Downloads · 30 days
6
14% of all-time downloads
qwen35-cskh-lora
Quan757/qwen35-cskh-lora
qwen35-cskh-lora is a text generation model from Quan757. Use it when you need the model to write or continue text. It is set up for peft.
Sinh viên: Nguyễn Văn Quân MSSV: 2A202601544 Base model: Qwen/Qwen3.5-0.8B Adapter: LoRA text-linear, rank 16, 58 optimizer steps, bf16
All-time downloads
44
Public
Repo size
43.3 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors43.3 MB · 100%
At a glance
- Task
- Text Generation
- Library
- peft
- Access
- Public
- Created
- Aug 21, 2026
- Updated
- Aug 21, 2026
- SHA
- 689d39a8
Base models
Model card
From the Hugging Face model README
Lab 21 — Vietnamese CSKH Triage LoRA
Sinh viên: Nguyễn Văn Quân
MSSV: 2A202601544
Base model: Qwen/Qwen3.5-0.8B
Adapter: LoRA text-linear, rank 16, 58 optimizer steps, bf16
Adapter chuyển ticket chăm sóc khách hàng tiếng Việt thành JSON gồm bốn trường:
intent, urgency, product, sentiment.
Kết quả
| Run | Target | Regression | Format | Latency ms/mẫu |
|---|---|---|---|---|
| Base + optimized prompt | 0.5000 | 0.6444 | 1.0000 | 6237.7 |
| LoRA fine-tune | 0.9900 | 0.0333 | 1.0000 | 908.5 |
Verdict của regression gate là FAILED: target tăng +0.490 nhưng regression giảm
-0.611. Adapter phù hợp để nghiên cứu worker triage được routing chặt, chưa nên dùng
như model đa dụng nếu chưa bổ sung replay data và đánh giá lại.
Artefact chấm bài
- Báo cáo đầy đủ:
submission/REPORT.md - Kết quả đánh giá:
results/ - Adapter:
adapter_config.jsonvàadapter_model.safetensors
Load adapter
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_id = "Qwen/Qwen3.5-0.8B"
adapter_id = "Quan757/qwen35-cskh-lora"
tokenizer = AutoTokenizer.from_pretrained(base_id, trust_remote_code=True)
base = AutoModelForCausalLM.from_pretrained(base_id, device_map="auto")
model = PeftModel.from_pretrained(base, adapter_id)
- Task
- Text Generation
- Library
- peft
- Languages
- vi
- Created
- Aug 21, 2026
- Updated
- Aug 21, 2026