LFM3 1.7B (MNN)
Liquid AI's LFM3 1.7B as a 4-bit MNN bundle for chatting on your phone with TokForge.
Published by TokForge for on-device inference.
| What it is | Chat model |
| Format | MNN bundle, 4-bit HQQ weights (block 32) |
| Runs on | Android phones (MNN engine; TokForge picks the CPU or GPU route per phone) |
| Phone class | 4 GB RAM or more (TokForge catalog) |
| Download | 1.31 GB (7 files) |
| Base model | LiquidAI/LFM3-1.7B |
| License | LFM Open License v1.0 |
TokForge
- Website: https://tokforge.ai
- Discord: https://discord.gg/Acv3CBtfVm
- Google Play: https://play.google.com/store/apps/details?id=dev.tokforge
- iOS TestFlight: https://testflight.apple.com/join/jnufjzRr
Runs on-device in the TokForge app.
Use it in TokForge
- Install TokForge from Google Play. This model needs the LFM3 support in TokForge 1.3.9.7 or newer.
- On Android, open Models > Recommended > Chat and download LFM3 1.7B. TokForge lists it on phones with 4 GB RAM or more.
- Load it and start a chat. The model runs on the phone; once it is downloaded, chatting needs no connection.
Collections: TokForge Chat Models (MNN).
Files
| File | Size | What it is |
|---|---|---|
config.json |
527 B | MNN runtime settings |
llm.mnn |
1.12 MB | Model graph |
llm.mnn.json |
2.69 MB | Model graph as JSON (reference copy) |
llm.mnn.weight |
1.30 GB | Model weights |
llm_config.json |
13.1 KB | Model settings and chat template |
tokenizer.mtok |
5.39 MB | Tokenizer |
LICENSE |
10.6 KB | LFM Open License v1.0 (Liquid AI's text) |
About the model
Liquid AI's LFM3 1.7B, a compact hybrid model built for edge devices: most layers are short convolutions, so memory stays low on long chats. Good general chat and instruction following in a small download.
Original model by Liquid AI, converted to MNN 4-bit for mobile deployment. This is a modified (converted and quantized) version of the original model; it is distributed under the LFM Open License v1.0 (see
LICENSE).
| Architecture | LFM3 hybrid (13 short-convolution layers, 6 full-attention and 5 sliding-window attention layers, GQA 16Q/4KV) |
| Parameters | 1.7B (4-bit quantized) |
| Quantization | int4 weights, block size 32, HQQ, tied embedding and head in the weight file |
| Vocab | 128,000 tokens |
| Chat template | Liquid AI's own, in llm_config.json (thinking on by default) |
The bundle is text only: in TokForge, photos you attach are described by the app's built-in image helper and the model answers from that description.
License
LFM3 1.7B is released by Liquid AI under the LFM Open License v1.0; the full text is in LICENSE. In plain words: you may
use, change and share it, keeping the license with it and marking changed files; commercial use is licensed only while you (or your
company) have annual revenue under USD 10 million (qualified non-profits doing non-commercial or research work are not
held to that limit).
Credits
- LFM3 1.7B by Liquid AI (Hugging Face), licensed under the LFM Open License v1.0
- MNN by Alibaba (GitHub)