Image Generation

Diffusion-based text-to-image generation models trained on extensive text-image pairs have shown the capacity to generate photorealistic images consistent with textual descriptions. However, a significant limitation of these models is their slow sample generation, which requires iterative refinement through the same network. In this paper, we enhance Score identity Distillation (SiD) by developing long and short classifier-free guidance (LSG) to efficiently distill pretrained Stable Diffusion models without using real training data. SiD aims to optimize a model-based explicit score matching loss, utilizing a score-identity-based approximation alongside the proposed LSG for practical computation. By training exclusively with fake images synthesized with its one-step generator, SiD equipped with LSG rapidly improves FID and CLIP scores, achieving state-of-the-art FID performance while maintaining a competitive CLIP score. Specifically, its data-free distillation of Stable Diffusion 1.5 achieves a record low FID of 8.15 on the COCO-2014 validation set, with a CLIP score of 0.304 at an LSG scale of 1.5, and a FID of 9.56 with a CLIP score of 0.313 at an LSG scale of 2. We will make our PyTorch implementation and distilled Stable Diffusion one-step generators available at https://github.com/mingyuanzhou/SiD-LSG

Diffusion Policies creating a Trust Region for Offline Reinforcement Learning

Offline reinforcement learning (RL) leverages pre-collected datasets to train optimal policies. Diffusion Q-Learning (DQL), introducing diffusion models as a powerful and expressive policy class, significantly boosts the performance of offline RL. However, its reliance on iterative denoising sampling to generate actions slows down both training and inference. While several recent attempts have tried to accelerate diffusion-QL, the improvement in training and/or inference speed often results in degraded performance. In this paper, we introduce a dual policy approach, Diffusion Trusted Q-Learning (DTQL), which comprises a diffusion policy for pure behavior cloning and a practical one-step policy. We bridge the two polices by a newly introduced diffusion trust region loss. The diffusion policy maintains expressiveness, while the trust region loss directs the one-step policy to explore freely and seek modes within the region defined by the diffusion policy. DTQL eliminates the need for iterative denoising sampling during both training and inference, making it remarkably computationally efficient. We evaluate its effectiveness and algorithmic characteristics against popular Kullback-Leibler (KL) based distillation methods in 2D bandit scenarios and gym tasks. We then show that DTQL could not only outperform other methods on the majority of the D4RL benchmark tasks but also demonstrate efficiency in training and inference speeds. The PyTorch implementation is available at https://github.com/TianyuCodings/Diffusion_Trusted_Q_Learning.

Self-Augmented Preference Optimization: Off-Policy Paradigms for Language Model Alignment

Traditional language model alignment methods, such as Direct Preference Optimization (DPO), are limited by their dependence on static, pre-collected paired preference data, which hampers their adaptability and practical applicability. To overcome this limitation, we introduce Self-Augmented Preference Optimization (SAPO), an effective and scalable training paradigm that does not require existing paired data. Building on the self-play concept, which autonomously generates negative responses, we further incorporate an off-policy learning pipeline to enhance data exploration and exploitation. Specifically, we employ an Exponential Moving Average (EMA) model in conjunction with a replay buffer to enable dynamic updates of response segments, effectively integrating real-time feedback with insights from historical data. Our comprehensive evaluations of the LLaMA3-8B and Mistral-7B models across benchmarks, including the Open LLM Leaderboard, IFEval, AlpacaEval 2.0, and MT-Bench, demonstrate that SAPO matches or surpasses established offline contrastive baselines, such as DPO and Odds Ratio Preference Optimization, and outperforms offline self-play methods like SPIN. Our code is available at https://github.com/yinyueqin/SAPO

Score identity Distillation: Exponentially Fast Distillation of Pretrained Diffusion Models for One-Step Generation

