Text Generation
Transformers
Safetensors
GGUF
Korean
gemma4
image-text-to-text
gemma
korean
roleplay
mud
lore
llama.cpp
lmstudio
conversational
Instructions to use sangwon1472/gemma4-e2b-mud with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use sangwon1472/gemma4-e2b-mud with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="sangwon1472/gemma4-e2b-mud") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("sangwon1472/gemma4-e2b-mud") model = AutoModelForImageTextToText.from_pretrained("sangwon1472/gemma4-e2b-mud") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - llama-cpp-python
How to use sangwon1472/gemma4-e2b-mud with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="sangwon1472/gemma4-e2b-mud", filename="gemma4-e2b-mud-Q4_K_M.gguf", )
llm.create_chat_completion( messages = [ { "role": "user", "content": "What is the capital of France?" } ] ) - Notebooks
- Google Colab
- Kaggle
- Local Apps
- llama.cpp
How to use sangwon1472/gemma4-e2b-mud with llama.cpp:
Install from brew
brew install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M # Run inference directly in the terminal: llama-cli -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M # Run inference directly in the terminal: llama-cli -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
Use Docker
docker model run hf.co/sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
- LM Studio
- Jan
- vLLM
How to use sangwon1472/gemma4-e2b-mud with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "sangwon1472/gemma4-e2b-mud" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sangwon1472/gemma4-e2b-mud", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
- SGLang
How to use sangwon1472/gemma4-e2b-mud with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "sangwon1472/gemma4-e2b-mud" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sangwon1472/gemma4-e2b-mud", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "sangwon1472/gemma4-e2b-mud" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sangwon1472/gemma4-e2b-mud", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Ollama
How to use sangwon1472/gemma4-e2b-mud with Ollama:
ollama run hf.co/sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
- Unsloth Studio new
How to use sangwon1472/gemma4-e2b-mud with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for sangwon1472/gemma4-e2b-mud to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for sangwon1472/gemma4-e2b-mud to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for sangwon1472/gemma4-e2b-mud to start chatting
- Docker Model Runner
How to use sangwon1472/gemma4-e2b-mud with Docker Model Runner:
docker model run hf.co/sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
- Lemonade
How to use sangwon1472/gemma4-e2b-mud with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull sangwon1472/gemma4-e2b-mud:UD-Q4_K_M
Run and chat with the model
lemonade run user.gemma4-e2b-mud-UD-Q4_K_M
List all available models
lemonade list
File size: 12,246 Bytes
59fd928 49871e2 59fd928 49871e2 59fd928 49871e2 b7b9673 49871e2 b7b9673 49871e2 59fd928 49871e2 59fd928 49871e2 d9fceea 49871e2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 | ---
license: apache-2.0
language:
- ko
library_name: transformers
pipeline_tag: text-generation
base_model:
- google/gemma-4-E2B-it
tags:
- gemma
- gemma4
- korean
- roleplay
- mud
- lore
- gguf
- llama.cpp
- lmstudio
- transformers
---
# gemma4-e2b-mud
`gemma4-e2b-mud` λ **Gemma 4 E2B κ³μ΄μ κΈ°λ°μΌλ‘ ν νκ΅μ΄ μ°μ£Όνν ν
μ€νΈ MUD μ€νμΌ λͺ¨λΈ**μ΄λ€.
μ§§μ λͺ
λ Ή λ°μ, NPC λμ¬, λ°© λΆμκΈ° λ¬μ¬, μ μΉ μ€λͺ
, νλ‘ ννΈ κ°μ **κ²μ μμ¬ λ μ΄μ΄**λ₯Ό λ€λ£¨κΈ° μ’κ² μ 리ν 체ν¬ν¬μΈνΈμ, μ΄λ₯Ό μ¬ννκ±°λ νμ₯ν μ μλ **Colab μ€νν° ν¨ν€μ§**, κ·Έλ¦¬κ³ **LM Studioμ© GGUF**λ₯Ό ν¨κ» μ 곡νλ€.
μ΄ μ μ₯μμ ν΅μ¬ λ°©ν₯μ βκ²μ μμ§μ νμ βμ΄ μλλΌ, **λ§ν¬, λΆμκΈ°, μ§§μ μΈκ³κ΄ μλ΅**μ AIκ° λ§‘λλ‘ λ§λλ κ²μ΄λ€.
