Kangaroo spotted on highway in Virginia prompts rescue by Nelson County Sheriff's deputies Saturday morning. The injured female marsupial was tranquilized and returned to its owner.

Breakthroughs, discoveries, and DIY tips sent six days a week. Kangaroos have likely been hopping across the planet for much longer than experts previously believed. Not only that, but the ancestors of today's marsupials landed their leaps while growing larger than their descendents. For thousands of years, the planet's largest hopping animal has remained Australia's red kangaroo (). A male "Big Red" easily reaches over five feet tall, weighs 200 pounds, and travels around 37 mph at a pace of up to six feet per leap.

Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models (LLMs) while maintaining an identical sampling distribution. However, the conventional approach of training separate draft model to achieve a satisfactory token acceptance rate can be costly and impractical. In this paper, we propose a novel self-speculative decoding framework \emph{Kangaroo} with \emph{double} early exiting strategy, which leverages the shallow sub-network and the \texttt{LM Head} of the well-trained target LLM to construct a self-drafting model. Then, the self-verification stage only requires computing the remaining layers over the \emph{early-exited} hidden states in parallel. To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network.

Gemini, etc.; some of these breakthroughs even seem to enable AI models to outperform human abilities in varied tasks that demand higher-order cognitive skills.

To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network.

Gemini, etc.; some of these breakthroughs even seem to enable AI models to outperform human abilities in varied tasks that demand higher-order cognitive skills.

To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network.

Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models (LLMs) while maintaining an identical sampling distribution. However, the conventional approach of training separate draft model to achieve a satisfactory token acceptance rate can be costly and impractical. In this paper, we propose a novel self-speculative decoding framework \emph{Kangaroo} with \emph{double} early exiting strategy, which leverages the shallow sub-network and the \texttt{LM Head} of the well-trained target LLM to construct a self-drafting model. Then, the self-verification stage only requires computing the remaining layers over the \emph{early-exited} hidden states in parallel. To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network.

Large Language Models (LLMs) excel in reasoning tasks through Chain-of-Thought (CoT) prompting. However, CoT prompting greatly increases computational demands, which has prompted growing interest in distilling CoT capabilities into Small Language Models (SLMs). This study systematically examines the factors influencing CoT distillation, including the choice of granularity, format and teacher model. Through experiments involving four teacher models and seven student models across seven mathematical and commonsense reasoning datasets, we uncover three key findings: (1) Unlike LLMs, SLMs exhibit a non-monotonic relationship with granularity, with stronger models benefiting from finer-grained reasoning and weaker models performing better with simpler CoT supervision; (2) CoT format significantly impacts LLMs but has minimal effect on SLMs, likely due to their reliance on supervised fine-tuning rather than pretraining preferences; (3) Stronger teacher models do NOT always produce better student models, as diversity and complexity in CoT supervision can outweigh accuracy alone. These findings emphasize the need to tailor CoT strategies to specific student model, offering actionable insights for optimizing CoT distillation in SLMs. The code and datasets are available at https://github.com/EIT-NLP/Distilling-CoT-Reasoning.

In this paper, we focus on the kangaroo, which has powerful legs capable of jumping and a soft and strong tail. To incorporate these unique structure into a robot for utilization, we propose a design method that takes into account both the feasibility as a robot and the kangaroo-mimetic structure. Based on the kangaroo's musculoskeletal structure, we determine the structure of the robot that enables it to jump by analyzing the muscle arrangement and prior verification in simulation. Also, to realize a tail capable of body support, we use an articulated, elastic structure as a tail. In order to achieve both softness and high power output, the robot is driven by a direct-drive, high-power wire-winding mechanism, and weight of legs and the tail is reduced by placing motors in the torso. The developed kangaroo robot can jump with its hind legs, moving its tail, and supporting its body using its hind legs and tail.