This paper investigates improving speculative decoding for multilingual LLM inference, finding that n-gram draft models offer consistent speed-ups across languages despite lower token acceptance rates…

The paper demonstrates that the AI-like style introduced by post-training alignment can be measured, localized, and causally removed using a novel ablation technique called PASTA.

This paper analyzes speech-text interleaved language models and reveals that they go through an implicit transcription phase in which spoken words become decodable as text in intermediate layers.

This paper investigates how LLMs handle multiple writing systems, finding that while they use shared latent representations, the model exhibits a structural bias that makes generating Latin script eas…

This study benchmarks token-optimized formats (TOON and TRON) against JSON in end-to-end agentic AI systems, finding that TRON significantly reduces token overhead with minimal performance degradation…

The paper quantitatively confirms the Currier A/B language distinction in the Voynich Manuscript, demonstrating it is governed by a higher-dimensional, context-dependent boolean switch rather than a s…

This paper investigates the redundancy of the prompt KV cache during language model decoding, finding that the structure provided by chat templates is the primary source of redundancy, not the actual…

This paper proposes a context engineering approach that scales memory with distinct information instead of tokens, using a novelty-gated cache and state-space summary.

The paper introduces CFGzip, an offline token space compression technique that significantly reduces the computational overhead of constrained decoding, making complex grammar enforcement feasible at…

This paper proposes a hybrid scheduler that dynamically switches between exclusive batching and mixed batching for LLM inference, achieving superior throughput, especially on bandwidth-constrained GPU…

PolySpeech-100 introduces a massive, multi-lingual benchmark covering 110 linguistic variants to rigorously test Speech-LLMs, demonstrating that open-source models struggle with low-resource languages…

The paper benchmarks local, offline LLMs for confidential translation workflows, demonstrating that while they are viable for privacy-sensitive use, they generally lag behind top commercial NMT system…

This paper proposes a new architecture for transformer-based language models called 'former', which allocates capacity non-uniformly across network depth by maintaining wider early and late layers whi…

This paper investigates the relationship between language model perplexity and word error rate in modern automatic speech recognition systems, studying the impact of external language models, encoder…

Weak self-training on synthetic data can amplify a language model's existing capabilities, but this effect is strictly dependent on the compatibility between the source and student models, not on the…

This paper demonstrates that compact, domain-specialized Automatic Speech Recognition (ASR) models significantly outperform large, general-purpose foundation models for conversational speech across 19…