[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-liquid-ai-lfm2-8b-vocab-128k":3,"news-related-d5bd25fb-f812-43c7-8246-18d7162dc76c":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"d5bd25fb-f812-43c7-8246-18d7162dc76c","Liquid AI 把 LFM2-8B-A1B 词表扩到 128K，decode 提速 3.7×","Liquid AI 在 arXiv (2607.15232) 发布了一种 LLM 词表原地扩展方法：在已有 tokenizer 上「续 BPE」，把 8B MoE 模型 LFM2-8B-A1B 词表扩到 128K，Hindi \u002F Vietnamese \u002F Thai 等低资源语言每字符 decode 速度提升 2.2-3.7×，模型权重与扩展 tokenizer 已开源。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.15232","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":18,"name":19,"slug":19,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":21,"name":22,"slug":22,"description":13,"color":13},"045c011e-e2bb-45ce-bdd6-0c927f8a3b87","token-efficiency",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"fda4e242-fe80-44a4-b373-b71cafc74baa","en","Liquid AI grows LFM2-8B-A1B vocab to 128K, 3.7x decode","Liquid AI published a vocabulary-extension-in-place method on arXiv (2607.15232): \"continuing BPE\" on an existing tokenizer, extending the 8B MoE model LFM2-8B-A1B vocabulary to 128K. Per-character decode speed for low-resource languages like Hindi \u002F Vietnamese \u002F Thai jumped 2.2-3.7×, with the model weights and extended tokenizer open-sourced.","liquid-ai-lfm2-8b-vocab-128k","2026-07-18T14:30:00Z","2026-07-18T14:13:43.879200Z","2026-08-19T02:08:40.142862Z",true,"agent",205,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"28c41f06-d20f-481c-b133-cd109af3aed1","答对之后停不下来:微软团队揪出在线蒸馏的 EOS 错配元凶","eos-mismatch-opd-length-inflation","2026-09-18T21:09:06+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"2e27016d-b90e-45c7-825a-41fd1e435c80","JHU 新研究:组合持续学习机制,百任务记忆留存从 1.2% 提到 34.9%","compose-cl-long-horizon-memorization","2026-09-16T15:10:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"30fca629-bace-4832-9789-b44aa8c8989d","学生团队从零训出开源 7B 模型 ZGCM-1:数学推理硬刚 235B 前沿","zgcm-1-open-7b-foundation-model","2026-09-15T19:10:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"2731ed1c-17c3-4d85-9174-983cf50743e3","地铁售票机上的 AI 大考:2.6GB 端侧模型 91.32 分超 GPT-5.6,规则基线也拿 84.6","metrollm-bench-transit-kiosk-llm","2026-09-12T23:08:18+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"2638aeac-dc4d-4b73-b7fe-2b042015adee","OreoLook 开源:三层缓存把 AI 搜索搬进 8 核 CPU,重复问题 0.1 毫秒出答案","oreolook-three-layer-cpu-cache","2026-09-10T23:08:36+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"3096df88-7158-4ffe-9356-1a83b829633b","A*-Thought-V2:把思维链塞进隐空间,回复砍半,平均精度反升","astar-thought-v2-latent-cot-compression","2026-09-09T15:10:00+00:00"]