Instructions to use WYNN747/ai4burmese-padauk with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use WYNN747/ai4burmese-padauk with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/gemma-4-E4B-it")
model = PeftModel.from_pretrained(base_model, "WYNN747/ai4burmese-padauk")

Transformers

How to use WYNN747/ai4burmese-padauk with Transformers:

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

pipe = pipeline("text-generation", model="WYNN747/ai4burmese-padauk")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("WYNN747/ai4burmese-padauk")
model = AutoModelForImageTextToText.from_pretrained("WYNN747/ai4burmese-padauk")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use WYNN747/ai4burmese-padauk with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "WYNN747/ai4burmese-padauk"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "WYNN747/ai4burmese-padauk",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/WYNN747/ai4burmese-padauk

SGLang

How to use WYNN747/ai4burmese-padauk 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 "WYNN747/ai4burmese-padauk" \
    --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": "WYNN747/ai4burmese-padauk",
		"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 "WYNN747/ai4burmese-padauk" \
        --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": "WYNN747/ai4burmese-padauk",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use WYNN747/ai4burmese-padauk with Docker Model Runner:
```
docker model run hf.co/WYNN747/ai4burmese-padauk
```

AI4Burmese Padauk

Mission: Open, free, and accessible Burmese AI so Myanmar is not left behind in the age of AI.

Padauk is part of the AI4Burmese initiative: an open effort to make Burmese AI more accessible, practical, and buildable for everyone, including users, students, researchers, and developers.

It is a practical Burmese-first agentic chat model specialized for day-to-day assistance, tool use, function calling, and compatibility with Skills, MCP servers, and external tools.

Model Details

Developed by: Dr. Wai Yan Nyein Naing
Shared by: WYNN747
Initiative: AI4Burmese
Model type: PEFT LoRA adapter for agentic chat behavior
Base model: unsloth/gemma-4-E4B-it
Languages: Burmese, English
Best for: Day-to-day assistance, workflow support, tool routing, and agentic interaction
Technical project page: https://waiyannyeinnaing.com/projects/padauk
Public model card: https://waiyannyeinnaing.com/projects/padauk
Website: https://ai4burmese.com/
Founder: https://waiyannyeinnaing.com
Hugging Face profile: https://huggingface.co/WYNN747

Model Description

Padauk is designed as a practical assistant layer rather than a pure text-completion model. It is built to help with everyday questions, writing, planning, task guidance, and assistant-style interactions in Burmese and English.

As part of the AI4Burmese mission, this release is intended to support more open and usable Burmese AI for real-world needs, especially for communities that benefit from free, practical, and adaptable local-language AI.

The model is positioned for real assistant workflows:

Direct answers for simple requests
Function-calling style behavior for structured tasks
Compatibility with external Skills, MCP servers, and tools when wired into a runtime
Burmese-first interaction with English support
Practical deployment as part of an open Burmese AI ecosystem

Why AI4Burmese

Burmese is still underserved in modern AI. Many users, students, and builders do not yet have enough open, high-quality, and practically deployable AI systems designed for Burmese language use.

AI4Burmese exists to help close that gap by supporting open-source Burmese AI models and tools that are:

Open for community learning and contribution
Free and accessible for broader public use
Practical for real workflows, not only demos
Buildable so others can adapt and extend them
Burmese-first while remaining useful in bilingual settings

Padauk is one part of that broader effort.

How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

base_model_id = "unsloth/gemma-4-E4B-it"
adapter_id = "WYNN747/ai4burmese-padauk"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    device_map="auto",
)
model = PeftModel.from_pretrained(base_model, adapter_id)

messages = [
    {"role": "system", "content": "You are Padauk, a practical Burmese-first agentic AI assistant from the AI4Burmese initiative."},
    {"role": "user", "content": "Help me plan a productive work day."},
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=False)
print(response)

Tool execution, Skills, and MCP server calls should be implemented in the surrounding application or runtime. Padauk is intended to decide when a tool should be used, not to replace the orchestration layer.

Best-Suited Uses

Everyday assistance and practical Q&A
Writing and rewriting
Planning and task breakdowns
Search-assisted workflows
Coding help and technical support
Tool-augmented assistant experiences
Skill-driven or MCP-connected workflows
Burmese-first AI applications and experiments

Training Notes

This release is framed as an agentic assistant fine-tune built on Gemma 4 E4B Instruct. The public positioning focuses on assistant behavior, tool use, practical task handling, and Burmese-first usability.

This model card intentionally stays concise so the release remains aligned with product and community use. A future expanded technical appendix may include:

Dataset details
Training recipe
Alignment method
Evaluation results
Quantization or deployment notes

Out-of-Scope & Limitations

Not a substitute for verified sources, professional advice, or domain experts
Tool access depends on the runtime and connected integrations
Function-calling quality depends on tool schema quality and orchestration
Results should be reviewed before use in high-stakes settings
Real-world usefulness depends on the surrounding application design and deployment setup

Citation

If you use Padauk in research or product work, please cite the model page and project page:

@misc{padauk2026,
  title = {AI4Burmese Padauk: Practical Burmese-First Agentic AI Assistant},
  author = {Wai Yan Nyein Naing},
  year = {2026},
  url = {https://huggingface.co/WYNN747/ai4burmese-padauk}
}