Liquid (LFM2)
3 MLX aliases · LiquidAI LFM2 / LFM2.5 · on-device / edge models · native MLX · lfm tool-call parser.
LiquidAI (the MIT spinout) builds LFM2 / LFM2.5 — a family of on-device, edge-oriented models designed to run natively on-device. rapid-mlx ships three MLX aliases spanning the range: a 24B-A2B MoE, an 8B-A1B MoE, and a 1.2B dense model, all wired to the lfm tool-call parser. These are the low-RAM end of the catalog: the 1.2B fits comfortably on a laptop, and the sparse-MoE variants keep active-parameter cost small.
Not R2-mirrored. The LFM2 aliases are not on the models.rapidmlx.com RAM picker or the R2 catalog — they pull from Hugging Face on first use via their alias. Run rapid-mlx pull <alias> (or just serve) and the weights stream from Hugging Face.
- family
- Liquid
- aliases
- 3
- vendor
- LiquidAI (MIT spinout)
- profile
- on-device / edge · native MLX
- install
- rapid-mlx serve <alias>
- OpenAI base URL
- http://localhost:8000/v1
LFM2 / LFM2.5 · 3 aliases
On-device models from LiquidAI. All three use the lfm tool-call parser; none ship a reasoning parser.
tool parser: lfm
| alias | hf repo | tool parser | reasoning | flags |
|---|---|---|---|---|
| lfm2-24b-a2b-4bit | lmstudio-community/LFM2-24B-A2B-MLX-4bit | lfm | — | moe |
| lfm2.5-8b-a1b-4bit | mlx-community/LFM2.5-8B-A1B-MLX-4bit | lfm | — | moe |
| lfm2.5-1b-4bit | mlx-community/LFM2.5-1.2B-Instruct-4bit | lfm | — | dense |
Pull & serve
| rapid-mlx pull lfm2.5-1b-4bit |
| rapid-mlx serve lfm2.5-1b-4bit |
Notes & caveats
- On-device by design. LFM2 targets edge / laptop deployment — the 1.2B fits a small Mac; the A2B / A1B MoE variants keep active-parameter cost low.
- Not R2-mirrored. Weights pull from Hugging Face (the
lmstudio-community/mlx-communityrepos above), not the R2 mirror, so first run downloads from HF. - Tool calling flows through the
lfmparser; there is no reasoning parser for this family.