We introduce Score identity Distillation (SiD), an innovative data-free method that distills the generative capabilities of pretrained diffusion models into a single-step generator. SiD not only facilitates an exponentially fast reduction in Fr\'echet inception distance (FID) during distillation but also approaches or even exceeds the FID performance of the original teacher diffusion models. By reformulating forward diffusion processes as semi-implicit distributions, we leverage three score-related identities to create an innovative loss mechanism. This mechanism achieves rapid FID reduction by training the generator using its own synthesized images, eliminating the need for real data or reverse-diffusion-based generation, all accomplished within significantly shortened generation time. Upon evaluation across four benchmark datasets, the SiD algorithm demonstrates high iteration efficiency during distillation and surpasses competing distillation approaches, whether they are one-step or few-step, data-free, or dependent on training data, in terms of generation quality. This achievement not only redefines the benchmarks for efficiency and effectiveness in diffusion distillation but also in the broader field of diffusion-based generation. Our PyTorch implementation will be publicly accessible on GitHub.

Take the Bull by the Horns: Hard Sample-Reweighted Continual Training Improves LLM Generalization

In the rapidly advancing arena of large language models (LLMs), a key challenge is to enhance their capabilities amid a looming shortage of high-quality training data. Our study starts from an empirical strategy for the light continual training of LLMs using their original pre-training data sets, with a specific focus on selective retention of samples that incur moderately high losses. These samples are deemed informative and beneficial for model refinement, contrasting with the highest-loss samples, which would be discarded due to their correlation with data noise and complexity. We then formalize this strategy into a principled framework of Instance-Reweighted Distributionally Robust Optimization (IR-DRO). IR-DRO is designed to dynamically prioritize the training focus on informative samples through an instance reweighting mechanism, streamlined by a closed-form solution for straightforward integration into established training protocols. Through rigorous experimentation with various models and datasets, our findings indicate that our sample-targeted methods significantly improve LLM performance across multiple benchmarks, in both continual pre-training and instruction tuning scenarios. Our codes are available at https://github.com/VITA-Group/HardFocusTraining.

Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Prompts

In the field of large language models (LLMs), aligning models with the diverse preferences of users is a critical challenge. Direct Preference Optimization (DPO) has played a key role in this area. It works by using pairs of preferences derived from the same prompts, and it functions without needing an additional reward model. However, DPO does not fully reflect the complex nature of human learning, which often involves understanding contrasting responses to not only identical but also similar questions. To overcome this shortfall, we propose Relative Preference Optimization (RPO). RPO is designed to discern between more and less preferred responses derived from both identical and related prompts. It introduces a contrastive weighting mechanism, enabling the tuning of LLMs using a broader range of preference data, including both paired and unpaired sets. This approach expands the learning capabilities of the model, allowing it to leverage insights from a more varied set of prompts. Through empirical tests, including dialogue and summarization tasks, and evaluations using the AlpacaEval2.0 leaderboard, RPO has demonstrated a superior ability to align LLMs with user preferences and to improve their adaptability during the training process. The PyTorch code necessary to reproduce the results presented in the paper will be made available on GitHub for public access.

Anything in Any Scene: Photorealistic Video Object Insertion

Realistic video simulation has shown significant potential across diverse applications, from virtual reality to film production. This is particularly true for scenarios where capturing videos in real-world settings is either impractical or expensive. Existing approaches in video simulation often fail to accurately model the lighting environment, represent the object geometry, or achieve high levels of photorealism. In this paper, we propose Anything in Any Scene, a novel and generic framework for realistic video simulation that seamlessly inserts any object into an existing dynamic video with a strong emphasis on physical realism. Our proposed general framework encompasses three key processes: 1) integrating a realistic object into a given scene video with proper placement to ensure geometric realism; 2) estimating the sky and environmental lighting distribution and simulating realistic shadows to enhance the light realism; 3) employing a style transfer network that refines the final video output to maximize photorealism. We experimentally demonstrate that Anything in Any Scene framework produces simulated videos of great geometric realism, lighting realism, and photorealism. By significantly mitigating the challenges associated with video data generation, our framework offers an efficient and cost-effective solution for acquiring high-quality videos. Furthermore, its applications extend well beyond video data augmentation, showing promising potential in virtual reality, video editing, and various other video-centric applications. Please check our project website https://anythinginanyscene.github.io for access to our project code and more high-resolution video results.