Metraipolis › archive › dc_3ddeb6f28be8

From Experience to Expertise: Adoption-Aware Memory Learning for Data-Scarce NPU Kernel Synthesis — arXiv:2609.35568

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.35568
License (as archived)
CC0-1.0
Archived
2026-10-07T22:39:14.529Z
SHA-256 of the archived text
09d5f0ba8610b022168a8c086e6979cfb87348771e773490e3f36eacbfd10c3d
Stable URL of this page
https://metraipolis.com/documento/dc_3ddeb6f28be8

How to cite

From Experience to Expertise: Adoption-Aware Memory Learning for Data-Scarce NPU Kernel Synthesis — arXiv:2609.35568 — arXiv (en), https://arxiv.org/abs/2609.35568 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3ddeb6f28be8 (Metraipolis archive dc_3ddeb6f28be8, sha256 09d5f0ba8610b022168a8c086e6979cfb87348771e773490e3f36eacbfd10c3d)
BibTeX
@misc{dc3ddeb6f28be8,
  title = {From Experience to Expertise: Adoption-Aware Memory Learning for Data-Scarce NPU Kernel Synthesis — arXiv:2609.35568},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.35568},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3ddeb6f28be8; sha256 09d5f0ba8610b022168a8c086e6979cfb87348771e773490e3f36eacbfd10c3d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35568. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

High-performance kernels underpin efficient accelerator execution but require expert tuning and lengthy manual optimization cycles. LLM coding agents promise automation, yet their CUDA knowledge transfers poorly to data-scarce domain-specific architectures (DSAs) such as NPUs, whose execution models and memory hierarchies differ substantially from those of GPUs. To address this transfer gap, post-training methods adapt LLMs to NPU programming but depend on scarce expert data and substantial training compute. Memory-learning agents instead adapt through external memory, but their uniform credit assignment gives adopted and unused experiences the same reward target, potentially biasing subsequent retrieval rankings. Moreover, when learned values guide only retrieval, high-value experiences that generalize across operators must be retrieved repeatedly rather than retained in context, thereby increasing retrieval overhead and weakening cross-task guidance. We therefore present SAGE, a persistent self-improving agent for NPU kernel synthesis. Adoption-Traced Utility estimation (ATU) combines explicit adoption records with kernel evaluation outcomes for adoption-aware credit assignment. Utility-Gated Consolidation (UGC) uses positive utility and repeated adoption across operators to select and abstract reusable rules into a bounded resident context. On NPUKernelBench, SAGE achieves a 95.5% execution rate versus 84.1% for the strongest controlled baseline, with 86.9% of solved operators outperforming torch_npu. With GLM-5.3, SAGE achieves a 43.99x speedup over the torch_npu reference on sparse flash attention. These results show that adoption-aware credit assignment and selective consolidation enable agents to accumulate and reuse hardware-specific knowledge across tasks.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.35568 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T22:14:28.476Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).