Liquid AI24B (2.45B active)Q4_K_M 下约 15 GB 显存

LFM2.5-24B-A2B — 显存、速度与本地部署

作者： Jakub Rusinowski · 最后更新：

The largest and last LFM2.5 model: 24.0B parameters with 2.45B active per token, 40 layers of which only 10 use grouped-query attention (the other 30 are short convolutions that cache nothing), 64 experts with top-4 routing. A pure reasoning model that always thinks before answering, built for on-device agents and tool use. Text only, 131,072-token context, 16 languages including Polish and Chinese. Liquid ships official GGUF, MLX, ONNX and FP8 builds; the Q4_K_M file is 14.5 GB.

LFM2.5-24B-A2B 在 Q4_K_M 下约需 15 GB 显存——量化权重加框架开销，不含 KV 缓存。在 Apple Silicon 上，这部分来自统一内存。

按量化级别的显存与速度

计算基准：NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存，因此数值高于上方的主数字。

量化显存速度（估算）适配
Q2_K
2.63 bpw
8.8 GB~292 tok/s可运行
Q3_K_M
3.41 bpw
11.2 GB~267 tok/s可运行
Q4_K_M
4.83 bpw
15.5 GB~232 tok/s可运行
Q5_K_M
5.67 bpw
18 GB~215 tok/s可运行
Q6_K
6.56 bpw
20.7 GB~199 tok/s可运行
Q8_0
8.50 bpw
26.5 GB~26 tok/s需卸载
F16
16.00 bpw
48.9 GB~17 tok/s需卸载

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型，详见 方法说明页. LFM2.5-24B-A2B 显存计算器 →

运行 LFM2.5-24B-A2B

目录中能运行 LFM2.5-24B-A2B 的最便宜 GPU 是 AMD Radeon RX 7900 XT (20 GB).

购买此硬件 AMD Radeon RX 7900 XT 20GB — 20 GB VRAM · 315 W board power立即云端部署 RunPod 上的 RTX 4090 — 低至 $0.34/小时 · 价格核实于 2026-07

或在 Vast.ai 比较，低至 $0.35/小时 (typical low · varies)

作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买，LLM Configurator 可能会获得佣金，而你无需支付任何额外费用。作为亚马逊联盟成员（Amazon Associate），LLM Configurator 会从符合条件的购买中获得收益。
AMD Radeon RX 7900 XT 20GB
20 GB VRAM · 315 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 LFM2.5-24B-A2B

安装 Ollama，然后运行：

ollama run lfm2.5:24b
Hugging Face 上的权重： LiquidAI/LFM2.5-24B-A2B-GGUF ↗

规格

Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.

参数量
24.0 Billion (2.45B active)
上下文窗口
131,072
架构
Hybrid short convolution + GQA MoE (64 experts, top-4)
提供商
Liquid AI
许可证
LFM Open License v1.0
Availability
Weights downloadable
Practical deployment
Single consumer GPU
规格量化
Q4_K_M
系统内存
32 GB
记录更新于
2026-10-15
许可证LFM Open License v1.0允许商业使用

Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.

Commercial use is licensed only below $10M annual revenue; above that, a separate agreement.

限制与注意事项

  • Always reasons before answering; there is no non-thinking mode.
  • Text only.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量，并非我们实测。

最适合agentstool callingreasoningmultilingualon device

我的 GPU 能运行 LFM2.5-24B-A2B 吗？

LFM2.5 的其他尺寸

LFM2.5-24B-A2B — 常见问题

How much VRAM does LFM2.5-24B-A2B need?

About 15 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does LFM2.5-24B-A2B run on an RTX 4090 (24 GB)?

Yes. LFM2.5-24B-A2B needs about 15 GB at Q4_K_M, inside a 24 GB card, at an estimated 232 tokens/sec.

How do I run LFM2.5-24B-A2B locally?

Install Ollama and run `ollama run lfm2.5:24b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does LFM2.5 come in?

LFM2.5 2.6B (2 GB), LFM2.5-8B-A1B (6 GB), LFM2.5-24B-A2B (15 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.