Controllable Text Generation for Large Language Models: A Survey
-
Updated
Aug 27, 2024 - TeX
Controllable Text Generation for Large Language Models: A Survey
[EMNLP 2023] Composable text generation & editing by ODE sampling in the latent space of a pretrained-LM VAE — compose sentiment, tense, formality, keyword controls without retraining
awesome-LLM-controlled-constrained-generation
[ACL 2024]Controlled Text Generation for Large Language Model with Dynamic Attribute Graphs
The code implementation of the EMNLP2022 paper: DisCup: Discriminator Cooperative Unlikelihood Prompt-tuning for Controllable Text Generation
Interpretable stylometric profiling + author-style transfer built on explicit and LLM-genenated JSON fingerprints and local measurements. Fingerprint your corpus, inspect/visualize signals, then LLM-generate stylistically similar text with meaning preserved, deterministic post-processing, normalization controls, and deviation reports. CLI + API.
[NLPCC'23] ZeroGen: Zero-shot Multimodal Controllable Text Generation with Multiple Oracles PyTorch Implementation
Implementing Simulation-guided Beam Search for text generation. Idea already proposed in Choo et al for Neural Combinatorial Optimization. In this repo, we use the technique for Controllable Text Generation.
[ACL 2024] "Benchmarking and Improving Compositional Generalization of Multi-aspect Controllable Text Generation"
📝 Transform writing by creating stylometric profiles for personalized author-style text generation and rewriting.
STATe-of-Thoughts (Structured Action Templates for Tree-of-Thoughts), a method for modulating reasoning trajectories (influencing latent properties of reasoning steps like structure, style, and content).
Controllable text generation project using different approaches
Official code for "Unsupervised Text Style Transfer with Controllable Intensity" — a two-stage SFT + PPO framework for fine-grained control over text readability and sentiment transfer.
Code and checkpoints for "The Illusion of Control: Why Bare Classifier Inversion Silently Fails in Concept-Bottleneck Text Generation" (EMNLP 2026). Multi-attribute controllable text generation, concept bottleneck, compositional generalization.
To associate your repository with the controllable-text-generation topic, visit your repo's landing page and select "manage topics."