LiquidAI/LFM2.5-24B-A2B
main boardmodel
LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device. Mixture-of-experts, 24B total, 2B active (64 experts, 4 per token). Context: 128K tokens (config). License: lfm1.0.
- Org
- LiquidAI
- Released
- 2026-10-12
- Status
- main board
- License
- lfm1.0 (restricted)
- Parameters
- 24.0B total, 2.0B active
- Architecture
- MoE · 64 experts, 4/token · Lfm2MoeForCausalLM
- Context
- 131,072 tokens
- Task
- text-generation · text
- Signal
- 579 HF likes · 20480 downloads · 19 X likes · 3 X post(s)
Runs on
real quant files; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~16.2 GB: 24GB GPU (3090 / 4090), 24GB GPU (7900 XTX), Mac 24GB, Mac 36GB, CPU only, 32GB RAM (slow)
- Q8 ~27.2 GB: 32GB GPU (5090), 48GB GPU (RTX 6000 Ada / A6000), Arc B70 32GB, Mac 48GB, Mac 64GB (M1 Max)
- FP16 ~49.7 GB: 96GB GPU (RTX Pro 6000 Blackwell), 1× DGX Spark (128GB), 2× DGX Spark (256GB), Strix Halo 128GB unified (~96GB usable), Mac 96GB, Mac 128GB, Mac 192GB, Mac 256GB, Mac 512GB, CPU only, 64GB RAM (slow), CPU only, 128GB RAM (slow), CPU only, 256GB RAM (slow)
Links
- Hugging Face
- Ollama
- quant: LiquidAI/LFM2.5-24B-A2B-FP8
- GGUF: LiquidAI/LFM2.5-24B-A2B-GGUF
- MLX: LiquidAI/LFM2.5-24B-A2B-MLX-4bit
- MLX: LiquidAI/LFM2.5-24B-A2B-MLX-5bit
- MLX: LiquidAI/LFM2.5-24B-A2B-MLX-6bit
- MLX: LiquidAI/LFM2.5-24B-A2B-MLX-8bit
- MLX: LiquidAI/LFM2.5-24B-A2B-MLX-bf16
- quant: LiquidAI/LFM2.5-24B-A2B-ONNX
- GGUF: bartowski/LFM2.5-24B-A2B-GGUF
- GGUF: ggml-org/LFM2.5-24B-A2B-GGUF
- MLX: mlx-community/LFM2.5-24B-A2B-OptiQ-4bit
- Unsloth: unsloth/LFM2.5-24B-A2B
First published here 2026-10-12 · updated 2026-10-12 · markdown · JSON