shan-gemma4-E2B-it-lora

Shan Gemma 4 E2B LoRA

A Shan/Tai language learning model fine-tuned from Gemma 4 for educational use, translation, grammar explanation, and community language preservation.

---

Overview

Shan Gemma 4 E2B LoRA is an instruction-tuned Shan language model built upon Google's Gemma 4 architecture.

The model specializes in:

• Shan-only conversations
• English → Shan translation
• Burmese → Shan translation
• Shan grammar explanation
• Vocabulary learning
• Reading comprehension
• Educational dialogue generation
• Student tutoring

---

Model Summary

---

Intended Use

This model is designed for:

• Shan language learning
• Educational chatbot
• Translation assistance
• Vocabulary explanations
• Grammar tutoring
• Reading comprehension
• Daily conversation practice
• Student learning support

---

Dataset

The model was trained on a custom Shan educational instruction dataset.

Dataset Categories

Dataset Statistics

---

Training Details

---

Training Log

---

Training Curve

---

Example Prompts

Vocabulary Learning

User

Explain the Shan word for water.

Assistant

ၼမ်ႉ ၼႆႉ မၢႆထိုင် water။
တူဝ်ယၢင်ႇ - ၵဝ်ၶႃႈ ၶႂ်ႈၵိၼ်ၼမ်ႉ။

---

Translation

User

Translate to Shan:

I want to learn Shan language.

Assistant

ၵဝ်ၶႃႈ ၶႂ်ႈႁဵၼ်း ၽႃႇသႃႇတႆး။

---

Grammar

User

Explain what a noun is.

Assistant

ၼၢမ်ႇၼႆႉ ပဵၼ်ၶေႃႈၵႂၢမ်းဢၼ်ၸီႉၸူး
ၵူၼ်း၊ သတ်း၊ တီႈယူႇ ဢမ်ႇၼၼ်
သင်ႇၶွင်ႇတၢင်းမူတ်း။

---

Usage

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="k1ngtai/shan-gemma4-E2B-it-lora",
    max_seq_length=1024,
    load_in_4bit=True,
)

FastLanguageModel.for_inference(model)

prompt = """<start_of_turn>user
You are a Shan language learning assistant.

Instruction: Teach Shan vocabulary.
Input: Give me 5 beginner Shan words about family.
<end_of_turn>
<start_of_turn>model
"""

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(
    **inputs,
    max_new_tokens=300,
    temperature=0.7,
    top_p=0.9,
    do_sample=True,
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

Evaluation

Strengths

• Shan vocabulary learning
• Educational conversations
• Translation assistance
• Grammar explanations
• Shan-only responses

Areas for Improvement

• Complex reasoning
• Literary translation
• Rare vocabulary
• Long context conversations

---

Limitations

• This is an experimental educational model.
• Shan translation quality may vary.
• The model may produce incorrect grammar or vocabulary.
• Not intended for legal, medical, financial, or safety-critical use.
• Best used with human review for learning materials.

---

Safety and Scope

This model is intended for peaceful educational use only.

It is focused on Shan/Tai language learning and community education.

---

Acknowledgements

• Google Gemma
• Unsloth
• Hugging Face
• Shan/Tai language community

---

Citation

@misc{shan-gemma4-e2b-lora,
  title={Shan Gemma 4 E2B LoRA},
  author={K1ngtai},
  year={2026},
  publisher={Hugging Face}
}