Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2605.27701 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913175593549824 |
|---|---|
| author | Renard, Arthur Gabriel, Franck Hartmann, Valentin Hongler, Clément |
| author_facet | Renard, Arthur Gabriel, Franck Hartmann, Valentin Hongler, Clément |
| contents | We present Frost Training, a method for improving Monte Carlo-based policy optimization for a large family of LLM-as-a-judge tasks called Cross-Entropy Games. The key idea is to exploit the gradient of the reward function in embedding space. This signal is used in the Greedy Coordinate Gradient (GCG) jailbreaking technique; we demonstrate for the first time that it can also be used to boost model training. We validate our method using GRPO training for maximum-likelihood infilling. Frost Training improves the model's ability to generate high-scoring outputs, reaching higher maximum scores in a best-of-k setting, and does so at an increased speed. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_27701 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Cross-Entropy Games and Frost Training Renard, Arthur Gabriel, Franck Hartmann, Valentin Hongler, Clément Artificial Intelligence We present Frost Training, a method for improving Monte Carlo-based policy optimization for a large family of LLM-as-a-judge tasks called Cross-Entropy Games. The key idea is to exploit the gradient of the reward function in embedding space. This signal is used in the Greedy Coordinate Gradient (GCG) jailbreaking technique; we demonstrate for the first time that it can also be used to boost model training. We validate our method using GRPO training for maximum-likelihood infilling. Frost Training improves the model's ability to generate high-scoring outputs, reaching higher maximum scores in a best-of-k setting, and does so at an increased speed. |
| title | Cross-Entropy Games and Frost Training |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2605.27701 |