English(EN) ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression

ESTS在WMT26上使用GPT-OSS-20B和GPT-5.1进行模型压缩的细节

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-14 04:00

ESTS的研究人员详细介绍了他们提交给WMT26模型压缩共享任务的工作，重点关注英译简中和英译埃及阿拉伯语的翻译。他们的方法包括根据特定任务的路由质量和跨语言路由发散度从GPT-OSS-20B中剪枝专家，然后使用GPT-5.1生成的合成数据对剩余的专家进行微调。通过对保留的专家权重使用MXFP4量化实现进一步压缩，模型参数量在4.186B到7.770B之间。 AI

影响这项研究展示了压缩大型语言模型的先进技术，有可能实现更高效的翻译系统部署。

排序理由该项目是一篇研究论文，详细介绍了提交给共享任务的模型压缩方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

ESTS在WMT26上使用GPT-OSS-20B和GPT-5.1进行模型压缩的细节

本文如何被排名

Signal score: 23 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该项目是一篇研究论文，详细介绍了提交给共享任务的模型压缩方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Liu O. Martin, Lucas Bandarkar, Nanyun Peng · 2026-09-14 04:00

ESTS at WMT26: 面向模型压缩的路由感知专家剪枝

arXiv:2609.12310v1 Announce Type: new Abstract: We describe six submissions under the team name ESTS to the unconstrained WMT26 Model Compression Shared Task for English--Simplified Chinese and English--Egyptian Arabic. We submit three compression operating points per translation…

本文如何被排名

报道来源 [1]

ESTS at WMT26: 面向模型压缩的路由感知专家剪枝

相关实体

相关话题