NanoChat β€” On-Device Model Mirror

This repository hosts the quantized LiteRT model files used by NanoChat β€” a free, open-source Android app that runs large language models fully on-device, with no internet connection required after the initial download and no API keys needed.

All models are mirrored here from their original sources so that NanoChat users on the Play Store can download them without needing a Hugging Face account.


πŸ“± About NanoChat

NanoChat brings powerful, private AI chat to your Android phone. Everything runs locally on your device β€” your conversations never leave your phone.

  • πŸ”’ Fully private β€” no data ever sent to any server
  • ⚑ Fast β€” GPU-accelerated inference via Google's LiteRT runtime
  • πŸ“¦ No account or API key required β€” models download directly in-app
  • πŸ–ΌοΈ Multimodal β€” select models support image and audio input
  • πŸ†“ Free and open source

β†’ View NanoChat on GitHub


πŸ—‚οΈ Repository Structure

Each model lives in its own folder:

nanochat-models/
β”œβ”€β”€ Gemma3-1B-IT/
β”‚   └── gemma3-1b-it-int4.litertlm
β”œβ”€β”€ Gemma-3n-E2B-it/
β”‚   └── gemma-3n-E2B-it-int4.litertlm
β”œβ”€β”€ Gemma-3n-E4B-it/
β”‚   └── gemma-3n-E4B-it-int4.litertlm
β”œβ”€β”€ MobileActions-270M/
β”‚   └── mobile_actions_q8_ekv1024.litertlm
β”œβ”€β”€ Phi-4-mini-instruct/
β”‚   └── Phi-4-mini-instruct_multi-prefill-seq_q8_ekv4096.litertlm
β”œβ”€β”€ Qwen2.5-1.5B-Instruct/
β”‚   └── Qwen2.5-1.5B-Instruct_multi-prefill-seq_q8_ekv4096.litertlm
β”œβ”€β”€ Qwen3.5-0.8B-Instruct/
β”‚   └── model_multimodal.litertlm
β”œβ”€β”€ Qwen3.5-2B-Instruct/
β”‚   └── model_multimodal.litertlm
β”œβ”€β”€ Qwen3.5-4B-Instruct/
β”‚   └── model_multimodal.litertlm
└── TinyGarden-270M/
    └── tiny_garden_q8_ekv1024.litertlm

πŸ€– Available Models

πŸ’¬ Chat Models

Model Size Min RAM Vision Audio
Gemma3-1B-IT 558 MB 6 GB ❌ ❌
Qwen2.5-1.5B-Instruct 1.49 GB 6 GB ❌ ❌
Qwen3.5-0.8B-Instruct 1.16 GB 4 GB βœ… ❌
Qwen3.5-2B-Instruct 2.78 GB 8 GB βœ… ❌
Gemma-3n-E2B-it 3.40 GB 8 GB βœ… βœ…
Phi-4-mini-instruct 3.81 GB 8 GB ❌ ❌
Qwen3.5-4B-Instruct 5.26 GB 8 GB βœ… ❌
Gemma-3n-E4B-it (experimental) 4.58 GB 12 GB βœ… βœ…

πŸ› οΈ Specialised Models

Model Size Purpose
TinyGarden-270M 276 MB Tiny Garden game
MobileActions-270M 276 MB On-device mobile actions

πŸ“₯ Original Sources


βš–οΈ Licenses

Model Family License
Gemma 3, Gemma 3n, TinyGarden, MobileActions Gemma Terms of Use
Qwen 2.5, Qwen 3.5 Apache 2.0
Phi-4-mini MIT

By downloading Gemma-family models through NanoChat you agree to the Gemma Terms of Use.


πŸ”„ Keeping This Mirror Updated

This mirror is synced from the NanoChat model_allowlist whenever a new version ships. Only enabled models are mirrored. Disabled models are not uploaded but existing files are never deleted, ensuring backwards compatibility with older app versions.


Built with ❀️ for the NanoChat community.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support