The Conductor and the Engine: A Path Towards Co-Designed Reasoning
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916967073447936 |
|---|---|
| author | Wang, Yuanxin Filipczuk, Pawel Garg, Anisha Dhada, Amaan Hassanpour, Mohammad Bick, David Venkatesh, Ganesh |
| author_facet | Wang, Yuanxin Filipczuk, Pawel Garg, Anisha Dhada, Amaan Hassanpour, Mohammad Bick, David Venkatesh, Ganesh |
| contents | Modern LLM reasoning relies on extensive test-time computation, driven by internal model training and external agentic orchestration. However, this synergy is often inefficient, as model verbosity and poor instruction following lead to wasted compute. We analyze this capability-cost trade-off and introduce an optimized reasoning workflow (\cepo) that empowers smaller open-source models to outperform models multiple times their size. We will open-source this workflow to enable further research. Our work demonstrates a clear path toward co-designing orchestration frameworks with the underlying model capabilities to unlock powerful reasoning in small-to-medium sized models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_19762 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | The Conductor and the Engine: A Path Towards Co-Designed Reasoning Wang, Yuanxin Filipczuk, Pawel Garg, Anisha Dhada, Amaan Hassanpour, Mohammad Bick, David Venkatesh, Ganesh Artificial Intelligence Modern LLM reasoning relies on extensive test-time computation, driven by internal model training and external agentic orchestration. However, this synergy is often inefficient, as model verbosity and poor instruction following lead to wasted compute. We analyze this capability-cost trade-off and introduce an optimized reasoning workflow (\cepo) that empowers smaller open-source models to outperform models multiple times their size. We will open-source this workflow to enable further research. Our work demonstrates a clear path toward co-designing orchestration frameworks with the underlying model capabilities to unlock powerful reasoning in small-to-medium sized models. |
| title | The Conductor and the Engine: A Path Towards Co-Designed Reasoning |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2509.19762 |