| --- |
| license: apache-2.0 |
| base_model: GRRNMAKE/Magnus |
| tags: |
| - synaxim |
| - grrn-inference |
| - int4 |
| - symbiotic-gate |
| - framework-free |
| - magnus |
| - mistral |
| - text-generation |
| pipeline_tag: text-generation |
| --- |
| |
| # Magnus-SYMB — SYNAXIM Text Inference |
|
|
| > **⚡ v0.1.x — Text-only.** Optimized for fast text generation through SYNAXIM's O(1) Symbiotic State Engine. Multimodal (vision, audio) planned for v0.2.x. |
|
|
| ## Quick Start |
|
|
| ```bash |
| pip install git+https://github.com/GRRN-MAKER/SYNAXIM.git |
| huggingface-cli download GRRNMAKE/Magnus-SYMB --local-dir ./magnus-symb |
| ``` |
|
|
| ```python |
| from grrn_inference import GRRNModel |
| |
| model = GRRNModel.from_pretrained("./magnus-symb", device="cuda") |
| result = model.generate("Hello, I am Magnus.", max_tokens=50, temperature=0.7) |
| print(result.text) |
| ``` |
|
|
| ## Model Details |
|
|
| | Property | Value | |
| |----------|-------| |
| | **Name** | Magnus | |
| | **Source** | [GRRNMAKE/Magnus](https://huggingface.co/GRRNMAKE/Magnus) | |
| | **Base** | Mistral-7B-Instruct-v0.3 | |
| | **Modality** | **Text-only** (v0.1.x) | |
| | **Parameters** | ~7.24B | |
| | **Layers** | 32, D=4096, 32Q/8KV | |
| | **Format** | `.symb` INT4 (3.85 GB) | |
| | **Adapter** | 32-layer Symbiotic Gate (trained via knowledge distillation) | |
|
|
| ## Backends |
|
|
| | Backend | Device | Speed | |
| |---------|--------|-------| |
| | Triton GPU | `device="cuda"` | 🔥 Fastest | |
| | Numba CPU | `device="cpu"` | ⚡ Fast | |
| | NumPy | `device="cpu-numpy"` | Baseline | |
|
|
| --- |
|
|
| **Engine**: [GRRN-MAKER/SYNAXIM](https://github.com/GRRN-MAKER/SYNAXIM) | **Author**: [GRRNMAKER](https://github.com/GRRN-MAKER) |
|
|
