Instructions to use tiennn/day22-qwen25-3b-sft-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiennn/day22-qwen25-3b-sft-mini with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen2.5-3B-Instruct-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "tiennn/day22-qwen25-3b-sft-mini")

Transformers

How to use tiennn/day22-qwen25-3b-sft-mini with Transformers:

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

pipe = pipeline("text-generation", model="tiennn/day22-qwen25-3b-sft-mini")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("tiennn/day22-qwen25-3b-sft-mini", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tiennn/day22-qwen25-3b-sft-mini with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiennn/day22-qwen25-3b-sft-mini"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiennn/day22-qwen25-3b-sft-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tiennn/day22-qwen25-3b-sft-mini

SGLang

How to use tiennn/day22-qwen25-3b-sft-mini 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 "tiennn/day22-qwen25-3b-sft-mini" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiennn/day22-qwen25-3b-sft-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "tiennn/day22-qwen25-3b-sft-mini" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiennn/day22-qwen25-3b-sft-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tiennn/day22-qwen25-3b-sft-mini with Docker Model Runner:
```
docker model run hf.co/tiennn/day22-qwen25-3b-sft-mini
```

Day 22 Qwen2.5-3B Vietnamese SFT-mini LoRA

This is the SFT stage of a local-GPU preference-alignment lab. It adapts unsloth/Qwen2.5-3B-Instruct-bnb-4bit on a deterministic 1,000-example slice of bkai-foundation-models/vi-alpaca before a distinct DPO adapter is trained.

Training

Setting	Value
LoRA rank / alpha	16 / 32
Epochs	1
Max sequence length	512
Learning rate	2e-4
Effective batch size	8
Train loss	0.9616
Runtime	433.5 s
Peak allocated VRAM	3.099 GB

The run used BF16 compute on an NVIDIA RTX 5060 Ti 16 GB. Full reproducibility metadata, loss history, tokenizer files, and a Vietnamese generation smoke test are included in this folder.

Use and limitations

Load the base model in 4-bit and attach this adapter with PEFT. This is an educational checkpoint, not a production safety model. The small training slice does not establish broad factuality, robustness, fairness, or medical/legal reliability. Review generated content before downstream use.

Source code and evaluation evidence: https://github.com/Tienlee41/K3-Track3-Day22-2A202601145_LeAnhTien