ETS: Efficient Tree Search for Inference-Time Scaling

Test-time compute scaling has emerged as a new axis along which to improve model accuracy, where additional computation is used at inference time to allow the model to think longer for more challenging problems. One promising approach for test-time compute scaling is search against a process reward model, where a model generates multiple potential candidates at each step of the search, and these partial trajectories are then scored by a separate reward model in order to guide the search process. The diversity of trajectories in the tree search process affects the accuracy of the search, since increasing diversity promotes more exploration. However, this diversity comes at a cost, as divergent trajectories have less KV sharing, which means they consume more memory and slow down the search process. Previous search methods either do not perform sufficient exploration, or else explore diverse trajectories but have high latency. We address this challenge by proposing Efficient Tree Search (ETS), which promotes KV sharing by pruning redundant trajectories while maintaining necessary diverse trajectories. ETS incorporates a linear programming cost model to promote KV cache sharing by penalizing the number of nodes retained, while incorporating a semantic coverage term into the cost model to ensure that we retain trajectories which are semantically different. We demonstrate how ETS can achieve 1.8$\times$ reduction in average KV cache size during the search process, leading to 1.4$\times$ increased throughput relative to prior state-of-the-art methods, with minimal accuracy degradation and without requiring any custom kernel implementation. Code is available at: https://github.com/SqueezeAILab/ETS.

DiffuseLoco: Real-Time Legged Locomotion Control with Diffusion from Offline Datasets

This work introduces DiffuseLoco, a framework for training multi-skill diffusion-based policies for dynamic legged locomotion from offline datasets, enabling real-time control of diverse skills on robots in the real world. Offline learning at scale has led to breakthroughs in computer vision, natural language processing, and robotic manipulation domains. However, scaling up learning for legged robot locomotion, especially with multiple skills in a single policy, presents significant challenges for prior online reinforcement learning methods. To address this challenge, we propose a novel, scalable framework that leverages diffusion models to directly learn from offline multimodal datasets with a diverse set of locomotion skills. With design choices tailored for real-time control in dynamical systems, including receding horizon control and delayed inputs, DiffuseLoco is capable of reproducing multimodality in performing various locomotion skills, zero-shot transfer to real quadrupedal robots, and it can be deployed on edge computing devices. Furthermore, DiffuseLoco demonstrates free transitions between skills and robustness against environmental variations. Through extensive benchmarking in real-world experiments, DiffuseLoco exhibits better stability and velocity tracking performance compared to prior reinforcement learning and non-diffusion-based behavior cloning baselines. The design choices are validated via comprehensive ablation studies. This work opens new possibilities for scaling up learning-based legged locomotion controllers through the scaling of large, expressive models and diverse offline datasets.

SPEED: Speculative Pipelined Execution for Efficient Decoding

Generative Large Language Models (LLMs) based on the Transformer architecture have recently emerged as a dominant foundation model for a wide range of Natural Language Processing tasks. Nevertheless, their application in real-time scenarios has been highly restricted due to the significant inference latency associated with these models. This is particularly pronounced due to the autoregressive nature of generative LLM inference, where tokens are generated sequentially since each token depends on all previous output tokens. It is therefore challenging to achieve any token-level parallelism, making inference extremely memory-bound. In this work, we propose SPEED, which improves inference efficiency by speculatively executing multiple future tokens in parallel with the current token using predicted values based on early-layer hidden states. For Transformer decoders that employ parameter sharing, the memory operations for the tokens executing in parallel can be amortized, which allows us to accelerate generative LLM inference. We demonstrate the efficiency of our method in terms of latency reduction relative to model accuracy and demonstrate how speculation allows for training deeper decoders with parameter sharing with minimal runtime overhead.