Instructions to use nvdat1601/lab22-dpo-vn with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvdat1601/lab22-dpo-vn with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen2.5-7B-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "nvdat1601/lab22-dpo-vn")

Transformers

How to use nvdat1601/lab22-dpo-vn with Transformers:

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

pipe = pipeline("text-generation", model="nvdat1601/lab22-dpo-vn")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("nvdat1601/lab22-dpo-vn", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use nvdat1601/lab22-dpo-vn with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvdat1601/lab22-dpo-vn"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nvdat1601/lab22-dpo-vn",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/nvdat1601/lab22-dpo-vn

SGLang

How to use nvdat1601/lab22-dpo-vn 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 "nvdat1601/lab22-dpo-vn" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nvdat1601/lab22-dpo-vn",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "nvdat1601/lab22-dpo-vn" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nvdat1601/lab22-dpo-vn",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Unsloth Studio

How to use nvdat1601/lab22-dpo-vn with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for nvdat1601/lab22-dpo-vn to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for nvdat1601/lab22-dpo-vn to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for nvdat1601/lab22-dpo-vn to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="nvdat1601/lab22-dpo-vn",
    max_seq_length=2048,
)

Docker Model Runner
How to use nvdat1601/lab22-dpo-vn with Docker Model Runner:
```
docker model run hf.co/nvdat1601/lab22-dpo-vn
```

Lab 22 DPO Vietnamese Adapter

This is a DPO LoRA adapter trained for Day 22 DPO/ORPO Alignment Lab.

Base model

unsloth/Qwen2.5-7B-bnb-4bit

Training data

SFT: 5CD-AI/Vietnamese-alpaca-gpt4-gg-translated
Preference data: argilla/ultrafeedback-binarized-preferences-cleaned

Training setup

Method: DPO
Beta: 0.1
Learning rate: 5e-7
Epochs: 1
LoRA rank: 16
LoRA alpha: 32
Compute tier: BIGGPU

Results

Final DPO loss: 0.7589
Final reward gap: +0.1910

Manual evaluation on 8 prompts showed no clear behavioral improvement over the SFT baseline in this short run. Outputs remained long and sometimes repetitive, especially on safety prompts.

Usage

Load the base model, then apply this PEFT adapter with PeftModel.from_pretrained.

Framework versions

PEFT 0.20.0

Downloads last month: 13

Model tree for nvdat1601/lab22-dpo-vn

Base model

Qwen/Qwen2.5-7B

Quantized

unsloth/Qwen2.5-7B-bnb-4bit

Adapter

(8)

this model