LFM2.5 gets community imatrix GGUFs, with a 14.6 GB Q4 pick

More download choices for local agents

Community quantizer mradermacher has uploaded weighted, or imatrix, GGUF builds of LFM2.5-24B-A2B. The repository was created October 14, a day after Liquid AI’s underlying model release. This adds community deployment choices for people experimenting with the reasoning and tool-use model locally; it is not a newly trained LFM checkpoint or the model’s first GGUF availability.

The English quantization card lists Q4_K_M at approximately 14.6 GB and recommends it as a fast option. It lists Q4_K_S at 13.8 GB and also documents smaller Q2 and IQ3 choices. Those recommendations belong to the quantizer. The reviewed file snapshot contains actual GGUF uploads and an imatrix file, rather than only a list of planned downloads.

Choose a file, then budget for the runtime

For local-model users, the concrete decision is which quantization fits their memory and quality requirements. The card’s file sizes describe downloads, not complete runtime memory: context cache and inference buffers require additional space. A smaller quantization also does not establish equal answer quality, so the author’s recommendations should be treated as starting points for testing a particular workload.

The underlying LFM2.5-24B-A2B is a text-only mixture-of-experts reasoning model. Liquid AI positions it for agents and tool calling, and already documents its own GGUF distribution. The new community files broaden that selection without proving a speed or accuracy gain over the official builds. The model retains its custom LFM1.0 licensing; conversion to GGUF does not turn it into a standard permissively licensed model.