Spaces:
Running
Running
|
Download README.md from Yo-Purple/meditor-webgpu: direct link, hf CLI and curl.
- Browser
- Download file 2.71 kB
-
https://huggingface.co/spaces/Yo-Purple/meditor-webgpu/resolve/main/README.md
- Command line
-
hf download hf://spaces/Yo-Purple/meditor-webgpu/README.md
-
curl -L -o README.md https://huggingface.co/spaces/Yo-Purple/meditor-webgpu/resolve/main/README.md
2.71 kB
| title: Meditor WebGPU Chat | |
| emoji: π§ | |
| colorFrom: green | |
| colorTo: indigo | |
| sdk: static | |
| pinned: false | |
| license: apache-2.0 | |
| short_description: Run Meditor MoE & Dense LLMs 100% in-browser on WebGPU | |
| # π§ Meditor WebGPU β Client-Side In-Browser AI Chat | |
| An interactive, high-performance web chat space running **Meditor models** directly in your browser using **WebGPU hardware acceleration**. | |
| - β‘ **Zero Cloud GPU Costs:** The neural network weights are downloaded into your browser and executed on your local GPU via WGSL shaders. | |
| - π **100% Private:** Your prompts and generated text never leave your computer. | |
| - πΎ **Persistent OPFS Caching:** Model weights are cached locally via the browser's Origin Private File System (OPFS); once downloaded, the model boots instantly on subsequent visits without re-downloading. | |
| - ποΈ **Full Telemetry:** Live tokens per second (`tok/s`), time-to-first-token (`TTFT`), token counts, and GPU device details. | |
| --- | |
| ## π Available Meditor Models | |
| | Model | Parameters | Quantization | Size | Description | | |
| | :--- | :--- | :--- | :--- | :--- | | |
| | **Meditor-1 Universal** | 500M Dense | `Q4_K_M` (4-bit) | ~468 MB | β‘ **Recommended**: Instant download, 60β150 tok/s on WebGPU. Great for mobile & laptops. | | |
| | **Meditor-1 Universal** | 500M Dense | `Q8_0` (8-bit) | ~644 MB | High-fidelity 8-bit quantization. | | |
| | **Meditor-1 Universal** | 500M Dense | `BF16` (16-bit) | ~1.18 GB | Uncompressed reference weights. | | |
| | **Meditor-2 Universal** | 3.4B MoE (1.8B Active) | `Q3_K_M` (3-bit) | ~1.60 GB | 22-expert MoE architecture, fast multi-turn reasoning. | | |
| | **Meditor-2 Universal** | 3.4B MoE (1.8B Active) | `Q4_K_M` (4-bit) | ~2.00 GB | Balanced 4-bit MoE with dialogue, reasoning, and coding expertise. | | |
| | **Meditor-2 Universal** | 3.4B MoE (1.8B Active) | `Q5_K_M` (5-bit) | ~2.33 GB | High-fidelity 5-bit MoE. | | |
| | **Custom HF Model** | Any GGUF | Any | Variable | Load any custom repository and GGUF file from Hugging Face Hub. | | |
| --- | |
| ## π οΈ Browser Requirements | |
| - **Google Chrome / Chromium / Edge / Brave** (v113+) β Native WebGPU enabled by default. | |
| - **Safari** (v17+) β WebGPU supported. | |
| - **Firefox** (v115+) β WebGPU can be enabled via `about:config` (`dom.webgpu.enabled = true`). | |
| --- | |
| ## π¦ Published Repositories by @Yo-Purple | |
| - [Yo-Purple/Meditor-1-500M-Universal](https://huggingface.co/Yo-Purple/Meditor-1-500M-Universal) | |
| - [Yo-Purple/Meditor-2-3.4B-Universal](https://huggingface.co/Yo-Purple/Meditor-2-3.4B-Universal) | |
| - [Yo-Purple/Meditor-1-2.2B-A500M-MoE](https://huggingface.co/Yo-Purple/Meditor-1-2.2B-A500M-MoE) | |
| - [Yo-Purple/Meditor-2-3.4B-A1.8B-HMoE](https://huggingface.co/Yo-Purple/Meditor-2-3.4B-A1.8B-HMoE) | |