## TL;DR
- κΈ°λ° λͺ¨λΈ: `google/gemma-4-E2B-it`
- μ£Ό μ¬μ© μΈμ΄: νκ΅μ΄
- μ£Ό μ¬μ© μμ: ν
μ€νΈ MUD, NPC λν, λ£Έ λ¬μ¬, μ μΉ/λ‘μ΄ μ€λͺ
- ν¬ν¨ νμ:
- Transformers 체ν¬ν¬μΈνΈ
- LM Studio / llama.cpp μ© `GGUF`
- Colab / Unsloth μ¬νμ© μ€νν° ν¨ν€μ§
- κ°μ₯ μ λ§λ μ
λ ₯:
- `talk oracle`
- `look`
- `rumor`
- `scan signal`
- `First Fire Horizonμ΄ μ΄λ€ κ³³μΈμ§ μ€λͺ
ν΄μ€.`
## μ΄ μ μ₯μμ λ€μ΄ μλ κ²
### 1. Transformers 체ν¬ν¬μΈνΈ
루νΈμλ λ³ν©λ Hugging Face νμ λͺ¨λΈ κ°μ€μΉκ° λ€μ΄ μλ€.
- `model-00001-of-00005.safetensors`
- `model-00002-of-00005.safetensors`
- `model-00003-of-00005.safetensors`
- `model-00004-of-00005.safetensors`
- `model-00005-of-00005.safetensors`
- `config.json`
- `processor_config.json`
- `tokenizer.json`
- `tokenizer_config.json`
- `chat_template.jinja`
### 2. GGUF
LM Studio / llama.cpp μ λ°λ‘ κ°μ Έκ° μ μλλ‘ λ€μ GGUF νμΌμ ν¨κ» μ 곡νλ€.
- `gemma4-e2b-mud-Q4_K_M.gguf`
μ΄ νμΌμ λ‘컬 μΆλ‘ , LM Studio ν
μ€νΈ, λΉ λ₯Έ λ°°ν¬μ μ ν©ν μμμ μ΄λ€.
### 3. Companion starter package
μ μ₯μ μμ [`gemma-mud-colab-starter/`](./gemma-mud-colab-starter) ν΄λμλ λ€μμ΄ ν¬ν¨λμ΄ μλ€.
- Colab λ
ΈνΈλΆ
- E2B / E4B μ€νμ© λ
ΈνΈλΆ λ³ν
- μμ λ°μ΄ν°μ
- LM Studio μμ€ν
ν둬ννΈ
- ν
μ€νΈ ν둬ννΈ
- μ€ν μλ΄μ νΈλ¬λΈμν
λ¬Έμ
μ¦ μ΄ μ μ₯μλ βλͺ¨λΈλ§ λμ Έ λμ ννβκ° μλλΌ, **μ€νκ³Ό λ°°ν¬κΉμ§ μ΄μ΄μ§λ ν¨ν€μ§ν μ μ₯μ**λ₯Ό λͺ©νλ‘ νλ€.
λ°μ΄ν°μ
ꡬ쑰μ μ μ λ°©λ²μ λ³λ λ¬Έμμ μ λ¦¬ν΄ λμλ€.
- [`DATASET_GUIDE.md`](./DATASET_GUIDE.md)
## λͺ¨λΈ κ°μ
μ΄ λͺ¨λΈμ **Gemma 4 E2B instruct κ³μ΄μ ν
μ€νΈ MUD ν€μ λ§κ² λ€λ¬μ νμ λͺ¨λΈ**μ΄λ€.
μ§ν₯νλ μλ΅μ μλμ κ°λ€.
- μ§§κ³ μ λͺ
ν NPC λμ¬
- λΆμκΈ°κ° μ΄μ μλ λ°©/μ₯λ©΄ μ€λͺ
- μΈκ³κ΄ κ³ μ λͺ
μ¬λ₯Ό μ€μ¬μΌλ‘ ν μ μΉ/λ‘μ΄ μλ΄
- νλ μ΄λ₯Ό λλ μ§§μ ννΈ
- λ²μ© λΉμν λ΅λ³λ³΄λ€, **μΈκ³ μμ 머무λ₯΄λ λ°μ**
λ°λλ‘ μλμ κ°μ λ°©ν₯μ νΌνλλ‘ μ€κ³νλ€.
