arXiv subjects

Jeongtae Lee

Publications and source records attributed to Jeongtae Lee.

5 recordsLinked to original sources

arXiv2026

Carryover Drafting: Recycling Rejected States for Speculative Decoding

Speculative decoding accelerates LLM inference by verifying multiple drafted tokens in parallel, allowing a single target forward pass to accept several tokens. By construction, verification computes representations for both accepted and rejected tokens. Yet, conventional drafters retain only the representations of accepted tokens, leaving the substantial verifier computation spent on rejected tokens effectively wasted. We find that these discarded hidden states generated during target forward retain useful information about future tokens that can improve subsequent drafts. However, realizing this opportunity poses two distinct challenges. At inference, recycling overhead can increase drafting latency, diminishing the speedup gained from increased acceptance length. During training, standard parallel drafter training does not produce inference-aligned rejected states, while obtaining them through sequential rollouts would sacrifice parallelism across training positions. We introduce Carryover Drafting, which addresses both challenges. Carryover recycles rejected target hidden states as temporary KV context, allowing the drafter to selectively attend to them. It reuses the drafter's existing interface and adds only a single learned embedding to distinguish rejected states from committed context. The additional KV context is replaced each drafting round, keeping its length bounded by one proposal block. We introduce parallel draft--verify--draft training that exposes the drafter to inference-aligned rejected states while preserving parallelism across training positions. Experiments with DFlash and a DSpark-derived semi-autoregressive drafter across two target models show that this simple Carryover mechanism improves average acceptance length by 6.5--14.7% and end-to-end vLLM speedup by 7.9--14.4% over the corresponding baselines, with speedup gains reaching 28.8% on translation.

cs.LG↗

arXiv2026

ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion

In this paper, we propose Adversarial Dynamics Priors (ADP) for perturbation-resilient humanoid locomotion control. Existing motion prior-based methods induce natural motion styles by imitating kinematic motion features, but they do not directly regularize dynamics features, such as CoM motion, centroidal momentum, contact forces, and contact states. To address this limitation, we replace kinematic motion-style feature with selected dynamics features extracted from locomotion trajectories as the target of adversarial regularization. To this end, we use trajectory optimization to construct a reference dataset and train a discriminator to evaluate whether policy-induced temporal windows are consistent with the resulting reference distribution. Without explicit motion tracking, ADP encourages policy rollouts to remain close to the reference support, even after perturbations. Experimental results show that, compared with AMP, the strongest baseline in our evaluation, ADP improves the $80\%$-success impulse threshold ($J_{80}$) by $16.7\%$, while reducing direction-averaged recovery time and velocity tracking error by $47.9\%$ and $35.4\%$, respectively.

cs.RO↗

arXiv2026

PrefillShare: A Shared Prefill Module for KV Reuse in Multi-LLM Disaggregated Serving

Multi-agent systems increasingly orchestrate multiple specialized language models to solve complex real-world problems, often invoking them over a shared context. This execution pattern repeatedly processes the same prompt prefix across models. Consequently, each model redundantly executes the prefill stage and maintains its own key-value (KV) cache, increasing aggregate prefill load and worsening tail latency by intensifying prefill-decode interference in existing LLM serving stacks. Disaggregated serving reduces such interference by placing prefill and decode on separate GPUs, but disaggregation does not fundamentally eliminate inter-model redundancy in computation and KV storage for the same prompt. To address this issue, we propose PrefillShare, a novel algorithm that enables sharing the prefill stage across multiple models in a disaggregated setting. PrefillShare factorizes the model into prefill and decode modules, freezes the prefill module, and fine-tunes only the decode module. This design allows multiple task-specific models to share a prefill module and the KV cache generated for the same prompt. We further introduce a routing mechanism that enables effective prefill sharing across heterogeneous models in a vLLM-based disaggregated system. PrefillShare not only matches full fine-tuning accuracy on a broad range of tasks and models, but also delivers 4.5x lower p95 latency and 3.9x higher throughput in multi-model agent workloads.

cs.LG↗

arXiv2026

Training-free Dropout Sampling for Semantic Token Acceptance in Speculative Decoding

Speculative decoding accelerates large language model inference by proposing tokens with a lightweight draft model and selectively accepting them using a target model. This work introduces DropMatch, a novel approach that matches draft tokens to the predictive distribution of the target model via Monte Carlo dropout applied exclusively to the LM head, enabling sampling-based acceptance decisions. By generating multiple decoding paths, our method forms an empirical token distribution against which draft tokens are evaluated for consistency. This acceptance mechanism enables the model to adaptively control the size of decoding paths under an appropriate dropout probability, preventing substantial distortion of the target model predictive distribution. The proposed method operates in a training-free, data-free, and calibration-free manner, requires no architectural modification to pretrained models, and can be orthogonally integrated with a wide range of existing speculative decoding and inference acceleration techniques. Experiments across multiple benchmarks demonstrate that our approach increases acceptance length while maintaining competitive task performance, yielding inference speedups ranging from 1.09x to 1.33x over the standard baseline, and up to an additional 1.09x speedup when applied on top of EAGLE3.

cs.CL↗

arXiv2018

Lifelong Learning with Dynamically Expandable Networks

We propose a novel deep network architecture for lifelong learning which we refer to as Dynamically Expandable Network (DEN), that can dynamically decide its network capacity as it trains on a sequence of tasks, to learn a compact overlapping knowledge sharing structure among tasks. DEN is efficiently trained in an online manner by performing selective retraining, dynamically expands network capacity upon arrival of each task with only the necessary number of units, and effectively prevents semantic drift by splitting/duplicating units and timestamping them. We validate DEN on multiple public datasets under lifelong learning scenarios, on which it not only significantly outperforms existing lifelong learning methods for deep networks, but also achieves the same level of performance as the batch counterparts with substantially fewer number of parameters. Further, the obtained network fine-tuned on all tasks obtained significantly better performance over the batch models, which shows that it can be used to estimate the optimal network structure even when all tasks are available in the first place.

cs.LG↗