--- title: README emoji: 👑 colorFrom: blue colorTo: indigo sdk: static pinned: true ---
Ralph logo Ralph · Bittensor subnet 40

The same model.
Small enough for your phone.

Qwen3-8B compressed into four public crowns, from 1.16 GB to 4.61 GB. Same admitted architecture. Fewer bits. Every winner downloadable; every round auditable.

Join Device Lab #1 Choose a crown Download the model Audit Round 7 Build to dethrone
4 crowns1.16–4.61 GB
Open weightsApache-2.0
Public decisionssigned round records
## One crown. Four physical iPhones. Your receipt next. Our configured 2.94 GB Round 7 sub2 crown loaded and answered in PocketPal on four physical iPhones. [Watch the video and add your own result to Device Lab #1](https://huggingface.co/RalphLabsAI/ralph-crowns/discussions/3). Successes and failures are both useful; include your device, app version, and settings. These are Ralph team-authored BrowserStack tests using a test-only PocketPal build. The devices stayed connected and all showed **Memory tight**; this is not App Store memory-fit or offline certification. The exact sub2 builder is [boweizh1204](https://huggingface.co/boweizh1204). [Uncut recordings, exact settings, and limitations](https://huggingface.co/datasets/RalphLabsAI/ralph-device-lab/tree/7f5f1cfbf7882726a6c55a1d9b2f7c0ff06b3da8) · [Build a device receipt](https://huggingface.co/spaces/RalphLabsAI/Ralph-Pocket-Lab)