Rendering document text as images allows vision-language models to encode documents as visual tokens, which can reduce input sequence length compared with text input. This reduction in input length is particularly useful for reranking, where each query involves scoring multiple candidate documents and token savings app...
Seongtae Hong, Youngjoon Jang, Jungseob Lee et al.· 0 citations
Small reasoning models (SRMs) have shown strong performance on complex reasoning tasks by generating extended chain-of-thought trajectories, but they often fail to recover once their reasoning enters an erroneous path. Existing test-time fusion methods rely on local fusion signals to determine when to trigger fusion, w...
Myunghoon Kang, Jungseob Lee, Jaehyung Seo et al.· 0 citations
Speculative decoding accelerates large language model inference by drafting future tokens cheaply and verifying them with the target model in parallel. Block drafters score a whole block of future tokens in one forward pass, yet standard decoding verifies only the top-scoring chain and discards the other candidates. Be...
Fine-tuning adapts aligned large language models (LLMs) to downstream tasks, but a few dozen harmful examples can remove their refusal of harmful requests. Prior work localizes safety-related behavior to specific layers, directions, and tokens, suggesting targets for protection. We test whether successful localization...
Jungseob Lee, D. Lee, Sugyeong Eo et al.· 1 citation
Knowledge distillation aims to transfer the factual knowledge of large language models to smaller models for efficient deployment. Yet a teacher may recall a relation in one direction while failing to generate the answer in the reverse direction. Distillation from its generated answers can therefore propagate this dire...
Jungseob Lee, Sugyeong Eo, Seongtae Hong et al.· 0 citations
Block drafters for speculative decoding are commonly trained on corpora written by external models, where a single off-policy token invalidates supervision for all subsequent slots in a block. Existing approaches discard these divergent slots, resulting in severe supervision loss. To resolve this problem while preservi...
Jungseob Lee, Chanjun Park, Sugyeong Eo et al.· 1 citation
PriceCheck is introduced, which builds a compact family of decision rules from label-free checks such as re-solving a problem, and shows that choosing how checks are combined and stopped matters alongside how well a verifier ranks answers.
D. Lee, Jungseob Lee, Chanjun Park et al.· 2 citations
Redline is a finite-sample procedure that selects operating points, hand-picked or learned, from the correctness of their answers on calibration prompts, and keeps the reference-relative risk, the joint probability that the reference answers correctly and a candidate configuration does not, within a user-chosen budget...
Jungseob Lee, D. Lee, Chanjun Park et al.· 1 citation
A byte crossover is derived, the context length at which the two savings are equal, together with ideal speedup bounds for each branch and for their composition, from model dimensions and keep ratios alone and is derived at https://github.com/js-lee-AI/ByteCross.
Jungseob Lee, Seungyoon Lee, Seongtae Hong et al.· 1 citation
GLANCE is presented, the first one-pass block drafter that is lossless on an unmodified VLM target, and it breaks the cycle at both ends of a self-defeating cycle.
Jungseob Lee, Seongtae Hong, D. Lee et al.· 0 citations
DART is introduced, a training-free routing framework that samples two cheap no-think drafts, accepts direct answering when the drafts agree, and predicts a thinking budget from draft entropy when they disagree, and preserves or improves always-thinking accuracy in most settings while reducing thinking-token use.
Jungseob Lee, Seongtae Hong, Seungjun Lee et al.· arXiv.org· 1 citation
Because the signal spans a contiguous layer band, LayerMix aggregates it to match oracle-layer performance without oracle access, and characterize the geometry within the controlled paired-example paradigm.
Jungseob Lee, Jaehyung Seo, Heu-Jeoung Lim· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
