Instructions to use imwithye/atlas with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use imwithye/atlas with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="imwithye/atlas",
	filename="gemma3/gemma3-1b-it-q4_k_m.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use imwithye/atlas with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf imwithye/atlas:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf imwithye/atlas:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf imwithye/atlas:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf imwithye/atlas:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf imwithye/atlas:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf imwithye/atlas:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf imwithye/atlas:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf imwithye/atlas:Q4_K_M

Use Docker

docker model run hf.co/imwithye/atlas:Q4_K_M

LM Studio
Jan
Ollama
How to use imwithye/atlas with Ollama:
```
ollama run hf.co/imwithye/atlas:Q4_K_M
```

Unsloth Studio

How to use imwithye/atlas with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for imwithye/atlas to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for imwithye/atlas to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for imwithye/atlas to start chatting

Atomic Chat new
Docker Model Runner
How to use imwithye/atlas with Docker Model Runner:
```
docker model run hf.co/imwithye/atlas:Q4_K_M
```

Lemonade

How to use imwithye/atlas with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull imwithye/atlas:Q4_K_M

Run and chat with the model

lemonade run user.atlas-Q4_K_M

List all available models

lemonade list

Atlas — On-Device Models

Curated GGUF weights bundled with the Atlas iOS app (https://github.com/imwithye/atlas). Each file is a community Q4_K_M quantization re-hosted here so the app can pin a stable URL per model and so users don't depend on third-party uploader availability.

All models are picked to fit comfortably on a modern iPhone (≤ 2 GB on disk, ≤ ~4 GB RAM at inference).

Layout

gemma3/     llama3.2/   qwen2.5/    qwen3/      smollm2/

One folder per model family. Files are named <family>-<size>[-it]-<quant>.gguf.

Models

Gemma 3 — Google (Gemma Terms of Use)

File	Params	Size	Notes
`gemma3/gemma3-1b-it-q4_k_m.gguf`	1B	~0.8 GB	Smallest curated entry; good for quick replies on older devices.

Source quant: ggml-org/gemma-3-1b-it-GGUF Base model: google/gemma-3-1b-it

Llama 3.2 — Meta (Llama 3.2 Community License)

File	Params	Size	Notes
`llama3.2/llama3.2-1b-it-q4_k_m.gguf`	1B	~0.8 GB	Lightweight instruct; strong English.
`llama3.2/llama3.2-3b-it-q4_k_m.gguf`	3B	~2.0 GB	Best general-purpose pick at this size tier.

Source quants: bartowski/Llama-3.2-1B-Instruct-GGUF · bartowski/Llama-3.2-3B-Instruct-GGUF Base models: meta-llama/Llama-3.2-1B-Instruct · meta-llama/Llama-3.2-3B-Instruct

Qwen 2.5 — Alibaba (Apache-2.0)

File	Params	Size	Notes
`qwen2.5/qwen2.5-1.5b-it-q4_k_m.gguf`	1.5B	~1.0 GB	Solid instruct baseline; broad multilingual coverage.

Source quant: bartowski/Qwen2.5-1.5B-Instruct-GGUF Base model: Qwen/Qwen2.5-1.5B-Instruct

Qwen 3 — Alibaba (Apache-2.0)

File	Params	Size	Notes
`qwen3/qwen3-1.7b-q4_k_m.gguf`	1.7B	~1.1 GB	Hybrid reasoning model; supports `/think` and `/no_think` modes.

Source quant: bartowski/Qwen_Qwen3-1.7B-GGUF Base model: Qwen/Qwen3-1.7B

SmolLM2 — Hugging Face (Apache-2.0)

File	Params	Size	Notes
`smollm2/smollm2-1.7b-it-q4_k_m.gguf`	1.7B	~1.0 GB	Compact, fast; trained for on-device use.

Source quant: HuggingFaceTB/SmolLM2-1.7B-Instruct-GGUF Base model: HuggingFaceTB/SmolLM2-1.7B-Instruct

Quantization

All weights are Q4_K_M — 4-bit K-quants with mixed precision for select tensors. A good size/quality tradeoff for mobile inference. Run with llama.cpp or any compatible runtime.

Licensing

Each file inherits the license of its base model. Check the linked base model page before redistribution. Atlas does not re-license the weights.

Downloads last month: 47

GGUF

Model size

1.0B params

Architecture

gemma3

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support