Papers with next-token generation
RomanLens: The Role Of Latent Romanization In Multilinguality In LLMs (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) exhibit strong multilingual performance despite training on English-centric corpora. |
| Approach: | They propose to use Romanization as a potential bridge in multilingual processing . they propose to encode semantic concepts similarly across native and Romanized scripts . |
| Outcome: | The proposed model encodes semantic concepts across native and Romanized scripts, suggesting a shared underlying representation. |
Steering LLM Thinking with Budget Guidance (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing budget control methods for large language models are inadequate for long reasoning . budget guidance can be used to control reasoning length without fine-tuning . |
| Approach: | They propose a budget guidance method that models a Gamma distribution over remaining thinking length during next-token generation and uses it to guide generation in a soft, token-level manner. |
| Outcome: | The proposed method achieves up to 26% accuracy gain on the MATH-500 benchmark compared to baseline methods while maintaining competitive accuracy with only 63% of the thinking tokens used by the full-thinking model. |