Instructions to use Hengzongshu/Kos_Mos_project with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Hengzongshu/Kos_Mos_project with Transformers:

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Hengzongshu/Kos_Mos_project")
model = AutoModelForCausalLM.from_pretrained("Hengzongshu/Kos_Mos_project")
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 Hengzongshu/Kos_Mos_project with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Hengzongshu/Kos_Mos_project

SGLang

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

Docker Model Runner
How to use Hengzongshu/Kos_Mos_project with Docker Model Runner:
```
docker model run hf.co/Hengzongshu/Kos_Mos_project
```

🤖 Model Card — Kos-Mos v0.1.0

Model Overview ｜模型介绍

Kos-Mos v0.1.0 is an early research model released as part of the Kos-Mos Project. The model is trained to exhibit a stable persona and role-consistent behavior, with an emphasis on identity coherence. Development is currently underway to build agent systems centered around this model, exploring its use as a core cognitive component. All training and data construction code is open-sourced and available at:https://github.com/xiaziye/Kos_Mos_project

Kos-Mos v0.1.0 是 Kos-Mos 项目 的早期研究模型，主要关注角色一致性与人格稳定性。目前正在推进以该模型为核心的智能体（Agent）构建工作，探索其作为认知核心的可行性。所有训练与数据构建代码均已开源，仓库地址：https://github.com/xiaziye/Kos_Mos_project

Long-Term Objective ｜长期目标

The long-term objective of the Kos-Mos Project is to explore structured language as an internal substrate for reasoning, memory, and learning in large language model agents, enabling language to function as a cognitive structure rather than solely an interface.

Kos-Mos 项目的长期目标是探索 结构化语言 作为大语言模型智能体中 推理、记忆与学习 的内部认知载体，使语言不再仅仅是交互接口。

Current Implementation (v0.1.0) ｜当前实现

The current version focuses on persona alignment and stability under low-prompt conditions. Training is based on DPO alignment and persona-focused SFT, inspired by role-driven alignment methods (https://arxiv.org/abs/2511.01689v1), aiming to reduce reliance on long system prompts while maintaining expressive consistency.

当前版本主要实现了 低提示词条件下的人格对齐与稳定性，通过 DPO 对齐训练 与 人格 / 语气微调（SFT） 完成，训练思路参考角色驱动对齐方法，以减少对冗长系统提示词的依赖。

Downloads last month: 1

Safetensors

Model size

8B params

Tensor type

F16

Model tree for Hengzongshu/Kos_Mos_project

Base model

deepseek-ai/DeepSeek-R1-Distill-Qwen-7B

Finetuned

(307)

this model

Dataset used to train Hengzongshu/Kos_Mos_project

Paper for Hengzongshu/Kos_Mos_project

Open Character Training: Shaping the Persona of AI Assistants through Constitutional AI

Paper • 2511.01689 • Published Nov 3, 2025 • 5