🐾 dama-kurong-solopreneur-ko (Kurong)

🔔 Latest Update (v1.1)

• 🛑 EOS Token Fix: Completely resolved the endless generation / repetition loop by reinforcing the <end_of_turn> stop token.
• 🌐 English Persona Expansion: Added dedicated English identity dataset so Kurong stays in authentic fairy companion character during English conversations.
• (Existing users: Please re-download the latest GGUF for the best experience!)

dama-kurong-solopreneur-ko is a lightweight, on-device AI companion and assistant model specifically tailored for solopreneurs, indie hackers, and freelancers.

It serves as the cognitive engine for Kurong (쿠롱이), a supportive desktop companion fairy. Fine-tuned from Google's gemma-2-2b-it using Unsloth LoRA, the model incorporates over 50+ practical bilingual playbooks (40+ Korean, 10+ English) covering customer support, outage response, MVP pruning, and deep-work routine resets delivered with a warm, caring persona.

🔍 Linguistic & Persona Capabilities

• Primary Language (Korean): Deeply optimized for natural, polite Korean conversational tone and localized solopreneur business advice.
• Secondary Language (English): Capable of general comprehension and basic responses inheriting Gemma-2 pre-trained knowledge with Kurong persona support.
• Format: GGUF (Q4_K_M) optimized for ultra-fast on-device local execution using LM Studio, Ollama, and llama.cpp.

🎮 Try with Dama-AI Web Companion (Interactive Playground)

Experience Kurong on an interactive desktop Tamagotchi screen! Launch your local inference server (LM Studio or Ollama) and connect seamlessly via the web client:

✨ Key Features

• 💼 Solopreneur Business Playbooks: Pre-embedded guidelines for first-paid-user CS, outage apology notices, MVP scope definition, and mental burnout recovery.
• 🧚 Dedicated 1st-Person Persona: Clear and consistent identity awareness as the fairy companion Kurong in both Korean and English.
• ⚡ On-Device & GGUF Optimized: 2B parameter lightweight footprint designed for fast, private local inference in LM Studio, Ollama, and llama.cpp (dama-kurong-solopreneur-ko.Q4_K_M.gguf).
• 🌐 Bilingual Comprehension: Primarily fine-tuned for high-quality natural Korean dialogue, while retaining Gemma-2's multilingual/English instruction-following capabilities.

💡 Prompt Template (Gemma-2 Format)

This model follows the standard Gemma-2 chat turn structure:

<start_of_turn>user
첫 유료 결제 고객 응대 팁 알려줘<end_of_turn>
<start_of_turn>model

💡 Pro-Tip for English Conversations: Since the model is primarily trained on Korean datasets, if it occasionally responds in Korean to an English question, simply append "Please reply in English" or "Answer in English" to your prompt for the best English output!
(💡 영어로 질문했을 때 한국어로 답변한다면, 질문 끝에 "Please reply in English"를 함께 덧붙여주시면 자연스러운 영어로 답변합니다.)

🚀 Quickstart (Inference Examples)

1. Python (transformers)

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "hwihwalab/dama-kurong-solopreneur-ko"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "user", "content": "첫 유료 결제 고객 응대 알려줘"}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=256,
    temperature=0.7,
    top_p=0.9,
    eos_token_id=tokenizer.convert_tokens_to_ids("<end_of_turn>")
)

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

2. LM Studio (Local GUI)

1. Open LM Studio and search for hwihwalab/dama-kurong-solopreneur-ko (or press Ctrl + Shift + M).
2. Download dama-kurong-solopreneur-ko.Q4_K_M.gguf.
3. Load the model in the Chat tab and start interacting locally.

3. Ollama (Modelfile)

FROM ./dama-kurong-solopreneur-ko.Q4_K_M.gguf

TEMPLATE """<start_of_turn>user
{{ .Prompt }}<end_of_turn>
<start_of_turn>model
"""

PARAMETER stop "<start_of_turn>"
PARAMETER stop "<end_of_turn>"

💬 Example Queries

📄 License & Usage Terms

This model is fine-tuned from Google Gemma-2-2B-it and is governed by the Gemma Terms of Use. Users and commercial adopters are free to use, modify, and distribute this model in accordance with Google's Prohibited Use Policy.

📖 모델 개요

dama-kurong-solopreneur-ko는 1인 창업가, 인디 해커, 프리랜서를 돕는 데스크톱 다마고치 비서 요정 쿠롱이의 온디바이스 AI 모델입니다. Google gemma-2-2b-it를 기반으로 Unsloth LoRA 파인튜닝을 거쳤으며, 50여 종의 1인 기업 실무 노하우(한/영 CS, 장애 대처, MVP 기능 정의, 멘탈 루틴 관리)를 내장하고 있습니다.

📜 라이선스 안내

본 모델은 Google의 Gemma 라이선스를 준수합니다. 자유로운 연구, 개발 및 개인/상업적 활용이 가능하나 Google의 Gemma 사용 규정(Prohibited Use Policy)을 준수해야 합니다.

Trained and fine-tuned for [Dama-AI] by Hwihwa Lab (@hwihwalab)