Tag
This paper proposes a composite objective for cone beam CT report generation that prioritizes factual entailment over lexical overlap, demonstrating that optimizing for lexical metrics harms factual accuracy. It releases a dataset and code, and presents a system that generates constrained clinical reports under polarity, laterality, and tooth level consistency constraints.
This paper investigates the impact of constrained decoding on tool-call abstention, decomposing the effects of grammar constraints into stop and emission components, and evaluates performance on small open-weight models across languages.
This paper introduces the trie automaton, a specialized constrained decoding mechanism for finite sets that precomputes token masks via Aho-Corasick matching, achieving up to 29x end-to-end throughput improvements over XGrammar in vLLM batch serving while guaranteeing 100% output validity.
The paper presents GrocLM, a fine-tuned language model for grocery category recommendation using a two-stage LoRA-based training strategy and trie-based constrained decoding. It achieves a 7.5% relative improvement in cart-adds per impression in a live production restocking task.
This paper describes LLM-INSTRUCT, the winning system for the UZH Shared Task at ArgMining 2026 on paragraph-level argument mining. The system uses constraint-aware retrieval and selective debate to improve accuracy and schema compliance.
This paper introduces benchmarks for natural language to MLIR code generation across multiple dialects and a schema-derived constrained decoding stack that allows a small language model to match or exceed large code LMs on structural verifier tasks, without retraining.
This paper introduces PCC-SQL, a method for enforcing column-use policies in text-to-SQL generation by constrained decoding, achieving deterministic elimination of violations with 0% Leakage Rate and high coverage on benchmarks.
This paper introduces generative compilation, a method to obtain compiler feedback on partial programs during AI code generation, using a 'sealor' transformation that enables standard compilers to diagnose incomplete code. Evaluated on Rust coding tasks, it reduces non-compiling outputs and improves functional correctness by catching errors early.
FlashTrie presents a GPU-accelerated constrained beam search for generative retrieval, using a succinct trie layout and cooperative CUDA kernels to reduce decoding latency and enable real-time serving at scale, achieving up to 24× speedup and a 0.71% revenue lift in a commercial search engine.
Weave of Formal Thought (WoFT) introduces a sound and complete constrained decoder for code generation that guarantees syntactic validity relative to the full Tree-sitter specification, and a fine-tuning method that trains models to interleave grammar symbols using reweighted wake-sleep, improving perplexity on Python code generation.
This paper presents EPIC, an efficient framework for context-free grammar constrained decoding in diffusion language models that reduces inference time by up to 67.5% while maintaining syntactic correctness.
A new hybrid decoding framework called In-Writing is proposed, which delays constraint application until after a trigger token, combining free-form reasoning with structured generation for improved accuracy in classification and reasoning tasks.