- νμ€ κΈ°μ
/DB/μμ μ€λͺ
μΌλ‘ μλ λ΅λ³
- `AIλ‘μ...` κ°μ λ©ν λ°μΈ
- μ₯ν©ν λ°±κ³Όμ¬μ μ μ€λͺ
- κ²μ μνλ₯Ό λͺ¨λΈμ΄ μμλ‘ νμ νλ λ°ν
## κΈ°λ° λͺ¨λΈκ³Ό ꡬ쑰
μ΄ μ μ₯μμ κΈ°λ°μ [`google/gemma-4-E2B-it`](https://huggingface.co/google/gemma-4-E2B-it) μ΄λ€.
Gemma 4 E2Bλ Googleμ μμ Gemma 4 κ³μ΄ λͺ¨λΈμ΄λ©°, 곡μ μΉ΄λ κΈ°μ€μΌλ‘ **Apache 2.0** λΌμ΄μ μ€λ₯Ό μ¬μ©νλ€.
μ£Όμ:
- **μ΄ μ μ₯μ 루νΈμ μ¬λΌμ¨ μ€μ κ°μ€μΉμ GGUFλ E2B κΈ°λ°μ΄λ€.**
- starter package μμ λ€μ΄ μλ `E4B` λ
ΈνΈλΆμ **E4B μ€ν/μ¬νμ΅μ© companion notebook** μ΄λ©°, λ£¨νΈ μ²΄ν¬ν¬μΈνΈκ° E4BλΌλ λ»μ μλλ€.
μ΄ μ μ₯μμ λ‘컬 `config.json` κΈ°μ€ μ£Όμ ν
μ€νΈ μ€μ μ λ€μκ³Ό κ°λ€.
- μν€ν
μ²: `Gemma4ForConditionalGeneration`
- `model_type`: `gemma4`
- ν
μ€νΈ hidden size: `1536`
- ν
μ€νΈ λ μ΄μ΄ μ: `35`
- attention heads: `8`
- key/value heads: `1`
- intermediate size: `6144`
- vocab size: `262144`
- sliding window: `512`
μ€μ:
- **μ΄ νμ λͺ¨λΈμ μ€μ μ¬μ© μ΄μ μ ν
μ€νΈ μμ±**μ΄λ€.
- Gemma 4 E2B κ³μ΄ μ체λ μμ λ©ν°λͺ¨λ¬ κ³μ΄μ΄μ§λ§, μ΄ μ μ₯μμ μ€νν° ν¨ν€μ§μ μμ ν둬ννΈλ **ν
μ€νΈ MUD μ¬μ© μλ리μ€**μ λ§μΆ°μ Έ μλ€.
- ν¨κ» λ€μ΄ μλ GGUF νμΌ μμ LM Studioμμ **ν
μ€νΈ λν μ€μ¬**μΌλ‘ λ°λ‘ ν
μ€νΈνλ μ©λμ λ§λ€.
## μ΄λ€ μμ
μ μ λ§λκ°
### μ λ§λ μμ
- ν
μ€νΈ MUD NPC λν
- λ£Έ/μ₯λ©΄ λΆμκΈ° λ¬μ¬
- μ§§μ rumor / lore / signal μλ΅
- λ‘컬 μ€ν 리ν
λ§ ν
μ€νΈ
- νκ΅μ΄ μΈκ³κ΄ νλ‘ν νμ΄ν
- MUD μμ§μ βμμ¬ λ μ΄μ΄β λΆλ¦¬ μ€ν
### νΉν μ λ§λ μ
λ ₯ μ€νμΌ
- νλ λ¨μ΄ λͺ
λ Ήν
- `talk oracle`
- `look`
- `rumor`
- `scan signal`
- μ§§μ μ€λͺ
μμ²ν
- `First Fire Horizonμ΄ μ΄λ€ κ³³μΈμ§ μ€λͺ
ν΄μ€.`
- `Helios Vergeμ μλ―Έμ μ€μμ±μ μ€λͺ
ν΄μ€.`
## μ΄λ€ μμ
μλ λ§μ§ μλκ°
μ΄ λͺ¨λΈμ μλ μν μ **μ§μ λ΄λΉνμ§ μλ νΈμ΄ μμ νλ€.**
- νμ€νΈ μλ£ νμ
- 보μ μ§κΈ
- μμ΄ν
/골λ μ μ°
- λ¬Έ μ΄λ¦Ό/μ κΈ λ± μν λ³κ²½ νμ
- μ ν¬ μΉν¨ κ³μ°
- λ²λ₯ , μλ£, κΈμ΅ κ°μ κ³ μ λ’° μ 보 μ 곡
μ¦, μΆμ² μ¬μ© ꡬ쑰λ λ€μκ³Ό κ°λ€.
