| --- |
| title: phenome playground |
| emoji: 🧬 |
| colorFrom: purple |
| colorTo: indigo |
| sdk: gradio |
| sdk_version: 6.15.1 |
| python_version: 3.12.12 |
| app_file: app.py |
| pinned: false |
| short_description: A genome is the system prompt — conditioned LLM personas |
| models: |
| - nyxia/phenome-qwen3.5-4b |
| --- |
| |
| # 🧬 phenome playground |
|
|
| **A genome is the system prompt.** |
|
|
| Instead of writing a 3,000-token persona prompt ("be warm but not too warm, use lists, |
| never say 'as an AI'…"), phenome encodes the whole persona as a compact **~126-character |
| symbolic genome** — 21 trait loci, each with two alleles and a 0–3 intensity dial — and a |
| LoRA learned to read that string directly. No written system prompt. Same model, different |
| genes → different persona. |
|
|
| Pick a curated preset (Method Actor, Bracket-Bard, Stage Director, Goblin, Roaster, |
| Contrarian, Terminal, Emoji Concierge, Storyteller, Sensory Drifter) or hit **🎲 Reroll** to |
| roll fresh genes. Open the *genome* accordion to see the exact alpha6 string the model is |
| reading and its decoded traits. |
|
|
| - **Model:** [`nyxia/phenome-qwen3.5-4b`](https://huggingface.co/nyxia/phenome-qwen3.5-4b) — Qwen3.5-4B + phenome LoRA, merged |
| - **The idea (blog):** [Should we use genetics instead of system prompts for AI Agents & Personas?](https://huggingface.co/blog/nyxia/genetics-instead-of-system-prompts-for-ai-agents) |
|
|
| Watch the *verbosity*, *humor*, *warmth*, *structure* and *pushback* axes move as you switch |
| personas on the same question. That shift is the genome doing the work. |
|
|
| > Personalities here are bold (sassy, blunt, chaotic) but safe-for-work — no explicit content. |
|
|
