Instructions to use cmz1024/OLMo3-190M-zh with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cmz1024/OLMo3-190M-zh with Transformers:

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

pipe = pipeline("text-generation", model="cmz1024/OLMo3-190M-zh")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cmz1024/OLMo3-190M-zh")
model = AutoModelForCausalLM.from_pretrained("cmz1024/OLMo3-190M-zh")
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 cmz1024/OLMo3-190M-zh with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cmz1024/OLMo3-190M-zh"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cmz1024/OLMo3-190M-zh",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cmz1024/OLMo3-190M-zh

SGLang

How to use cmz1024/OLMo3-190M-zh 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 "cmz1024/OLMo3-190M-zh" \
    --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": "cmz1024/OLMo3-190M-zh",
		"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 "cmz1024/OLMo3-190M-zh" \
        --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": "cmz1024/OLMo3-190M-zh",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cmz1024/OLMo3-190M-zh with Docker Model Runner:
```
docker model run hf.co/cmz1024/OLMo3-190M-zh
```

OLMo3-190M-zh

File size: 2,005 Bytes

---
license: apache-2.0
language:
- zh
tags:
- pretrained
- olmo3
- chinese
library_name: transformers
pipeline_tag: text-generation
---

# OLMo3-190M-zh (v3, base)

**从零训练的 190M 中文 base 模型**，OLMo3 canonical 架构。活水 42ailab 出品，配套《零基础 AI 大模型研发训练营》L04 预训练讲。

## 用法

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("42ailab/OLMo3-190M-zh")
model = AutoModelForCausalLM.from_pretrained("42ailab/OLMo3-190M-zh")
```

## 训练

- 架构：OLMo3-190M（d=768, L=12, ffn=3072, QK-Norm, SWA=4096）
- Tokenizer：48k 中文 BPE（自训）
- 数据：Ultra-FineWeb-zh + Fineweb-Edu-Chinese V2.1（合计 3.4B tokens）
- 训练：Modal H100，14500 steps × 262K tokens/step，cosine 5e-4 → 5e-5
- Final mean train loss: 3.953

## 7-prompt 抽测（v3 基线）

| Prompt | 结果 |
|---|---|
| 人工智能是 | 🟢 流畅科普 |
| 山里有座庙 | 🟡 主题漂移 |
| 今天天气不错，我准备 | 🟡 漂移 |
| 北京大学位于 | 🔴 "江苏省"（错）|
| 四大发明是 | 🔴 "铜管/铁器/铜锤/铜镐"（全错）|
| 《红楼梦》人物 | 🟡 对错混杂 |
| Python 是一种 | 🔴 "开源库"（错）|

**合计 1 绿 / 3 黄 / 3 红** — 这是 base 的状态。改进版见 [`42ailab/OLMo3-190M-zh-v3.1`](https://huggingface.co/42ailab/OLMo3-190M-zh-v3.1)（continue pretrain 后"北大在北京"/"Python 是编程语言"等改善）。

## 关系

```
v3 (本 repo) ──continue pretrain──▶ v3.1 (42ailab/OLMo3-190M-zh-v3.1)
```

## License

- 权重 Apache-2.0
- 训练数据主要 Apache-2.0；不含 Wikipedia（v3.1 才引入 wiki）

## Citation

```bibtex
@misc{huoshui-olmo3-190m-zh,
  title={OLMo3-190M-zh: Chinese Pretrain Teaching Model},
  author={活水 AI 实验室 (42ailab) and 阳志平},
  year={2026},
  howpublished={\url{https://huggingface.co/42ailab/OLMo3-190M-zh}},
  note={LLM001 Course, Lecture 04}
}
```