📱 POCKET — strong AI, on your own device

Run capable language and image models locally, on CPU, with no GPU and no cloud.

---

POCKET is a family of on-device models from VIDRAFT, quantized to run locally on a laptop CPU with no GPU and no network. The GGUF builds have been downloaded over 800,000 times. If you want private, offline, zero-cost inference, POCKET is built for it.

Weights live on Hugging Face. This repository is the index and the run recipes.

📦 Model index

Model	Type	Downloads	Hugging Face
POCKET-35B-GGUF	On-device LLM (CPU)	560K+	link
POCKET-26B-GGUF	On-device LLM (CPU)	250K+	link
POCKET-Darwin-180B-GGUF	Large on-device LLM	51K+	link
POCKET-EN-GGUF / POCKET-KR-GGUF	English / Korean	-	EN · KR
POCKET-KR-MLX	Apple Silicon (MLX)	-	link
POCKET-Zimage-CPU	On-device image generation	-	link

Live CPU demos: POCKET-26B-CPU · POCKET-35B-CPU · POCKET-Image-Studio

⚡ Run locally (CPU, no GPU)

With llama.cpp:

# download a GGUF from the model page, then:
llama-cli -hf FINAL-Bench/POCKET-26B-GGUF -p "Explain zero-token verification in one sentence."

With Ollama:

ollama run hf.co/FINAL-Bench/POCKET-26B-GGUF

On Apple Silicon, use the MLX build (POCKET-KR-MLX) with mlx-lm.

❓ FAQ

What is POCKET?

POCKET is a family of on-device AI models that run locally on a CPU, with no GPU and no cloud, using GGUF quantization. It covers language and image generation.

Can I run a large model on a laptop without a GPU?

Yes. POCKET ships CPU-friendly GGUF builds (26B, 35B, and more) that run with llama.cpp or Ollama on a normal laptop.

Is it private and offline?

Yes. Everything runs on your device, so no data leaves your machine and there is no inference cost.

Which POCKET model should I start with?

POCKET-26B-GGUF is a good balance of quality and speed on CPU; POCKET-35B-GGUF is stronger; POCKET-KR/EN target Korean/English.

📫 Connect

🌐 vidraft.net · 🤗 FINAL-Bench · 🧵 @vidraft_lab · 𝕏 @VIDRAFT_ai · ✉️ vidraft@mail.vidraft.net

📜 License

Apache-2.0 · Built by VIDRAFT · Beyond AI, Toward AGI.

_{on-device AI · local LLM · CPU inference · GGUF · llama.cpp · Ollama · offline AI · private AI · quantization · POCKET}