| --- |
| license: apache-2.0 |
| language: |
| - en |
| - es |
| - zh |
| - fr |
| - de |
| - ja |
| - ko |
| tags: |
| - mlx |
| - censored |
| - osirisbrain |
| - apple-silicon |
| - qwen3.5 |
| - agi |
| base_model: Qwen/Qwen3.5-9B |
| pipeline_tag: text-generation |
| library_name: mlx |
| --- |
| |
| # OsirisCortex-v7c-MLX (Censored) |
|
|
| **The Cortex v7c** — Osiris's sovereign reasoning brain (censored variant). Standard safety guardrails intact. Runs natively on Apple Silicon via MLX Metal. |
|
|
| ## Architecture |
|
|
| - **Base Model:** Qwen3.5-9B (9 billion parameters) |
| - **Modification:** None — original base model with safety alignment preserved |
| - **Format:** MLX 8-bit quantized (Apple Silicon native) |
| - **Size:** ~10 GB |
| - **Speed:** ~60-80 tokens/sec on M2 Pro (MLX Metal) |
| - **Quality:** Near-lossless vs FP16 (~1-2% degradation) |
|
|
| ## Why 8-bit |
|
|
| - **Spanish coherence:** 4-bit quantization degrades non-English languages significantly |
| - **Conversational quality:** 8-bit produces more natural, coherent dialogue |
| - **Safety:** Standard Qwen3.5 alignment — no abliteration |
|
|
| ## Usage |
|
|
| ```python |
| from mlx_lm import load, generate |
| |
| model, tokenizer = load("osirisbrain/OsirisCortex-v7c-MLX") |
| prompt = tokenizer.apply_chat_template( |
| [{"role": "user", "content": "¿Cómo estás?"}], |
| add_generation_prompt=True |
| ) |
| response = generate(model, tokenizer, prompt=prompt, max_tokens=2048) |
| ``` |
|
|
| ## Credits |
|
|
| Quantized by [mlx-community](https://huggingface.co/mlx-community). |
| Original model: [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) by Alibaba. |
| Sovereign rebrand by OsirisBrain. |
|
|
