Smooth Gate Functions for Soft Advantage Policy Optimization
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910069998747648 |
|---|---|
| author | Denisov, Egor Glazyrina, Svetlana Kryzhanovskiy, Maksim Ischenko, Roman |
| author_facet | Denisov, Egor Glazyrina, Svetlana Kryzhanovskiy, Maksim Ischenko, Roman |
| contents | Group Relative Policy Optimization (GRPO) has significantly advanced the training of large language models and enhanced their reasoning capabilities, while it remains susceptible to instability due to the use of hard clipping. Soft Adaptive Policy Optimization (SAPO) addresses this limitation by replacing clipping with a smooth sigmoid-based gate function, which leads to more stable updates. We have decided to push this theory further and investigate the impact of different gate functions on both training stability and final model performance. We formalize the key properties that admissible gates should satisfy and identify several families of such functions for empirical evaluation. This paper presents an analysis of our findings based on experiments conducted with the Qwen2.5-7B-Instruct model on mathematical reasoning tasks. These results provide practical guidance for designing smoother and more robust policy optimization objectives for large language model training. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_19345 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Smooth Gate Functions for Soft Advantage Policy Optimization Denisov, Egor Glazyrina, Svetlana Kryzhanovskiy, Maksim Ischenko, Roman Machine Learning Artificial Intelligence Group Relative Policy Optimization (GRPO) has significantly advanced the training of large language models and enhanced their reasoning capabilities, while it remains susceptible to instability due to the use of hard clipping. Soft Adaptive Policy Optimization (SAPO) addresses this limitation by replacing clipping with a smooth sigmoid-based gate function, which leads to more stable updates. We have decided to push this theory further and investigate the impact of different gate functions on both training stability and final model performance. We formalize the key properties that admissible gates should satisfy and identify several families of such functions for empirical evaluation. This paper presents an analysis of our findings based on experiments conducted with the Qwen2.5-7B-Instruct model on mathematical reasoning tasks. These results provide practical guidance for designing smoother and more robust policy optimization objectives for large language model training. |
| title | Smooth Gate Functions for Soft Advantage Policy Optimization |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2602.19345 |