Generalized Parallel Scaling with Interdependent Generations
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911451588853760 |
|---|---|
| author | Dong, Harry Brandfonbrener, David Helenowski, Eryk He, Yun Kumar, Mrinal Fang, Han Chi, Yuejie Sankararaman, Karthik Abinav |
| author_facet | Dong, Harry Brandfonbrener, David Helenowski, Eryk He, Yun Kumar, Mrinal Fang, Han Chi, Yuejie Sankararaman, Karthik Abinav |
| contents | Parallel LLM inference scaling involves sampling a set of $N>1$ responses for a single input prompt. However, these $N$ parallel responses tend to be generated independently from each other, partitioning compute resources and leaving potentially useful information in one generation untapped by others. This is in contrast to response length scaling where past computation is used in all future steps. For higher quality responses and response sets, we propose Bridge to generate interdependent responses in parallel by rethinking batched LLM hidden states as holistic tensors rather than independent slices. With only a small amount (2.8%-5.1%) of new parameters, Bridge improves the relative mean accuracy gains from reinforcement learning with verifiable rewards by up to 39% and boosts consistency of correct responses. Trained once, Bridge scales to any generation width, all with greater performance than independent generations, unlocking a more general mode of parallel scaling that effectively leverages information between sequences, compatible with any post-generation aggregation technique. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_01143 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Generalized Parallel Scaling with Interdependent Generations Dong, Harry Brandfonbrener, David Helenowski, Eryk He, Yun Kumar, Mrinal Fang, Han Chi, Yuejie Sankararaman, Karthik Abinav Artificial Intelligence Machine Learning Parallel LLM inference scaling involves sampling a set of $N>1$ responses for a single input prompt. However, these $N$ parallel responses tend to be generated independently from each other, partitioning compute resources and leaving potentially useful information in one generation untapped by others. This is in contrast to response length scaling where past computation is used in all future steps. For higher quality responses and response sets, we propose Bridge to generate interdependent responses in parallel by rethinking batched LLM hidden states as holistic tensors rather than independent slices. With only a small amount (2.8%-5.1%) of new parameters, Bridge improves the relative mean accuracy gains from reinforcement learning with verifiable rewards by up to 39% and boosts consistency of correct responses. Trained once, Bridge scales to any generation width, all with greater performance than independent generations, unlocking a more general mode of parallel scaling that effectively leverages information between sequences, compatible with any post-generation aggregation technique. |
| title | Generalized Parallel Scaling with Interdependent Generations |
| topic | Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2510.01143 |