- **μμ§**: μ΄λ, μ ν¬, μν, 보μ, νμ€νΈ λ‘μ§
- **λͺ¨λΈ**: λμ¬, λΆμκΈ°, μ μΉ, μ§§μ ννΈ, λ¬μ¬
## λΉ λ₯Έ μμ
### Transformers λ‘ λ°λ‘ μ°κΈ°
Gemma 4 곡μ λ¬Έμμ μ¬μ© λ°©μμ λ§μΆ° `AutoProcessor` + `AutoModelForImageTextToText` κ²½λ‘λ‘ μμν μ μλ€.
```python
from transformers import AutoProcessor, AutoModelForImageTextToText
import torch
MODEL_ID = "sangwon1472/gemma4-e2b-mud"
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = AutoModelForImageTextToText.from_pretrained(
MODEL_ID,
dtype="auto",
device_map="auto",
)
messages = [
{
"role": "system",
"content": "λΉμ μ μ°μ£Όνν ν
μ€νΈ MUDμ νλ‘ μλ΄μμ΄μ μΈκ³κ΄ ν΄μ€μλ€. λ΅λ³μ νκ΅μ΄λ‘ νλ€."
},
{
"role": "user",
"content": "talk oracle"
},
]
text = processor.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False,
)
inputs = processor(text=text, return_tensors="pt").to(model.device)
input_len = inputs["input_ids"].shape[-1]
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=128,
temperature=1.0,
top_p=0.95,
top_k=64,
)
response = processor.decode(outputs[0][input_len:], skip_special_tokens=False)
print(processor.parse_response(response))
```
## LM Studio / llama.cpp μ¬μ©
μ΄ μ μ₯μμλ λ°λ‘ import κ°λ₯ν GGUF νμΌμ΄ ν¬ν¨λμ΄ μλ€.
- `gemma4-e2b-mud-Q4_K_M.gguf`
LM Studio μμλ μ΄ GGUFλ₯Ό κ°μ Έμμ λ°λ‘ ν
μ€νΈν μ μλ€.
ν¨κ» μ 곡νλ μμ€ν
ν둬ννΈμ ν
μ€νΈ λ¬Έμ₯μ starter package μμ μλ€.
- [`lmstudio_system_prompt_ko.txt`](./gemma-mud-colab-starter/examples/lmstudio_system_prompt_ko.txt)
- [`lmstudio_system_prompt_npc_ko.txt`](./gemma-mud-colab-starter/examples/lmstudio_system_prompt_npc_ko.txt)
- [`lmstudio_system_prompt_lore_ko.txt`](./gemma-mud-colab-starter/examples/lmstudio_system_prompt_lore_ko.txt)
- [`lmstudio_test_prompts.md`](./gemma-mud-colab-starter/examples/lmstudio_test_prompts.md)
### κΆμ₯ μμ ν둬ννΈ
- `talk oracle`
- `look`
- `rumor`
- `scan signal`
- `First Fire Horizonμ΄ μ΄λ€ κ³³μΈμ§ μ€λͺ
ν΄μ€.`
### κΆμ₯ μμ± μ€μ
Gemma 4 κ³μ΄μ 곡μ/μ€μ μ¬μ©μμ μλ κ°μ΄ 무λν μΆλ°μ μ΄λ€.
