---
title: Qwen-2.5-0.5b-instrunt-lora-J
canonical_url: "https://www.modelscope.cn/models/jacksong123/Qwen-2.5-0.5b-instrunt-lora-J"
md_url: "https://www.modelscope.cn/models/jacksong123/Qwen-2.5-0.5b-instrunt-lora-J.md"
repository: jacksong123/Qwen-2.5-0.5b-instrunt-lora-J
chinese_name: "轻语0.5B"
last_updated: 2026-07-25
license: "Apache License 2.0"
pipeline_tag: text-generation
tasks:
  - text-generation
  - text-classification
  - translation
  - text-summarization
  - nli
  - table-question-answering
  - relation-extraction
  - question-answering
  - faq-question-answering
  - chatbot
base_model:
  - Qwen/Qwen2.5-0.5B-Instruct
base_model_relation: adapter
parameters: 4.4M
tensor_type:
  - F32
library_name:
  - pytorch
  - transformer
  - lora
  - safetensors
frameworks:
  - PyTorch
downloads: 67
stars: 3
tags:
  - lora
  - transformers
  - qwen
  - chinese
  - chat
---

# Qwen-2.5-0.5b-instrunt-lora-J

> Qwen-2.5-0.5b-instrunt-lora-J - jacksong123 在 ModelScope 开源的模型。一个基于 Qwen2.5-0.5B-Instruct 的轻量级中文聊天适配器，面向日常轻松对话场景优化。说话风格自然口语化，像朋友聊天一样简洁直接，不端着、不说教。

jacksong123/Qwen-2.5-0.5b-instrunt-lora-J 是 ModelScope 魔搭社区上的 4.4M 参数text-generation、text-classification、translation模型，采用 Apache License 2.0 许可，基于 Qwen/Qwen2.5-0.5B-Instruct 构建。

- **Repository**: jacksong123/Qwen-2.5-0.5b-instrunt-lora-J
- **License**: Apache License 2.0
- **Tasks**: text-generation, text-classification, translation, text-summarization, nli, table-question-answering, relation-extraction, question-answering, faq-question-answering, chatbot
- **Parameters**: 4.4M
- **Base model**: Qwen/Qwen2.5-0.5B-Instruct
- **Tags**: lora, transformers, qwen, chinese, chat
- **Downloads**: 67
- **Stars**: 3
- **Last updated**: 2026-07-25

Source: https://www.modelscope.cn/models/jacksong123/Qwen-2.5-0.5b-instrunt-lora-J

---

# 轻语-0.5B

一个基于 Qwen2.5-0.5B-Instruct 的轻量级中文聊天适配器，面向日常轻松对话场景优化。说话风格自然口语化，像朋友聊天一样简洁直接，不端着、不说教。

## Model Details

### Model Description

- **Developed by:** 宋梓铭
- **Model type:** 因果语言模型（对话适配器）
- **Language(s):** 中文
- **License:** Apache License 2.0
- **Finetuned from model:** Qwen/Qwen2.5-0.5B-Instruct

## Uses

### Direct Use

适合直接用于日常闲聊、轻量对话场景：
- 日常聊天、打发时间
- 简单问答、快速响应
- 群聊搭子、轻松互动
- 低资源设备上的本地对话部署

### Out-of-Scope Use

- 专业领域知识问答（医疗、法律、金融等）
- 长篇内容生成、写作
- 需要严格事实准确性的场景
- 高风险决策辅助

## Bias, Risks, and Limitations

作为0.5B参数的小模型，本模型存在以下局限：
- 知识覆盖范围有限，可能产生事实错误或幻觉
- 长对话上下文理解能力有限
- 复杂逻辑推理能力较弱
- 可能存在训练数据中的偏见

使用时请保持批判性思维，重要内容请自行核实。

### Recommendations

- 适合轻松娱乐的日常对话，不适合严肃用途
- 建议在单轮或短多轮对话中使用
- 输出内容仅供参考，请自行判断真伪

## How to Get Started with the Model

```python
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
import torch

base = "Qwen/Qwen2.5-0.5B-Instruct"
lora = "qwen05b_affective_lora"

bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16)
model = AutoModelForCausalLM.from_pretrained(base, quantization_config=bnb, torch_dtype=torch.bfloat16, trust_remote_code=True)
model = PeftModel.from_pretrained(model, lora)
tokenizer = AutoTokenizer.from_pretrained(base, trust_remote_code=True)

messages = [{"role": "user", "content": "在吗"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to("cuda")
out = model.generate(**inputs, max_new_tokens=512, temperature=0.7, top_p=0.9, repetition_penalty=1.15)
print(tokenizer.decode(out[0][inputs.input_ids.size(1):], skip_special_tokens=True))
```

## Training Details

### Training Data

基于 BelleGroup/train_0.5M_CN 中文指令数据集，选取约2万条多样化对话样本，涵盖日常问候、兴趣爱好、生活闲聊、简单知识问答等场景。

### Training Procedure

- **Training regime:** bf16 mixed precision + 4bit QLoRA
- **LoRA rank:** 8
- **Target modules:** q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
- **Epochs:** 3
- **Learning rate:** 2e-4
- **Optimizer:** AdamW

### Framework versions

- PEFT 0.19.1+
- Transformers 4.x
- PyTorch 2.x
- BitsAndBytes 0.4x