Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.26720 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914602929881088 |
|---|---|
| author | Chong, Yee Hin Wu, Jiaming Zhang, Youhui Qu, Peng |
| author_facet | Chong, Yee Hin Wu, Jiaming Zhang, Youhui Qu, Peng |
| contents | Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift.
We introduce \texttt{CUDAnalyst}, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. \texttt{CUDAnalyst} enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_26720 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation Chong, Yee Hin Wu, Jiaming Zhang, Youhui Qu, Peng Artificial Intelligence Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce \texttt{CUDAnalyst}, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. \texttt{CUDAnalyst} enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied. |
| title | Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2605.26720 |