- `temperature = 1.0`
- `top_p = 0.95`
- `top_k = 64`
λ³΄λ€ μ§§κ³ μμ μ μΈ NPC λ΅λ³μ΄ νμνλ©΄ μλμ²λΌ μ‘°κΈ λ³΄μμ μΌλ‘ μμν΄λ μ’λ€.
- `temperature = 0.7`
- `max tokens = 96 ~ 128`
μμ βλ μ§§κ²β μ€μ μ μ΄ μ μ₯μμ MUD μ©λμ λ§μΆ μ€μ κΆμ₯κ°μ΄λ€.
## Companion starter package μλ΄
ν¨κ» μ 곡νλ [`gemma-mud-colab-starter/`](./gemma-mud-colab-starter) ν΄λλ λͺ¨λΈμ λ€μ νμ΅νκ±°λ λ³νν΄ λ³΄κ³ μΆμ μ¬μ©μλ₯Ό μν ꡬμ±μ΄λ€.
μ£Όμ νμΌ:
- [`README.md`](./gemma-mud-colab-starter/README.md)
- [`run_instructions.md`](./gemma-mud-colab-starter/run_instructions.md)
- [`troubleshooting.md`](./gemma-mud-colab-starter/troubleshooting.md)
- [`Gemma4_MUD_QLoRA_Colab_Notebook-E2B.ipynb`](./gemma-mud-colab-starter/notebooks/Gemma4_MUD_QLoRA_Colab_Notebook-E2B.ipynb)
- [`Gemma4_MUD_QLoRA_Colab_Notebook-E4B.ipynb`](./gemma-mud-colab-starter/notebooks/Gemma4_MUD_QLoRA_Colab_Notebook-E4B.ipynb)
ν¬ν¨λ μμ λ°μ΄ν°μ
:
- `combined_1000.jsonl` : 1000ν μλ³Έ
- `combined_1000.unsloth_chatml_dedup.jsonl` : 963ν μ 리본
- `combined_1000.unsloth_gemma4_messages_dedup.jsonl` : 963ν Gemma 4 messages μ 리본
- `gemma4_mud_alpaca_100.jsonl` : 100ν μν μμ
λ°μ΄ν°μ
μ ꡬ쑰, λΆν¬, μ μ λ°©μ, μ μν μμ± κ·μΉμ λ³λ λ¬Έμμμ μμΈν λ€λ£¬λ€.
- [`DATASET_GUIDE.md`](./DATASET_GUIDE.md)
μ€μ:
- μ λ°μ΄ν°μ
μ **starter/retraining/example μ©λ**λ‘ ν¨κ» λ£μ΄ λ κ²μ΄λ€.
- μ΄ μ μ₯μμ 체ν¬ν¬μΈνΈλ₯Ό μ΄ν΄νκ±°λ νμ₯νλ λ° λμμ μ£Όλ μλ£μ΄λ©°, μ¬ν μ€νμ μν μΆλ°μ μΌλ‘ 보λ νΈμ΄ μ’λ€.
## μλ΅ νμ§μ λ³Ό λ μ’μ κΈ°μ€
μ λ΅ λ¬Έμ₯μ μΈμ λ΄λμ§κ° ν΅μ¬μ μλλ€.
λ μ€μν κ²μ **μλ΅μ κ²°**μ΄λ€.
μ’μ μ νΈ:
- νκ΅μ΄λ‘ λ΅νλ€
- μΈκ³κ΄ μμ λ¨Έλ¬Έλ€
- μ§§κ³ λΆμκΈ°κ° μλ€
- λͺ
λ Ήμ΄λ³ λ°μ κ²°μ΄ λ€λ₯΄λ€
- νμ€ μΌλ° μμ κ°μλ‘ μμ§ μλλ€
κ²½κ³ μ νΈ:
- `Hello!` λ‘ μμνλ λ²μ© μ±λ΄ ν€
- `Oracle Database` κ°μ νμ€ μμ λ°μ
- `AIλ‘μ` κ°μ λ©ν λ°μΈ
- μμ§μ΄ ν΄μΌ ν μν νμ μ λͺ¨λΈμ΄ μ§μ μ μΈ
## νκ³μ μ£Όμμ¬ν
μ΄ λͺ¨λΈμ νΉμ μ€νμΌκ³Ό μ©λμ λ§μΆ **λλ©μΈ μ μν λͺ¨λΈ**μ΄λ€. λ°λΌμ μλ νκ³λ₯Ό μ΄ν΄νκ³ μ°λ κ²μ΄ μ’λ€.
