🍫 CHOCO-BRAIN-0.8B-v0.1
“CHOCO-BRAIN-0.8B: Solopreneur Decision Engine & JEV Judgment System
🌐 English Documentation | 🇰🇷 한국어 매뉴얼”
This repository contains CHOCO-BRAIN-0.8B-v0.1, an ultra-lightweight on-device decision intelligence model and operational control pipeline for solopreneur business automation. Powered by the JEV (Judgment-Evaluation-Verification) framework and fine-tuned on Qwen3.5-0.8B via LoRA, it acts as an executive prefrontal cortex—processing complex business situations and outputting deterministic, multi-agent actionable decisions in under 0.05 seconds locally.
🧠 Why a Brain, Not a Chatbot?
Most LLMs are language engines — they write, summarize, translate. CHOCO-BRAIN is a decision engine inspired by the human prefrontal cortex: the part of the brain that sets priorities, filters risk, and commits to action.
Feed it any solopreneur work request — marketing strategy, content production, customer support, payment dispute — and it answers five structured questions, then outputs a single operational action key. No hallucinated essays. No verbose explanations. Just a decision.
📦 Inside the Box
| Detail | |
|---|---|
| Model | CHOCO-BRAIN-0.8B-v0.1 |
| Base | Qwen/Qwen3.5-0.8B |
| Method | LoRA Supervised Fine-Tuning (Rank 8, α 16) |
| Persona | Chief Strategic Decision Officer for Solopreneurs |
| Data | 30 real-world solopreneur scenarios → 150 structured decision tasks |
| Splits | Train 18 req (90) · Val 6 req (30) · Test 6 req (30) |
| Formats | GGUF (Q8_0) · Ollama ZIP · LM Studio compatible |
| License | Apache 2.0 |
🔀 5-Axis Decision Pipeline
Every request passes through five deterministic axes — no free-form generation, no hallucination:
| Axis | Question | Candidates |
|---|---|---|
domain | What kind of work is this? | marketing · content · support · unknown |
brain | Which specialist brain handles it? | marketing_ai · content_ai · support_ai · none |
knowledge | Which internal rulebook to reference? | marketing_docs · content_docs · support_docs · none |
agent | Who executes? | strategy_writer · script_writer · reply_drafter · ask_user · human_review |
next_step | What happens after? | finish · content_handoff · ask_user · human_review |
🚦 4 Action Keys
The final output is always one of four standardized keys — designed for multi-agent routing and automation:
| Key | When | Example |
|---|---|---|
NOW | Critical emergency | Payment token leak, server breach |
NORMAL | Routine execution | Strategy draft, video script, FAQ reply |
PAYMENT_CHECK | Financial audit needed | Wire transfer verification, refund request |
REVIEW | Human CEO approval required | Legal dispute, ambiguous context, contract signing |
🚀 Get Running in 60 Seconds
Option 1: Ollama (Recommended)
# Download CHOCO-BRAIN.zip from this repository, then:
unzip CHOCO-BRAIN.zip
cd CHOCO-BRAIN
ollama create choco-brain -f Modelfile
ollama run choco-brain
Option 2: LM Studio
- Download
CHOCO-BRAIN-0.8B.gguffrom this repository. - Place the file in your LM Studio models directory.
- Load the model and start chatting — responses come as structured JSON decisions.
Option 3: Python (transformers + peft)
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = "Qwen/Qwen3.5-0.8B"
adapter = "hwihwalab/CHOCO-BRAIN-0.8B-v0.1"
tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(
base, torch_dtype=torch.float16, device_map="auto"
)
model = PeftModel.from_pretrained(model, adapter)
model.eval()
# Input a solopreneur business request → structured decision JSON
Output Schema
{
"brain_name": "CHOCO-BRAIN",
"request": "Plan a crowdfunding marketing strategy for my AI handbook. Don't write the sales copy.",
"action_key": "NORMAL",
"decisions": {
"domain": "marketing",
"brain": "marketing_ai",
"knowledge": "marketing_docs",
"agent": "strategy_writer",
"next_step": "finish"
}
}
⚠️ Honest Limits
- Not a chatbot. CHOCO-BRAIN outputs structured decisions, not conversational text. It will not write your marketing copy or draft your emails.
- Small data, focused scope. Trained on 30 curated scenarios. Edge cases outside solopreneur workflows may produce unexpected classifications.
- Not production-critical. Do not use as an autonomous authority for financial, legal, or safety-critical decisions without human review.
- Korean-primary. Optimized for Korean business contexts. English input works but is not the primary training language.
- v0.1 — research release. This is an early exploration, not a production-grade system. Accuracy benchmarks and expanded training are planned for future versions.
📄 License
This project and model weights are licensed under the Apache 2.0 License — see the repository LICENSE file for full details.
Base model weights retain their original license (Qwen License).
📖 모델 개요
CHOCO-BRAIN-0.8B-v0.1은 1인 기업가를 위한 초경량 의사결정 AI 모델입니다. 수다스러운 챗봇이 아니라, 인간 전전두엽(뇌의 CEO)을 모사하여 업무 요청을 0.05초 만에 분석하고 표준 행동 키(NOW, NORMAL, PAYMENT_CHECK, REVIEW)를 판정합니다.
🚀 실행 방법
- Ollama:
CHOCO-BRAIN.zip다운로드 →ollama create choco-brain -f Modelfile→ollama run choco-brain - LM Studio:
CHOCO-BRAIN-0.8B.gguf파일을 LM Studio 모델 폴더에 넣고 로드 - Python:
transformers+peft라이브러리로 추론
📜 라이선스
Apache 2.0 라이선스 하에 배포됩니다.
Built and maintained by HWIHWA LAB — Turning local AI into real business tools, one model at a time.