kshitijthakkar/lfm25-mobile-actions-cactus

LFM2.5 Mobile Actions — Cactus Format

Pre-converted weights for LiquidAI/LFM2.5-1.2B-MobileActions in Cactus format for on-device inference.

Model Details

Property	Value
Architecture	LFM2 (Liquid Foundation Model 2.5)
Parameters	~1.2B
Hidden Dim	2,048
Layers	16 (mixed conv + full attention)
Attention Heads	32 (8 KV heads)
FFN Dim	12,288
Context Length	128,000 tokens
Precision	FP16
Vocab Size	65,536
Layer Types	Conv (0,1,3,4) + Full Attention (2,5,7,9,11,13,15)

Format

These weights are in Cactus binary format (.weights files + config.txt), converted from HuggingFace safetensors using the Cactus Python conversion pipeline.

This is not a standard HuggingFace Transformers model. It is designed to be loaded directly by the Cactus C++ inference engine for mobile and edge deployment (iOS, Android, macOS, Linux ARM64).

Usage

Download the entire repo and point Cactus to the directory:

# Clone weights
git clone https://huggingface.co/kshitijthakkar/lfm25-mobile-actions-cactus

# Use with Cactus engine
cactus run ./lfm25-mobile-actions-cactus

Conversion

Converted using the Cactus Python converter from the original safetensors:

cactus convert LiquidAI/LFM2.5-1.2B-MobileActions ./output --precision FP16

License

See the original model card for license details.