Liquid AI Releases LFM2.5-2.6B: An On-Device Agentic Model With 128K Context, Tool Calling, And Open Weights
https://www.marktechpost.com/2026/08/06/liquid-ai-lfm2-5-2-6b-on-device-agentic-model/📌 【Liquid AI】推出 LFM2.5-2.6B：專為裝置端設計，具備 128K 上下文與強大工具呼叫能力
TL;DR：Liquid AI 發布 LFM2.5-2.6B 開源模型，可在手機與筆電本地端執行，具備強大的 Agentic 任務處理能力。
隨著邊緣運算（Edge Computing）需求增加，如何在裝置端（On-device）實現具備代理能力（Agentic）的 AI 成為關鍵。Liquid AI 推出的 LFM2.5-2.6B 模型，旨在讓手機、筆電、PC 甚至機器人，都能在不將數據傳送到雲端的情況下，執行多步驟的複雜任務。
🧩 混合架構設計：結合捲積與 Attention
LFM2.5-2.6B 擁有 2.69B 總參數，結構包含 30 層，其設計核心在於結合了兩種不同的機制：
- 22 層雙閘門短捲積區塊（double-gated short convolution blocks）。
- 8 層群組查詢注意力區塊（grouped-query attention blocks）。
為了提升效率，Liquid AI 並非從頭訓練，而是透過以下方式優化：
- 詞彙表擴張：透過對現有 tokenizer 進行就地（in place）擴展，將詞彙量提升至 128,000。
- 上下文延伸：透過專門的中期訓練（mid-training）階段，將上下文長度延伸至 131,072 tokens。
📊 效能表現：以小體量挑戰大規模模型
根據 Liquid AI 的測試，LFM2.5-2.6B 在指令遵循（instruction-following）與工具使用（tool-use）的表現，足以媲美規模大其近四倍的模型。
| 模型名稱 | 參數規模 | 說明 |
|---|---|---|
| LFM2.5-2.6B | 2.69B | 本專案核心模型 |
| gemma-4-E2B-it | 5.1B | 對照組 |
| gemma-4-E4B-it | 8B | 對照組 |
| Qwen3.5-4B | 4.7B | 對照組 |
| Qwen3.5-9B | 9.7B | 對照組 |
- 指令遵循與工具使用：LFM2.5-2.6B 在多項基準測試中領先，僅在 BFCLv4 測試中略遜於 Qwen3.5-9B。
- 程式碼能力：在 LiveCodeBenchv6 測試中，LFM2.5-2.6B 取得 59.41 分，而 Qwen3.5-9B 則為 69.86 分，顯示大型模型在編程領域仍保有優勢。
📦 開源與落地：支援多種本地端格式
為了讓開發者能立即部署，Liquid AI 釋出了兩個版本：
- LFM2.5-2.6B-Base：適合用於進一步微調（fine-tuning）。
- LFM2.5-2.6B post-trained：針對代理任務（agentic workloads）進行過後訓練。
這兩個版本皆透過 lfm1.0 授權在 Hugging Face 上公開。此外，權重支援多種原生格式（GGUF、MLX、ONNX），並在發布首日即支援 llama.cpp、vLLM、SGLang 與 LM Studio 等主流工具。
🎯 實務啟示
對於開發者而言，LFM2.5-2.6B 提供了一個極具潛力的本地端 Agent 解決方案。由於推理（inference）完全在裝置端完成，這不僅解決了隱私疑慮（數據無需離開裝置），更將每次運行的邊際成本降至接近於零，非常適合需要高頻率、低延遲且具備隱私需求的應用場景。
🔗 來源
- 標題：Liquid AI Releases LFM2.5-2.6B: An On-Device Agentic Model With 128K Context, Tool Calling, And Open Weights
- 作者／機構：Asif Razzaq @ MarkTechPost
- 連結：https://www.marktechpost.com/2026/08/06/liquid-ai-lfm2-5-2-6b-on-device-agentic-model/
#LiquidAI #LFM #OnDeviceAI #AgenticModel #OpenWeights #MachineLearning #LLM #EdgeAI #EdgeComputing #AIModels
原始資料 MarkTechPost · 收集於 2026-08-07
摘要原文
Liquid AI released LFM2.5-2.6B , an agentic model that runs entirely on-device. It plans, calls tools, and works through multi-step tasks on phones, laptops, PCs, and robots. The model has 2.69B total parameters, a 131,072-token context window, and a 128,000-token vocabulary. Pre-training used approximately 34 trillion tokens. Two checkpoints shipped: LFM2.5-2.6B-Base for fine-tuning, and LFM2.5-2.6B post-trained for agentic workloads. Because inference stays local, data never leaves the device and the marginal cost of each run is near zero. Liquid AI reports tool-use and instruction-following scores competitive with models nearly four times its size. The answer is Yes. Both checkpoints are public on Hugging Face under the lfm1.0 license. Weights ship in native, GGUF , MLX , and ONNX formats, with day-one support in llama.cpp, vLLM, SGLang, and LM Studio. LFM2.5-2.6B has 2.69B total parameters across 30 layers. The stack is 22 double-gated short convolution blocks plus 8 grouped-query attention blocks. Vocabulary size is 128,000 and context length is 131,072 tokens. Pre-training used approximately 34 trillion tokens. Liquid AI doubled the vocabulary to 128K by extending the existing tokenizer in place rather than retraining from scratch. A dedicated mid-training phase extends context to 128K. The model covers 16 languages and is text-only. The base checkpoint becomes an agent through four stages. Liquid AI compared LFM2.5-2.6B against gemma-4-E2B-it (5.1B), gemma-4-E4B-it (8B), Qwen3.5-4B (4.7B) and Qwen3.5-9B (9.7B). It leads every instruction-following benchmark reported and nearly every tool use benchmark, trailing Qwen3.5-9B only on BFCLv4. Coding is where larger models keep an edge: LiveCodeBenchv6 is 59.41 versus 69.86 for Qwen3.5-9B. Check out the Technical details , LFM2.5-2.6B , and LFM2.5-2.6B-Base . Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Liquid AI Releases LFM2.5-2.6B: An On-Device Agentic Model With 128K Context, Tool Calling, And Open Weights appeared first on MarkTechPost .
由 tencent/hy3:free 自動生成