REFFT AI Store

LFM2.5-2.6B

LiquidAI LFM2.5-2.6B hybrid dense decoder for on-device text generation.

View READMEModel repository

LFM2.5-2.6B is LiquidAI's 30-layer hybrid decoder (22 double-gated short-conv mixers + 8 GQA attention layers, dense SwiGLU FFN). The hexagon package is a host VISA client (lfm2_5_2_6b) talking to refft-hexagon on the board. Modeling is built from refft.hexagon.ops.

Model performance

LFM2.5-2.6B on Hexagon HTP via refft-hexagon cli --model (single-file RFTB).

Snapdragon 8 Elite Gen 5 (SM8850) 1×HTP
SoC SM8850 · Android 16 (aarch64) · Hexagon v81 · w4 · runtime 0.6.5.dev672+gbc3f6c0ac · measured 2026-08-13
Measured · SM8850
31.8tok/s
TPS / Decode speed
127 ms
TTFT
150tok/s
Prefill speed

Install and run

Install
curl -fsSL https://raw.githubusercontent.com/refinefuture-ai/refft.cpp/main/refft-hexagon/install.sh | sh
Run
~/.local/share/refft-hexagon/bin/refft-hexagon cli --model ./LFM2.5-2.6B-w4.refft --backend hexagon --prompt "Who are you?" --max_new_tokens 128
Serving, environment variables and full instructions

Tags

llmgenerative-aiquantizedlfmlfm2lfm2.5