Liquid AI

LFM2.5 8B A1B

Name: LFM2.5 8B A1B
Rating: 74 (193 reviews)
Author: Liquid AI

前沿

HuggingFace

116.1K下载量658点赞May 2026发布日期128K tokens上下文Other许可证58 良好质量

LFM2.5 8B A1B (8.5B parameters) requires approximately 6.9 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 1.5B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 8 GB of VRAM.

快速开始

— 复制粘贴即可本地运行

Copy-paste commands to run LFM2.5 8B A1B on your machine.

Run

lms load LFM2.5-8B-A1B && lms server start

Quick specs

Parameters8.5B (1.5B active)

Architecturemoe (MoE)

Context128K tokens

Modalitytext

Min RAM3.3 GB

Rec. RAM5.2 GB (Q4_K_M)

LicenseOther

FamilyLFM2

✓ Chat

About this model

LFM2.5-8B-A1B is Liquid AI's on-device MoE assistant: 8.3B total parameters with only 1.5B activated per token (32 experts, 4 active). Its hybrid convolution + attention backbone is optimized for fast, low-memory edge inference on consumer hardware.

•Only ~1.5B active parameters per token — MoE efficiency for on-device use.
•Hybrid short-convolution + grouped-query-attention backbone (LFM2 architecture).
•128K context, multilingual (en, ar, zh, fr, de, ja, ko).
•GGUF and MLX builds recommended for llama.cpp and Apple Silicon.

LFM2.5 8B A1B 的最佳选择

MacBook Pro M2 Pro 16GBA

16 GB

运行此模型

LFM2.5 8B A1B on RTX 3080 10GB LFM2.5 8B A1B on Intel Arc B570 10GB LFM2.5 8B A1B on GTX 1080 Ti 11GB

量化选项

各量化级别的 VRAM 估算

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	3.3 GB	Low	—
Q3_K_S	3	4.2 GB	Low	—
NVFP4	4	4.8 GB	Medium	—
Q4_K_M	4	5.2 GB	Medium	—
Q5_K_M	5	6.1 GB	High	—
Q6_K	6	7.0 GB	High	—
Q8_0	8	9.1 GB	Very High	—
F16	16	17.4 GB	Maximum	—

硬件兼容性

全部硬件的适配估算

打开计算器

Computing compatibility...

内存详细分析

Reference: RTX 2060 6GB

Weights5.2 GB

KV Cache0.2 GB

Runtime0.9 GB

Headroom0.6 GB

常见问题

FAQ — LFM2.5 8B A1B

另请参阅

量化指南评分方法打开计算器