Samsung (South Korea)

3 related articles

Subscribe to updates

arXiv Oct 15, 2026
Byte order, not just byte count, sets boundary prediction: UBE cuts one-symbol balanced grapheme error from 8.76% to 0.06%
Byte-level language models group bytes into patches with a learned boundary predictor, and what that predictor can see at limited lookahead depends on the text encoding. In UTF-8 many scripts take three bytes per codepoint, and the first byte carries only 4 high codepoint bits, so after the same left context a mark extending a grapheme cluster and a letter starting the next can share their first byte. Training small causal predictors on matched FineWeb-2 text and evaluating on 17 FLORES-200 languages, balanced grapheme-boundary error at one-symbol lookahead is 8.76% with UTF-8 and 0.06% with UBE; the same-length UBE-BE variant gives 8.62%, so the gain follows byte order rather than stream length.
Byte-level language models group bytes into patches with a learned boundary predictor, and what that predictor can see at limited lookahead depends on the text encoding. In UTF-8 many scripts take three bytes per codepoint, and the first byte carries only 4 high codepoint bits, so after the same left context a mark extending a grapheme cluster and a letter starting the next can share their first byte. Training small causal predictors on matched FineWeb-2 text and evaluating on 17 FLORES-200 languages, balanced grapheme-boundary error at one-symbol lookahead is 8.76% with UTF-8 and 0.06% with UBE; the same-length UBE-BE variant gives 8.62%, so the gain follows byte order rather than stream length.
Samsung (South Korea)
Hyunsik Kim（First author, Corresponding author）
Universal Byte-Level Encoding Multilingual boundary prediction byteification Grapheme clusters
arXiv Oct 5, 2026
ServeTwin couples specification-driven analytical timing with a stateful serving loop, reproducing the throughput-interactivity frontier at 3.6% mean error without target-hardware profiling
The authors present ServeTwin, a closed-loop simulator that couples specification-driven analytical timing with a stateful serving loop, uses iSTAGE to derive per-iteration traces from model and engine specifications while avoiding target-hardware operator profiling, and implements a vLLM-compatible interface that runs unmodified serving benchmarks; against real deployments it reproduces InferenceX's steady-state throughput-interactivity frontier with a 3.6% mean error and predicts LMBenchmark's multi-turn performance with a 9.9% error while tracking KV-cache evolution, and pre-silicon sweeps show that the preferred HBM bandwidth-capacity tradeoff reverses across workload states and that increasing concurrency shifts the bottleneck from memory bandwidth to scheduler and runtime overheads.
The authors present ServeTwin, a closed-loop simulator that couples specification-driven analytical timing with a stateful serving loop, uses iSTAGE to derive per-iteration traces from model and engine specifications while avoiding target-hardware operator profiling, and implements a vLLM-compatible interface that runs unmodified serving benchmarks; against real deployments it reproduces InferenceX's steady-state throughput-interactivity frontier with a 3.6% mean error and predicts LMBenchmark's multi-turn performance with a 9.9% error while tracking KV-cache evolution, and pre-silicon sweeps show that the preferred HBM bandwidth-capacity tradeoff reverses across workload states and that increasing concurrency shifts the bottleneck from memory bandwidth to scheduler and runtime overheads.
Samsung Advanced Institute of Technology (South Korea)Samsung (South Korea)
Sungjoon Park（First author）
ServeTwin iSTAGE distributed LLM serving simulation
arXiv Sep 29, 2026
SCOPD lifts VLM performance at 10% visual-token retention from 86.37% to 92.43% via sparse-context on-policy self-distillation
Using a fixed-context Pass@K analysis, this work shows that degradation after visual-token pruning is not only lost information but also unreliable use of surviving evidence (a representation-utilization gap), and introduces SCOPD and SCOPD+ sparse-context on-policy self-distillation, raising the normalized average over 13 benchmarks at 10% visual-token retention from 86.37% (Vanilla) to 90.49% and 92.43% without architectural changes, extra inference-time computation, or ground-truth answers.
Using a fixed-context Pass@K analysis, this work shows that degradation after visual-token pruning is not only lost information but also unreliable use of surviving evidence (a representation-utilization gap), and introduces SCOPD and SCOPD+ sparse-context on-policy self-distillation, raising the normalized average over 13 benchmarks at 10% visual-token retention from 86.37% (Vanilla) to 90.49% and 92.43% without architectural changes, extra inference-time computation, or ground-truth answers.
Stanford University University of Waterloo Tsinghua University University of Toronto University of British Columbia Samsung (South Korea)York University
Ahmadreza Jeddi（First author）
On-Policy Self-Distillation Visual Token Pruning Vision-Language Model Selective Supervision Multimodal Reasoning

You’re all caught up

Byte order, not just byte count, sets boundary prediction: UBE cuts one-symbol balanced grapheme error from 8.76% to 0.06%

ServeTwin couples specification-driven analytical timing with a stateful serving loop, reproducing the throughput-interactivity frontier at 3.6% mean error without target-hardware profiling

SCOPD lifts VLM performance at 10% visual-token retention from 86.37% to 92.43% via sparse-context on-policy self-distillation