arXiv:2608. 11361v1 Announce Type: new Abstract: Tokenizer vocabulary size is a foundational design choice in large language model (LLM) infrastructure, yet it is typically fixed at training time based on convention rather than deployment analysis.
arXiv:2511. 20849v2 Announce Type: replace-cross Abstract: We introduce a new tokenizer for language models that minimizes the average tokens per character, thereby reducing the number of tokens needed to represent text during training and to generate text during inference.
arXiv:2606. 00516v1 Announce Type: new Abstract: Mixed batching (MB)--interleaving prefill and decode in a single batch--has become the standard scheduling strategy for large language model (LLM) inference due to its efficiency in maximizing compute and memory utilization.
arXiv:2608. 10605v1 Announce Type: cross Abstract: In large-scale pretraining, the algorithm, architecture, and systems decisions are conventionally made in disconnected stages.
In large-scale pretraining, the algorithm, architecture, and systems decisions are conventionally made in disconnected stages. A scaling law stage selects an architecture and training recipe, optimizing loss under compute constraints, and a separate systems stage then optimizes the implementation for hardware efficiency.
arXiv:2606. 10706v1 Announce Type: cross Abstract: Resource constraints increasingly determine what can be trained, fine-tuned, and deployed in large language models (LLMs), yet efficiency is often studied through isolated techniques rather than as an interacting system of limits.
The paper argues that AI deployment performance depends on interactions among compression, compiler transformations, and serving policies rather than just model architecture. It introduces a three‑layer taxonomy—model‑level techniques, compiler transformations, and system policies—and frames deployment as a constrained multi‑objective optimization problem over accuracy, latency, throughput, memory footprint, and energy. The authors propose an evidence protocol for comparable benchmarking and synthesize data from edge and data‑center platforms to show that cross‑layer interactions drive deployment outcomes, concluding with a constraint‑aware selection procedure and open research problems.
HeadWiseKV is a training‑free framework that compresses the residual global key–value caches of hybrid long‑context language models by assigning each physical KV head a static, multilevel history window. It formulates cache allocation as a restricted operational rate–distortion problem and uses the SeqCalib algorithm to generate per‑head residency policies that account for interactions across layers. In evaluations on four hybrid models, HeadWiseKV preserves near‑full‑KV quality while reducing peak device memory usage by 8.59% at a 112K context length and extending the largest verified context from 114K to 161K.
arXiv:2609.37887v1 Announce Type: new Abstract: Activation and key-value cache precision change what a quantized language model computes without altering its stored weights. Direct weight-code bounds...
Long-context inference retains a growing key--value (KV) cache during decoding, which consumes substantial GPU memory and can reduce generation throughput. This bottleneck remains in hybrid language m...
The paper investigates whether training Mixture-of-Experts (MoE) routers can improve memory‑bandwidth locality on consumer GPUs. Using a new zero‑surgery telemetry tool, the authors measure that a large Qwen3‑235B model is bottlenecked by disk‑based expert access, and that an LRU cache can serve a majority of requests. They pre‑register experiments training 137 M‑parameter MoE models with locality‑aware losses, finding that while cache misses can drop up to 60 % (99 % static‑pin hit rate), every configuration fails to meet a strict 1 % perplexity threshold, indicating a tight coupling between cache efficiency and model quality.
arXiv:2607. 25271v1 Announce Type: cross Abstract: Classical compute-optimal scaling laws assume an unbounded supply of fresh pretraining data, yet pretraining is increasingly entering a regime in which compute grows faster than the availability of high-quality data.
GroupKV is a lightweight hierarchical KV cache management system designed for long‑context diffusion large language model (dLLM) inference. It partitions the context into contiguous groups and uses coarse‑to‑fine sparse selection, cross‑layer consistency for predictive prefetching, and a staleness correction mechanism to keep the cache coherent amid dynamic KV updates. The approach also incorporates streaming prefill to lower peak memory usage, achieving up to 48× longer serviceable context, 3.73× faster inference in offload‑based settings, and competitive task accuracy.