---
license: other
tags:
- ane
- apple
- coreml
- neural-engine
- npu
- llm
base_model: LiquidAI/LFM2.5-2.6B
pipeline_tag: text-generation
---

# LFM2.5-2.6B — Apple Neural Engine artifacts

Prebuilt Apple Neural Engine artifacts for [`LiquidAI/LFM2.5-2.6B`](https://huggingface.co/LiquidAI/LFM2.5-2.6B).

The artifacts are portable Core ML packages; Core ML compiles them for the specific device on first load.

## Contents

Each directory is a complete, independently runnable bundle. Pick one.

| directory | precision | graphs | largest graph | intended target |
|---|---|---|---|---|
| `lut8_g32_c6/` | 8-bit weight-only, fp16 compute | 7 | 411 MB | **iPhone and Mac** |
| `lut8_g32/` | 8-bit weight-only, fp16 compute | 5 | 654 MB | Mac |
| root files | fp16 | 5 | 1240 MB | Mac |

The three differ in how the model is split across graphs, not in what it computes. `lut8_g32_c6`
is split more finely so that no single graph is large enough to be a problem on a phone; the
coarser splits are fine on a Mac and load slightly faster there.

Each 8-bit directory carries a `PRECISION.json` describing exactly how it was produced.

For model behaviour, capabilities, intended use, limitations, and license, see the upstream
[LiquidAI/LFM2.5-2.6B](https://huggingface.co/LiquidAI/LFM2.5-2.6B) model card. This repository adds no model
capabilities and changes no model behaviour.