| --- |
| title: Bonsai κ-Stream — 1-bit 8B, streamed like a film |
| emoji: 🌳 |
| colorFrom: gray |
| colorTo: purple |
| sdk: static |
| pinned: false |
| license: apache-2.0 |
| short_description: Stream a 1-bit 8B LLM into your browser, verified |
| models: |
| - HOLOGRAMTECH/q-bonsai-8b |
| - prism-ml/Bonsai-8B-gguf |
| --- |
| |
| # Bonsai κ-Stream |
|
|
| **[Bonsai-8B](https://huggingface.co/prism-ml/Bonsai-8B-gguf) by Prism ML — 8 billion parameters |
| at 1.125 true bits/weight — streamed into your browser like a film and run on WebGPU. No server, |
| no account, no build step: one static page.** |
|
|
| What makes this different from downloading a model: |
|
|
| - **Content-addressed streaming.** The model is a [κ-object](https://huggingface.co/HOLOGRAMTECH/q-bonsai-8b): |
| 399 blocks raced over parallel connections, each verified against its content address |
| **before** it is decompressed or touches the GPU. A shard-grid shows the mind materializing. |
| - **Yours after one visit.** Blocks persist in browser storage — the second visit loads with |
| **zero network** and works offline. Any static mirror of the same bytes is an equal origin, |
| because the bytes, not the host, carry identity. |
| - **The weights are the published model.** The trained binary sign bits pass through byte-exact |
| (no re-quantization); the GEMV kernel dequantizes inside the matmul, so weights never expand |
| to dense floats in memory. ~35–45 tok/s on a consumer laptop GPU, thinking mode intact. |
| - **Part of a living system.** This same engine and ladder run [Q](https://hologram-technologies.github.io/hologram-os/), |
| a serverless browser OS where Bonsai serves as a second brain that takes over mid-conversation |
| from an instant 2B wafer. |
|
|
| Requires WebGPU (Chrome/Edge 113+). Created using **Bonsai by Prism ML** (Apache-2.0) on the |
| Hologram κ-substrate. |
|
|
