qwen3-4b-maid

Qwen3-4B-Maid

English | 中文

---

中文

模型简介

Qwen3-4B-Maid 是基于 Qwen3-4B-Instruct-2507 进行 LoRA 微调的对话模型，使用自制的女仆风格对话数据集训练，使模型在保留原有语言能力的同时，具备温柔、礼貌、带有女仆语气的对话风格。

模型信息

训练数据

使用自制的中文女仆风格对话数据集，共 1,717 条，涵盖日常聊天、角色扮演等场景，旨在让模型学习温柔、礼貌、带有「ご主人様」语气的表达方式。

使用方式

推荐使用 transformers 加载（需配合对应的 LoRA adapter）：

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
model = PeftModel.from_pretrained(base_model, "your-username/Qwen3-4B-Maid")

效果说明

• ✅ 模型已学会女仆语气与说话风格
• ⚠️ 回复有时偏冗长，建议在推理时适当限制 max_new_tokens
• 本模型为个人实验性项目，效果仅供参考

局限性与注意事项

• 训练步数较少（100 steps），模型对风格的学习尚不完全稳定
• 不建议用于需要精确事实输出的场景
• 本模型不具备安全过滤能力，请使用者自行负责使用场景

开源协议

本模型遵循基座模型 Qwen3 的 Apache 2.0 License。

---

English

Model Description

Qwen3-4B-Maid is a LoRA fine-tuned conversational model based on Qwen3-4B-Instruct-2507. It was trained on a custom maid-style dialogue dataset to give the model a gentle, polite, and maid-like conversational tone, while retaining the original model's language capabilities.

Model Details

Training Data

A custom Chinese maid-style dialogue dataset with 1,717 entries, covering daily conversation and roleplay scenarios. The goal is to teach the model a gentle, polite tone reminiscent of a maid character.

Usage

Recommended to load via transformers with the LoRA adapter:

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
model = PeftModel.from_pretrained(base_model, "your-username/Qwen3-4B-Maid")

Performance Notes

• ✅ The model has successfully learned the maid speech style and tone
• ⚠️ Responses can be verbose; consider limiting max_new_tokens during inference
• This is a personal experimental project; results are for reference only

Limitations

• Limited training (100 steps); style consistency may vary
• Not recommended for tasks requiring precise factual output
• No safety filtering; users are responsible for appropriate use

License

This model follows the Apache 2.0 License of the base Qwen3 model.