Liquid AI Releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: Bidirectional Encoders That Stay Fast at 8K Context on CPU
https://www.marktechpost.com/2026/07/29/liquid-ai-releases-lfm2-5-encoder-230m-and-lfm2-5-encoder-350m-bidirectional-encoders-that-stay-fast-at-8k-context-on-cpu/📌 【Liquid AI 新發佈】LFM2.5 雙向編碼器：在 CPU 上處理 8K 長文本依然高效
TL;DR：Liquid AI 推出兩款輕量級雙向編碼器，透過 LFM2 混合架構，讓 CPU 在 8K 上下文下仍具備高效率。
🤔 解決持續運行的分類與過濾任務
在現代 AI 流程中，編碼器 (Encoder) 經常位於分類器、意圖路由 (intent routers)、安全過濾器 (safety filters) 與 PII（個人識別資訊）檢測器的底層。這些任務通常需要持續不斷地運行，且為了節省成本，大多在沒有 GPU 的環境下於 CPU 上執行。隨著輸入長度增加，如何在維持效能的同時處理更長的文本，成為關鍵挑戰。
🧩 基於 LFM2 混合架構的設計理念
Liquid AI 推出的 LFM2.5-Encoder-230M 與 LFM2.5-Encoder-350M 並非從零開始訓練，而是從 LFM2.5 的解碼器 (decoder) 骨幹模型進行轉換。其核心架構設計包含：
- 混合架構：將門控短卷積塊 (gated short-convolution blocks) 與分組查詢注意力 (grouped-query attention) 交錯排列。
- 長文本優勢：由於 LFM2 架構的成本隨輸入長度增加的成長速度較慢，因此在處理長文本時具備優勢。
- 模型規格：兩款模型均擁有 1024 的隱藏層維度 (hidden size)、65,536 的詞彙表大小 (vocabulary)，並支援 15 種語言。
📊 以較小規模實現高精準度排名
Liquid AI 針對 17 項任務（涵蓋 GLUE、SuperGLUE 及多語言分類）對 14 個模型進行評估，所有模型皆經過針對任務的完全微調 (fine-tuned)。
LFM2.5-Encoder-350M 表現：
- 17 項任務平均分數為 81.02 (±1.00)，在評估中排名第四。
- 領先的型號（XLM-R XL、ModernBERT-large、XLM-R large）規模均比它大。
- 值得注意的是，排名第一的模型規模幾乎是它的 10 倍。
LFM2.5-Encoder-230M 表現：
- 17 項任務平均分數為 79.29 (±1.02)，排名第六。
- 表現優於 ModernBERT-base (78.19) 以及所有 EuroBERT 模型（包括 EuroBERT-610M 與 EuroBERT-2.1B）。
- 效能亦超越了 Liquid AI 自身的檢索型模型（ColBERT 與 Embedding 系列）。
🎯 實務啟示
對於需要在 CPU 環境下進行高頻率、長文本處理（如安全過濾或意圖識別）的工程師來說，這類具備高效長文本處理能力且規模較小的雙向編碼器，提供了一個在低硬體成本下維持高精準度的選擇。
🔗 來源
- 標題：Liquid AI Releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: Bidirectional Encoders That Stay Fast at 8K Context on CPU
- 作者／機構：Asif Razzaq @ MarkTechPost
- 連結：https://www.marktechpost.com/2026/07/29/liquid-ai-releases-lfm2-5-encoder-230m-and-lfm2-5-encoder-350m-bidirectional-encoders-that-stay-fast-at-8k-context-on-cpu/
#LiquidAI #Encoder #MachineLearning #NLP #LFM2 #CPUInference #BidirectionalModel #Transformer #DeepLearning #OpenSource
原始資料 MarkTechPost · 收集於 2026-07-30
摘要原文
Liquid AI has released two open-weight bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M . Both are masked language models built on the LFM2 hybrid backbone. Both carry an 8,192-token context. Encoders sit underneath classifiers, intent routers, safety filters, and PII detectors. Those jobs run continuously, usually without a GPU, and increasingly on longer inputs. BERT established the class. ModernBERT pushed its accuracy, speed, and context. Liquid AI’s argument is that the LFM2 architecture continues that line, because its cost grows more slowly as inputs get longer. The encoders are not trained from scratch. They are initialized from the LFM2.5-230M and LFM2.5-350M decoder backbones, then converted with three changes: Architecturally, the backbone interleaves gated short-convolution blocks with grouped-query attention, the same design described in the LFM2 technical report . Both checkpoints use a hidden size of 1024 and a 65,536-token vocabulary, and support 15 languages. The license is the LFM Open License v1.0. The embed below breaks down the conversion, the full ranking table, the per-task scores, and the CPU latency figures. Liquid AI evaluated 14 models on 17 tasks pulled from GLUE, SuperGLUE, and multilingual classification. Every model is fully fine-tuned per task, and the reported score is that fine-tuned model’s result. LFM2.5-Encoder-350M posts a 17-task mean of 81.02 (±1.00), ranking fourth. The three models ahead of it are all larger. They are XLM-R XL at 3.5B (83.06), ModernBERT-large at 395M (81.68), and XLM-R large at 560M (81.34). The top model is nearly 10 times its size. LFM2.5-Encoder-230M posts 79.29 (±1.02), ranking sixth. It beats ModernBERT-base at 78.19 and every EuroBERT model in the table, including EuroBERT-610M (75.87) and EuroBERT-2.1B (72.19). Both new encoders also score above Liquid AI’s own retrieval siblings, LFM2.5-ColBERT-350M (76.18) and LFM2.5-Embedding-350M (75.68). That gap is the stated reason Liquid AI
由 tencent/hy3:free 自動生成