Qwen3-1.7B Chinese Customer Service Multi-Task Model
A multi-task fine-tuned Qwen3-1.7B model for Chinese customer service NLP, trained on 4 tasks simultaneously.
Supported Tasks
| Task | Description | Example |
|---|---|---|
| 客服 FAQ | 回答产品、订单、退款等常见问题 | "如何申请退款？" |
| 文档问答 | 根据 SOP/手册/合約回答问题 | 提供文档 + 问题 → 答案 |
| 工单分类 | 自动将工单分到合适类别 | 输入工单内容 → 输出类别 |
| 信息抽取 | 抽取日期、金额、地址、姓名等 | 输入文本 → JSON 实体列表 |
Training Data
| Task | Dataset | Samples |
|---|---|---|
| FAQ | COIG-CQIA (wiki + douban) | ~10K |
| Doc QA | CMRC2018 + DRCD | ~10K |
| Classification | CLUE TNEWS | ~5K |
| NER/IE | InstructIE-zh + COIG-CQIA | ~5K |
Training Configuration
- Base Model: Qwen/Qwen3-1.7B
- Method: SFT (Supervised Fine-Tuning)
- Learning Rate: 2e-5
- Epochs: 3
- Batch Size: 4 × 8 (effective 32)
- Max Length: 2048
- Packing: enabled (BFD)
Usage
Training Script
See in this repository for the complete training script.
To run training:
For GPU training, use:
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support