Research map / Probabilistic and scientific modeling / Dynamical systems
Text Generation as Continuous Latent Dynamics via Reinforcement Learning
Chen Jia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add KL regularization to the reference LLM (ablation) | CT-LLM · AlpacaEval 2 | length-controlled win rate: 15.4 -> 19.2 | Improved |
| add KL regularization to the reference LLM (ablation) | CT-LLM · AlpacaEval 2 | ROUGE-L: 30.5 -> 36.5 | Improved |
| add KL regularization to the reference LLM (ablation) | CT-LLM · AlpacaEval 2 | raw win rate: 12.6 -> 16.6 | Improved |
| add KL regularization | Transformer Dec-only · AlpacaEval 2 | length-controlled win rate: 13.1 -> 12.8 | No clear effect |
| add KL regularization | Transformer Enc-Dec · AlpacaEval 2 | length-controlled win rate: 6.2 -> 8.9 | Improved |
| replace Transformer Dec-only with continuous latent ODE policy (both RL+KL) | CT-LLM · AlpacaEval 2 | length-controlled win rate: 12.8 -> 19.2 | Improved |
| replace Transformer Dec-only with continuous latent ODE policy (both RL+KL) | CT-LLM · AlpacaEval 2 | ROUGE-L: 32.5 -> 36.5 | Improved |
| scale CT-LLM from 0.4B to 2.1B parameters | CT-LLM · AlpacaEval 2 (model scaling) | length-controlled win rate: 3.6 -> 22.0 | Improved |
| scale CT-LLM from 0.4B to 2.1B parameters | CT-LLM · AlpacaEval 2 (model scaling) | perplexity: 9.8 -> 6.8 | Improved |
| scale CT-LLM from 1.2B to 2.1B parameters | CT-LLM · AlpacaEval 2 (model scaling) | length-controlled win rate: 19.2 -> 22.0 | Improved |
| use CT-LLM instead of 8B Transformer under same training FLOPs | CT-LLM · AlpacaEval 2 (compute-matched, 1.4e20 FLOPs) | ROUGE-L: 30.2 -> 36.8 | Improved |
| use CT-LLM instead of 1.7B Transformer under same training FLOPs | CT-LLM · AlpacaEval 2 (compute-matched, 1.4e20 FLOPs) | ROUGE-L: 28.5 -> 36.8 | Improved |
Closest papers by meaning
- Continuous Latent Diffusion Language Model (NeurIPS 2026 · similarity 0.87)
- Successor Features for Efficient Multi-Subject Controlled Text Generation (ICML 2024 · similarity 0.86)
- From Self-Attention to Markov Models: Unveiling the Dynamics of Generative Transformers (ICML 2024 · similarity 0.86)
- Embedding-Aligned Language Models (NeurIPS 2024 · similarity 0.86)
- Energy-Based Diffusion Language Models for Text Generation (ICLR 2025 · similarity 0.86)
- TEncDM: Understanding the Properties of the Diffusion Model in the Space of Language Model Encodings (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