- μΈκ³κ΄ ν€μ κ°νλμ§λ§, μμ λλ©μΈ μ μ λͺ¨λΈ νΉμ±μ **κ³ μ λͺ
μ¬λ₯Ό κ³Όμ₯λκ² μ΄μ΄λΆμ΄κ±°λ λΆμκΈ°λ₯Ό κ³Όμμ±**ν μ μλ€.
- ν
μ€νΈ MUD μμ¬μλ μ λ§μ§λ§, μΌλ° λΉμν μ§μμλ΅μ μ νλλ₯Ό λͺ©νλ‘ νμ§λ μμλ€.
- μ₯ν©ν μ€λͺ
μ μ€μ΄κΈ° μν΄ ν둬ννΈμ μμ€ν
λ©μμ§λ₯Ό μ§§κ³ λͺ
ννκ² μ μ§νλ νΈμ΄ μ’λ€.
- 곡μ λ²€μΉλ§ν¬ κΈ°λ° μ¬νκ°λ λ³λ μμ μ± νκ°λ₯Ό μ΄ μΉ΄λμμ μ 곡νμ§λ μλλ€.
## κΆμ₯ μ¬μ© ν¨ν΄
μ΄ λͺ¨λΈμ μλμ²λΌ λΆμ΄λ©΄ κ°μ₯ μμ°μ€λ½λ€.
1. νλ μ΄μ΄ μ
λ ₯μ μ§§μ ν
μ€νΈ ν둬ννΈλ‘ μ 리νλ€.
2. μμ§μ μν/νμ μ μ²λ¦¬νλ€.
3. λͺ¨λΈμ λμ¬, λ¬μ¬, ννΈ, μ μΉ μ€λͺ
μ μμ±νλ€.
4. μμ§μ λͺ¨λΈ μΆλ ₯μ κ·Έλλ‘ μ°κΈ°λ³΄λ€, νμνλ©΄ νμ²λ¦¬ κ·μΉμ λλ€.
νΉν μλ μ μ μ§ν€λ νΈμ΄ μ’λ€.
- **μμ§μ΄ κ²°μ ν΄μΌ ν κ²**: νμ€νΈ μ±κ³΅, 보μ, μ ν¬ νμ , λ¬Έ μν
- **λͺ¨λΈμ΄ ννν΄λ μ’μ κ²**: λΆμκΈ°, λͺ©μ리, μ¬μ΄, λ¨μ, λ¬Έν, μ ν
## λΌμ΄μ μ€μ μΆμ²
- νμ κΈ°λ° λͺ¨λΈ: [`google/gemma-4-E2B-it`](https://huggingface.co/google/gemma-4-E2B-it)
- λΌμ΄μ μ€ νκΈ°: `Apache-2.0`
- GGUF λ° starter package λ μ΄ μ μ₯μ λ°°ν¬ νΈμλ₯Ό μν΄ ν¨κ» ν¬ν¨νλ€.
Gemma 4 μ체μ ꡬ쑰μ κΈ°λ³Έ μ¬μ© λ°©μμ 곡μ Gemma 4 μΉ΄λμ Unsloth Gemma 4 λ¬Έμλ₯Ό μ°Έκ³ νλ©΄ μ’λ€.
- [Google Gemma 4 E2B model card](https://huggingface.co/google/gemma-4-E2B-it)
- [Unsloth Gemma 4 docs](https://unsloth.ai/docs/models/gemma-4)
## ν μ€ μμ½
`gemma4-e2b-mud` λ **νκ΅μ΄ μ°μ£Όνν ν
μ€νΈ MUDμ λΆμκΈ°, NPC λμ¬, μ μΉ μ€λͺ
μ λ‘컬μμλ λΉ λ₯΄κ² μνν μ μκ² λ§λ Gemma 4 E2B νμ λͺ¨λΈ + GGUF + Colab starter package λ¬Άμ**μ΄λ€.
|