Metraipolis › archive › dc_f8839aa680bf

Shared Context Candidate Scoring with Jointly Adapted Readout Networks — arXiv:2611.04218

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.04218
License (as archived)
CC0-1.0
Archived
2026-11-04T21:42:24.790Z
SHA-256 of the archived text
d3816ea69d526880e54a7360567962cd19fd2ffc5c0bff2916591f95b90bd2e0
Stable URL of this page
https://metraipolis.com/documento/dc_f8839aa680bf

How to cite

Shared Context Candidate Scoring with Jointly Adapted Readout Networks — arXiv:2611.04218 — arXiv (en), https://arxiv.org/abs/2611.04218 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f8839aa680bf (Metraipolis archive dc_f8839aa680bf, sha256 d3816ea69d526880e54a7360567962cd19fd2ffc5c0bff2916591f95b90bd2e0)
BibTeX
@misc{dcf8839aa680bf,
  title = {Shared Context Candidate Scoring with Jointly Adapted Readout Networks — arXiv:2611.04218},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04218},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f8839aa680bf; sha256 d3816ea69d526880e54a7360567962cd19fd2ffc5c0bff2916591f95b90bd2e0},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04218. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Repeated decisions over a shared document create two distinct opportunities: reusing document computation and adapting how question representations score candidate answers. We present Q64, a candidate-scoring system that preserves a frozen 64-layer Qwen3.8-27B text backbone, reuses its hybrid prefix state and intermediate document representations, and jointly trains a residual adapter with a four-layer decoder and candidate-conditioned readout. We used ZiYor, a self-evolving AI system, to explore and refine the Q64 design. We specified the scoring method, implemented the system, and evaluated its accuracy and latency. The final adaptation stage updates 88.66 million parameters using 8,192 selected training views. On a fixed 5,112-item evaluation combining ARC-Challenge, ANLI-R3, and stratified subsets of MMLU-Pro and CMMLU, joint adaptation improves four-benchmark macro accuracy from 66.46% to 74.32%. It remains below the native 27B baseline at 75.68% and AnyJev Qwen3-32B L0 at 76.68%. On a separate synthetic workload with 4,096 document tokens and 50 questions, median cold-request latency is 2.237 seconds, compared with 5.383 seconds for the native baseline and 56.872 seconds for the measured batched AnyJev configuration. Short single-question requests do not share this advantage. The evidence supports an accuracy-latency tradeoff in the evaluated systems, rather than a backbone reduction or an isolated component advantage. Public benchmarks informed earlier development, and real-document multi-question validation remains open.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.04218 — 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 2026-11-04T21:36:04.415Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).