Saved in:
Bibliographic Details
Main Authors: Peng, Kerui, Li, Feifei, Fan, Xingyu, Que, Wenhui
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.27971
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914606971092992
author Peng, Kerui
Li, Feifei
Fan, Xingyu
Que, Wenhui
author_facet Peng, Kerui
Li, Feifei
Fan, Xingyu
Que, Wenhui
contents When large language models are fine-tuned to generate persona- or tone-conditioned responses, their output diversity is severely limited--a failure we term Cross-Style Collapse. We trace this collapse to the cross-entropy objective, which under shared representations tends to suppress diverse continuations. We propose Semantic Flow Regularization (SFR), a lightweight auxiliary objective that supervises the backbone with continuous sentence-encoder embeddings of future segments via conditional flow matching. The stochastic flow source preserves multi-modality by construction; the flow-matching head is discarded at inference, adding zero deployment cost. On a large-scale industrial dialogue dataset (Qwen3-32B, 9 personas), SFR improves output diversity, style fidelity, and response quality over SFT. We further validate on the public LiveCodeBench-v5 (Qwen2.5-Coder-7B-Instruct), where SFR consistently improves pass@k, confirming generality beyond stylized dialogue. A controlled comparison on MBPP reveals Multi-Token Prediction to be a degenerate special case of SFR.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27971
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Semantic Flow Regularization: Teaching LLMs to Generate Diverse Yet Coherent Responses
Peng, Kerui
Li, Feifei
Fan, Xingyu
Que, Wenhui
Computation and Language
Artificial Intelligence
When large language models are fine-tuned to generate persona- or tone-conditioned responses, their output diversity is severely limited--a failure we term Cross-Style Collapse. We trace this collapse to the cross-entropy objective, which under shared representations tends to suppress diverse continuations. We propose Semantic Flow Regularization (SFR), a lightweight auxiliary objective that supervises the backbone with continuous sentence-encoder embeddings of future segments via conditional flow matching. The stochastic flow source preserves multi-modality by construction; the flow-matching head is discarded at inference, adding zero deployment cost. On a large-scale industrial dialogue dataset (Qwen3-32B, 9 personas), SFR improves output diversity, style fidelity, and response quality over SFT. We further validate on the public LiveCodeBench-v5 (Qwen2.5-Coder-7B-Instruct), where SFR consistently improves pass@k, confirming generality beyond stylized dialogue. A controlled comparison on MBPP reveals Multi-Token Prediction to be a degenerate special case of SFR.
title Semantic Flow Regularization: Teaching LLMs to Generate Diverse Yet Coherent Responses
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2605.27971