Instructions to use lianghsun/Marble-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use lianghsun/Marble-3B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="lianghsun/Marble-3B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("lianghsun/Marble-3B")
model = AutoModelForCausalLM.from_pretrained("lianghsun/Marble-3B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use lianghsun/Marble-3B with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/lianghsun/Marble-3B

SGLang

How to use lianghsun/Marble-3B 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 "lianghsun/Marble-3B" \
    --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": "lianghsun/Marble-3B",
		"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 "lianghsun/Marble-3B" \
        --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": "lianghsun/Marble-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use lianghsun/Marble-3B with Docker Model Runner:
```
docker model run hf.co/lianghsun/Marble-3B
```

Marble-3B / README.md

lianghsun

docs: align model card with template style

1894413 verified 5 months ago

preview code

Raw

History Blame Contribute Delete

2.73 kB

	---
	license: mit
	language:
	- zh
	- en
	base_model:
	- ibm-granite/granite-3.1-3b-a800m-base
	library_name: transformers
	pipeline_tag: text-generation
	tags:
	- Taiwan
	- ROC
	- zhtw
	- Marble
	- Granite
	- MoE
	- SLM
	- continued-pretraining
	---

	# Model Card for Marble-3B

	<!-- Provide a quick summary of what the model is/does. -->
	Marble-3B 是以 [ibm-granite/granite-3.1-3b-a800m-base](https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-base)（IBM Granite 3.1 3B-A800M MoE）為基底，針對繁體中文與中華民國台灣語境完成持續預訓練（CPT）之 MoE 基底模型，作為 [Marble-3B-Instruct](https://huggingface.co/lianghsun/Marble-3B-Instruct) 等下游模型的繁中底座。

	> ⚠️ 規格重點：
	> 本模型為 3B Mixture-of-Experts（MoE）基底模型、純文本單模態、僅做 CPT、未做指令微調，需自行 SFT 後才有對話能力。

	## Model Details

	IBM Granite 3.1 系列引入 MoE 架構（3B 總參數、800M 活化參數），在推論成本與能力之間提供有趣的折衷點。Marble-3B 把繁中與台灣語境語料注入 Granite 3.1 3B-A800M，使下游任務能在 MoE 架構下取得繁中底層能力，兼顧推論效率與多領域覆蓋。

	核心特點 (Key Features)
	1. MoE 架構繁中底座：少量活化參數（800M）卻有 3B 總參數的容量，部署效率較同等級稠密模型佳。
	2. 多領域適配：MoE 結構天然適合多領域知識分流，配合繁中 CPT 可作為法律、教育、生活等多領域微調的共同基底。
	3. 可下游微調：作為 Instruct、領域應用之 SFT／DPO 起點。

	### Model Description

	- Developed by: [Liang Hsun Huang](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US)
	- Funded by: [APMIC](https://www.apmic.ai/)
	- Base model: [ibm-granite/granite-3.1-3b-a800m-base](https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-base)
	- Model type: GraniteMoeForCausalLM (Transformers)
	- Language(s) (NLP): Traditional Chinese, English
	- License: MIT

	### Model Sources

	- Repository: [lianghsun/Marble-3B](https://huggingface.co/lianghsun/Marble-3B)

	## Citation

	```bibtex
	@misc{marble_3b,
	title = {Marble-3B: A Traditional Chinese Continued-Pretrained Granite 3B-A800M MoE Model for Taiwan},
	author = {Huang, Liang Hsun},
	year = {2025},
	howpublished = {\url{https://huggingface.co/lianghsun/Marble-3B}}
	}
	```

	## Acknowledge

	- 特此感謝 [APMIC](https://www.apmic.ai/) 的算力支援。

	## Model Card Authors

	[Huang Liang Hsun](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US)

	## Model Card Contact

	[Huang Liang Hsun](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US)