README.md

metadata

library_name: sentis
tags:
  - unity-sentis
  - onnx
  - smollm2
  - causal-lm
license: apache-2.0
language:
  - en
base_model: HuggingFaceTB/SmolLM2-135M-Instruct
pipeline_tag: text-generation

SmolLM2-135M-Instruct for Unity Sentis

This repository contains optimized versions of the SmolLM2-135M-Instruct model, specifically formatted for use in Unity Sentis.

SmolLM2-135M is an ultra-compact model that excels at on-device tasks like text rewriting, summarization, and simple NPC dialogue while maintaining an exceptionally small memory footprint.

Available Versions

File Name	Format	Precision	Size	Best For
model.onnx	ONNX	FP32	~530MB	High-fidelity source / Desktop
model_FP16.sentis	Sentis	FP16	~320MB	Balanced Performance / Mobile
model_Uint8.sentis	Sentis	Uint8	~160MB	Maximum Performance / WebGL / Low-end Mobile

How to use in Unity

1. Install Sentis: Use the Unity Package Manager to install com.unity.sentis (2.1.0+ recommended).
2. Download Model: Choose the .sentis file based on your target platform's memory constraints.
3. Import: Drag the .sentis file and the tokenizer.json into your Unity project.

Basic C# Loading Snippet

using Unity.Sentis;
using UnityEngine;

public class TinyLLM : MonoBehaviour {
    public ModelAsset modelAsset;
    private IWorker engine;

    void Start() {
        Model runtimeModel = ModelLoader.Load(modelAsset);
        // GPUCompute is recommended for mobile performance
        engine = WorkerFactory.CreateWorker(BackendType.GPUCompute, runtimeModel);
    }
}