mlsanityData to 7 Oct 2026

Semantic Flow Regularization: Teaching LLMs to Generate Diverse Yet Coherent Responses

arXiv 2605.27971 ↗First posted 27 May 2026cs.CL cs.AIcited by 0

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 40 of its 55 references that are in the map. First posted on arXiv 27 May 2026.

Builds on

40 of 55 references.

References · 40

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,550
  4. 2021Program Synthesis with Large Language Modelscited by 3,509
  5. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 2,994
  6. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  7. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  8. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,838
  9. 2023Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generationcited by 1,332
  10. 2019ZeRO: Memory Optimizations Toward Training Trillion Parameter Modelscited by 1,278
  11. 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,234
  12. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,107
28 more
  1. 2015A Diversity-Promoting Objective Function for Neural Conversation Modelscited by 1,068
  2. 2022Diffusion-LM Improves Controllable Text Generationcited by 1,006
  3. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 982
  4. 2018Personalizing Dialogue Agents: I have a dog, do you have pets too?cited by 820
  5. 2019Plug and Play Language Models: A Simple Approach to Controlled Text Generationcited by 793
  6. 2019CTRL: A Conditional Transformer Language Model for Controllable Generationcited by 754
  7. 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
  8. 2024Revisiting Feature Prediction for Learning Visual Representations from Videocited by 490
  9. 2019Neural Text Generation with Unlikelihood Trainingcited by 489
  10. 2018Texygen: A Benchmarking Platform for Text Generation Modelscited by 484
  11. 2020GeDi: Generative Discriminator Guided Sequence Generationcited by 377
  12. 2021FUDGE: Controlled Text Generation With Future Discriminatorscited by 374
  13. 2024Better & Faster Large Language Models via Multi-token Predictioncited by 344
  14. 2022Rectified Flow: A Marginal Preserving Approach to Optimal Transportcited by 277
  15. 2023Efficient Diffusion Training via Min-SNR Weighting Strategycited by 242
  16. 2022A Contrastive Framework for Neural Text Generationcited by 224
  17. 2017Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoderscited by 222
  18. 2019The Second Conversational Intelligence Challenge (ConvAI2)cited by 178
  19. 2016A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialoguescited by 174
  20. 2019Learning to Speak and Act in a Fantasy Text Adventure Gamecited by 122
  21. 2024Large Concept Models: Language Modeling in a Sentence Representation Spacecited by 86
  22. 2025OpenCodeInstruct: A Large-scale Instruction Tuning Dataset for Code LLMscited by 57
  23. 2025rStar-Coder: Scaling Competitive Code Reasoning with a Large-Scale Verified Datasetcited by 49
  24. 2026Embarrassingly Simple Self-Distillation Improves Code Generationcited by 36
  25. 2019Structuring Latent Spaces for Stylized Response Generationcited by 25
  26. 2023MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropiescited by 16
  27. 2023Text Style Transfer Evaluation Using Large Language Modelscited by 11
  28. 2025One Agent to Serve All: a Lite-Adaptive Stylized AI Assistant for Millions of Multi-Style Official Accountscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026ConvergeFlow: Language Flow with Provable Convergence to Token Embeddings
  2. 2026Towards Faster Language Model Inference Using Mixture-of-Experts Flow Matching
  3. 2025Learning LLM Preference over Intra-Dialogue Pairs: A Framework for Utterance-level Understandings
  4. 2026Flow Map Language Models: One-step Language Modeling via Continuous Denoising
  5. 2025Large Language Diffusion Models
  6. 2026Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding Spaces
  7. 2025FlowBind: Efficient Any-to-Any Generation with Bidirectional Flows
  8. 2026LangFlow: Continuous Diffusion Rivals Discrete in Language Modeling