Instructions to use wish418/carbix-v5 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use wish418/carbix-v5 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf wish418/carbix-v5
# Run inference directly in the terminal:
llama cli -hf wish418/carbix-v5

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf wish418/carbix-v5
# Run inference directly in the terminal:
llama cli -hf wish418/carbix-v5

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 wish418/carbix-v5
# Run inference directly in the terminal:
./llama-cli -hf wish418/carbix-v5

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 wish418/carbix-v5
# Run inference directly in the terminal:
./build/bin/llama-cli -hf wish418/carbix-v5

Use Docker

docker model run hf.co/wish418/carbix-v5

LM Studio
Jan
Ollama
How to use wish418/carbix-v5 with Ollama:
```
ollama run hf.co/wish418/carbix-v5
```
Unsloth Desktop

How to use wish418/carbix-v5 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf wish418/carbix-v5

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "wish418/carbix-v5"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use wish418/carbix-v5 with Docker Model Runner:
```
docker model run hf.co/wish418/carbix-v5
```

Lemonade

How to use wish418/carbix-v5 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull wish418/carbix-v5

Run and chat with the model

lemonade run user.carbix-v5-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use wish418/carbix-v5 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf wish418/carbix-v5

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default wish418/carbix-v5

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use wish418/carbix-v5 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf wish418/carbix-v5

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "wish418/carbix-v5" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

carbix-v5

한국 탄소배출 증빙 문서(전기요금 청구서 · 도시가스 청구서 · 화물 운송증)에서 구조화 JSON을 추출하는 Qwen3-VL-4B LoRA 파인튜닝 모델.

구성

파일	크기	설명
`carbix-v5-q8.gguf`	4.28GB	llama.cpp 서빙용 가중치 (q8_0)
`carbix-v5-mmproj.gguf`	836MB	비전 프로젝터 (f16)
`lora_adapter_v5/`	132MB	PEFT 어댑터 (merge 전)

v4 대비 변경

운송 필드 8키 → 25키. 문서에 인쇄된 항목을 전부 추출한다 (vat_amount, vehicle_tonnage, cargo_weight, item_name, ordering_company_*, payment_type, tax_invoice_requested, note, driver_*, receiver_*, received_date, receiver_signature 등)
운송 학습 데이터 916 → 4,509건 (폰트당 8.7 → 43건)
프롬프트 v13

학습

베이스     Qwen/Qwen3-VL-4B-Instruct
LoRA       r=16, alpha=32, dropout=0.05, 비전 레이어 포함
최적화     AdamW, LR 2e-4, cosine, warmup 0.05
배치       1 × grad_accum 4 (effective 4)
스텝       4,025 (1 epoch), 약 10.2시간 (L4 24GB)
데이터     16,100건 — 가스 10,696 · 운송 4,509 · 전기 895
손실       train 0.0095 / eval 0.0025

정확도 (테스트 657건, llama.cpp q8_0)

카테고리	정확도
전기	100.0%
가스	99.87%
운송	98.0%
전체	99.42%

건당 평균 4.8초.

필드별 (운송)

driver_phone 76.0% 가 최저 — 손글씨 폰트의 숫자 글리프가 모호해 1~~2자리를 오독한다(4가 Y처럼 렌더링). vehicle_number 92.5%, ordering_company_contact 93.0% 도 좁은 칸에 인쇄되는 항목이다. 나머지 필드는 96~~100%.

서빙

llama-server -m carbix-v5-q8.gguf --mmproj carbix-v5-mmproj.gguf \
    --host 127.0.0.1 --port 8001 -ngl 99 -c 8192 \
    --parallel 2 --image-min-tokens 1024 --jinja

max_tokens 는 768 이상을 권장한다 — 운송 25키 출력이 경유지 3개일 때 461 토큰까지 간다.

Downloads last month: 133

GGUF

Model size

4B params

Architecture

qwen3vl

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for wish418/carbix-v5

Base model

Qwen/Qwen3-VL-4B-Instruct

Adapter

(158)

this model