Instructions to use betty0/Qwen2.5-3B-Instruct-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use betty0/Qwen2.5-3B-Instruct-GGUF 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 betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf betty0/Qwen2.5-3B-Instruct-GGUF: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 betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf betty0/Qwen2.5-3B-Instruct-GGUF: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 betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

Use Docker

docker model run hf.co/betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use betty0/Qwen2.5-3B-Instruct-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "betty0/Qwen2.5-3B-Instruct-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "betty0/Qwen2.5-3B-Instruct-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

Ollama
How to use betty0/Qwen2.5-3B-Instruct-GGUF with Ollama:
```
ollama run hf.co/betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
```
Unsloth Desktop

How to use betty0/Qwen2.5-3B-Instruct-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

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": "betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use betty0/Qwen2.5-3B-Instruct-GGUF with Docker Model Runner:
```
docker model run hf.co/betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M
```

Lemonade

How to use betty0/Qwen2.5-3B-Instruct-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Qwen2.5-3B-Instruct-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use betty0/Qwen2.5-3B-Instruct-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

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 betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use betty0/Qwen2.5-3B-Instruct-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

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 "betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M" \
  --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"

Qwen2.5-3B-Instruct — GGUF 量化版

Qwen/Qwen2.5-3B-Instruct 的 GGUF 量化版本, 由 llama.cpp(commit 4f37f51) 的 convert_hf_to_gguf.py + llama-quantize 產生。量化 pipeline 原始碼:gguf-quantization-factory。

GGUF quantized versions of Qwen/Qwen2.5-3B-Instruct, produced with llama.cpp. All benchmark numbers below were measured on a real NVIDIA GeForce RTX 4090 (all layers offloaded, -ngl 99).

量化等級比較 Quantization comparison

PPL:wikitext-2-raw-v1 test split,context = 512(越低越好 / lower is better)
速度 Speed:llama-bench pp512 / tg128,單位 tokens/s

量化 Quant	大小 Size	PPL ↓	ΔPPL vs F16	pp512 tok/s	tg128 tok/s	峰值 VRAM	檔案 File
F16	5.75 GiB	9.0631	基準 baseline	19999.59	127.83	8698 MiB	—(未上傳 not uploaded)
Q8_0	3.06 GiB	9.0806	+0.19%	22389.26	206.78	5888 MiB	`Qwen2.5-3B-Instruct-Q8_0.gguf`
Q5_K_M	2.07 GiB	9.1883	+1.38%	20976.91	270.39	4630 MiB	`Qwen2.5-3B-Instruct-Q5_K_M.gguf`
Q4_K_M	1.80 GiB	9.5741	+5.64%	19691.23	300.21	4348 MiB	`Qwen2.5-3B-Instruct-Q4_K_M.gguf`
Q4_K_M_IMAT	1.80 GiB	9.3241	+2.88%	21121.76	302.57	4299 MiB	`Qwen2.5-3B-Instruct-Q4_K_M_IMAT.gguf`
Q4_K_M_ZHTW	1.80 GiB	9.2671	+2.25%	20469.41	289.34	4295 MiB	`Qwen2.5-3B-Instruct-Q4_K_M_ZHTW.gguf`
IQ4_XS	1.62 GiB	9.3713	+3.40%	21769.35	303.62	4128 MiB	`Qwen2.5-3B-Instruct-IQ4_XS.gguf`
IQ4_XS_ZHTW	1.62 GiB	9.3731	+3.42%	21919.70	295.07	4120 MiB	`Qwen2.5-3B-Instruct-IQ4_XS_ZHTW.gguf`
IQ3_M	1.39 GiB	9.9827	+10.15%	22226.49	341.54	3620 MiB	`Qwen2.5-3B-Instruct-IQ3_M.gguf`
IQ3_S	1.36 GiB	9.9546	+9.84%	21605.76	338.30	3627 MiB	`Qwen2.5-3B-Instruct-IQ3_S.gguf`
IQ3_XS	1.30 GiB	10.0373	+10.75%	21557.88	350.32	3536 MiB	`Qwen2.5-3B-Instruct-IQ3_XS.gguf`
IQ3_XXS	1.19 GiB	10.5463	+16.36%	21914.09	354.49	3660 MiB	`Qwen2.5-3B-Instruct-IQ3_XXS.gguf`
IQ2_M	1.06 GiB	12.5399	+38.36%	19803.65	338.97	11007 MiB	—(未上傳 not uploaded)
IQ2_S	0.99 GiB	14.3376	+58.20%	18590.73	338.72	11122 MiB	—(未上傳 not uploaded)
IQ2_XS	0.96 GiB	15.5602	+71.69%	19556.77	344.39	11078 MiB	—(未上傳 not uploaded)
IQ2_XXS	0.88 GiB	23.9641	+164.41%	22406.26	365.61	10991 MiB	—(未上傳 not uploaded)

怎麼選 Which one should I pick?

Q4_K_M(推薦預設 / recommended default):檔案最小、速度最快,品質損失通常可接受。適合桌機/筆電日常使用、VRAM 有限的裝置。
Q5_K_M:比 Q4_K_M 更貼近原模型,只多一點大小。想要更好品質時選這個。
Q8_0:幾乎無損(ΔPPL 通常 <0.1%),但檔案接近 F16 的一半大小。對品質敏感的正式服務。
F16:未上傳(檔案過大);想要 F16 請用 pipeline 自行轉檔。

使用方式 Usage

Ollama

# 直接從 Hugging Face 跑(Ollama ≥ 0.3.34)
ollama run hf.co/betty0/Qwen2.5-3B-Instruct-GGUF:Q4_K_M

或自己寫 Modelfile(含 Qwen ChatML template 與 stop tokens),見 gguf-quantization-factory 的 ollama/Modelfile。

llama.cpp

# 下載
hf download betty0/Qwen2.5-3B-Instruct-GGUF Qwen2.5-3B-Instruct-Q4_K_M.gguf --local-dir .

# 互動對話(全部層放 GPU)
llama-cli -m Qwen2.5-3B-Instruct-Q4_K_M.gguf -ngl 99 -cnv

來源模型與授權 License

來源模型 Base model:Qwen/Qwen2.5-3B-Instruct
授權 License:Apache-2.0(依來源模型)

Downloads last month: 163

GGUF

Model size

3B params

Architecture

qwen2

Hardware compatibility

3-bit

4-bit

5-bit

8-bit

Model tree for betty0/Qwen2.5-3B-Instruct-GGUF

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-3B-Instruct

Quantized

(279)

this model