v2.3 · 2026-08-09

LFM2.5 Encoder 350M

lfm2-5-encoder-350m

EMBEDDING

LFM2.5 Encoder 350M is a 0.3545B embedding model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.

PROVIDER

Liquid AI

FAMILY

LFM2.5 Encoder

MODEL TYPE

embedding

PARAMETERS

0.3545B

MODALITIES

text

ARCHITECTURE

Bidirectional LFM2 hybrid encoder with a masked-language-modeling head

CONTEXT WINDOW

8,192 tokens

TRAINING TOKENS

—

RELEASE DATE

—

LICENSE & LINKS

LFM Open License v1.0license text

MODEL CARD

huggingface.co/LiquidAI/LFM2.5-Encoder-350M

HUGGING FACE

LiquidAI/LFM2.5-Encoder-350M

SETUP HINTS

LLAMA.CPP

The official card uses custom Transformers code; do not assume llama.cpp support.

RAM / VRAM ESTIMATES

MIN RAM

4 GB

COMFORTABLE RAM

8 GB

These are conservative local-inference estimates, not official hardware requirements.

HARDWARE FIT

CPU only (no GPU)

yes

8 GB RAM

comfortable

16 GB RAM

comfortable

32 GB RAM

comfortable

8 GB VRAM

not required

12 GB VRAM

not required

24 GB VRAM

not required

Apple Silicon (unified memory)

comfortable

Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.

BEST FOR

·Fine-tuning higher-quality compact classifiers, routers, token taggers, retrieval systems, and similarity models.

AVOID IF

·You want a ready-made chat assistant or the smallest encoder in this family.

CAVEATS

·This is a masked-language-model encoder, not a generative chat model.

·The official card requires trust_remote_code=True; review remote code before use.

·Vendor evaluation results have not been reproduced by LocalLLMGuide.com.

·The custom license includes a commercial-use revenue threshold.

SOURCE URLS

[source 1] [source 2]

FIELD EVIDENCE