Liquid AI
LFM 2.5 MoE IQ3_S 8B-A1B
A larger 3-bit MoE option for a better balance of size and quality. Experimental reasoning model; stores all 8B weights and activates about 1B per token.
About LFM 2.5 MoE IQ3_S 8B-A1B
A larger 3-bit MoE option for a better balance of size and quality. Experimental reasoning model; stores all 8B weights and activates about 1B per token.
Every model in this catalog runs locally in the Offline AI Chat app. After the download finishes, LFM 2.5 MoE IQ3_S answers on your phone with no connection, and your prompts and replies are never uploaded.
- Download size
- 3.6 GB (UD-IQ3_S GGUF)
- Memory needed
- 6 GB minimum · 8 GB recommended
- Context in app
- 4,096 tokens
- Runs on
- Android 8+, 64-bit · iOS in private testing
- Access
- Offline Plus subscription
- License
- LFM Open License v1.0
- Model source
- Liquid AI on Hugging Face