Gemma 4 E2B Yemeni Arabic Assistant GGUF

gemma-4-e2b-yemeni-arabic-assistant-GGUF provides GGUF builds of a Gemma 4 E2B Yemeni Arabic assistant model for local inference with llama.cpp-compatible runtimes.

The model is adapted for broad Yemeni Arabic responses. It is intended to answer, rewrite, and translate text using Yemeni-style Arabic rather than defaulting to formal Modern Standard Arabic.

Yemeni Arabic is underrepresented in public LLM resources. This release is intended to support practical local inference and research experimentation around low-resource Yemeni Arabic dialect adaptation.

Model Details

• Model repo: yemeni-ai-lab/gemma-4-e2b-yemeni-arabic-assistant-GGUF
• Base model: unsloth/gemma-4-E2B-it
• Developer: Loay Amin
• Organization: Yemeni AI Lab
• Format: GGUF
• Runtime target: llama.cpp-compatible inference
• Language: Arabic
• Dialect focus: Broad Yemeni Arabic
• Task type: Text generation / assistant responses / dialect rewriting

Intended Use

This model is intended for:

• Yemeni Arabic assistant responses
• Yemeni dialect rewriting
• Translation or adaptation into Yemeni Arabic
• Short localized Arabic replies
• Yemeni vocabulary and expression experiments
• Local inference on consumer hardware
• Research on low-resource Arabic dialect adaptation

Example use cases:

User: ترجم للهجة اليمنية: I am tired today and I do not want to go out.
Assistant: أنا تعبان اليوم وما اشتي أخرج.

User: اكتبها بطريقة يمنية أبسط: لا أريد أن أتكلم عن الموضوع الآن.
Assistant: ما اشتي أتكلم عن الموضوع ذلحين.

Recommended Prompt

Use a short system prompt that controls assistant behavior:

أنت مساعد يمني ذكي ومختصر.
جاوب باللهجة اليمنية الطبيعية.
لا تضف تحية إلا إذا المستخدم بدأ بتحية.
لا تختم بسؤال أو عرض مساعدة.
لا تضف رأيك إلا إذا المستخدم طلب رأيك.
في الترجمة أو إعادة الصياغة: أعطِ النص فقط بدون شرح.

For translation-only tasks:

ترجم للهجة اليمنية فقط. لا تضف مقدمة ولا شرح.

Recommended Inference Settings

Suggested settings:

temperature: 0.3 - 0.5
top_p: 0.85 - 0.9
top_k: 40
repeat_penalty: 1.08 - 1.12
max_tokens: 160 - 256

For translation or rewriting:

temperature: 0.2 - 0.3
max_tokens: 120 - 180

If the model becomes too verbose, reduce max_tokens first.

Usage with llama.cpp

Download one GGUF file from the repository, then run it with a llama.cpp-compatible build:

llama-cli \
  -m gemma-4-e2b-yemeni-arabic-assistant.Q4_K.gguf \
  -cnv \
  -p "أنت مساعد يمني ذكي ومختصر. جاوب باللهجة اليمنية الطبيعية."

For a direct translation prompt:

llama-cli \
  -m gemma-4-e2b-yemeni-arabic-assistant.Q4_K.gguf \
  -p "ترجم للهجة اليمنية فقط. لا تضف مقدمة ولا شرح.\n\nالنص: I am tired today and I do not want to go out."

Different llama.cpp builds may use slightly different flags. If your runtime supports Gemma chat templates, use the model's chat template rather than a hand-written prompt format.

Usage with llama-cpp-python

from llama_cpp import Llama

llm = Llama(
    model_path="gemma-4-e2b-yemeni-arabic-assistant.Q4_K.gguf",
    n_ctx=1024,
    n_gpu_layers=-1,
)

prompt = """أنت مساعد يمني ذكي ومختصر.
جاوب باللهجة اليمنية الطبيعية.
لا تضف تحية إلا إذا المستخدم بدأ بتحية.
لا تختم بسؤال أو عرض مساعدة.

المستخدم: ترجم للهجة اليمنية: I am tired today and I do not want to go out.
المساعد:"""

out = llm(
    prompt,
    max_tokens=160,
    temperature=0.3,
    top_p=0.9,
    repeat_penalty=1.1,
)

print(out["choices"][0]["text"].strip())

Training Data Summary

The source adapter was trained on 19,609 cleaned Yemeni Arabic assistant-style samples.

The dataset was built from a mixture of locally collected and generated resources, including:

• Yemeni social-media style text collected from public/community sources
• Yemeni examples derived from Arabic dialect identification data
• Yemeni vocabulary and expression references
• Local dictionary-style resources
• Grounded teacher-model rewrites and expansions
• Filtered task examples for translation, rewriting, Q&A, reasoning, and calculations

The final training file was cleaned with rule-based and LLM-assisted review passes. The cleaning focused on reducing:

• Non-target dialect leakage such as Gulf/Egyptian expressions
• Overused assistant phrases and service endings
• Excessive greetings and formal closings
• Repetitive openings such as overuse of "والله" and "شوف"
• Long, generic explanations where a direct answer was better
• Obvious hallucinated or malformed Yemeni expressions
• Known unwanted terms from the current data cleaning pass

The final cleaned dataset used for the source fine-tune contained:

• 19,609 rows
• Average assistant response length around 150 characters
• Maximum assistant response length capped at 360 characters

Notes on Yemeni Arabic

Yemeni Arabic is not one dialect. It includes substantial regional variation across Sana'a, Taiz, Aden, Ibb, Hadramout, Tihama, Dhamar, Lahj, Al-Bayda, Marib, and other areas. This model targets a broad, practical Yemeni style for assistant use rather than a single city-specific dialect.

The dataset intentionally avoids some expressions that appeared in generated data but were judged unsuitable for the target style, including examples such as non-target vocatives, Gulf-like fillers, and malformed Yemeni phrases.

Limitations

This is an experimental dialect adaptation model.

Known limitations:

• Quantized GGUF files may behave differently from the original adapter, especially at lower bitrates.
• Very small quantizations can reduce dialect naturalness, reasoning quality, and factual consistency.
• The model may still produce Modern Standard Arabic on technical, formal, or unfamiliar topics.
• The base instruction model may still show polite assistant behavior unless the prompt and output length are controlled.
• The model may occasionally overuse common openings such as "والله" or "شوف".
• It may hallucinate factual or cultural information if used without retrieval or source grounding.
• The training data includes synthetic and teacher-assisted samples, so not every phrase should be treated as corpus-attested Yemeni speech.
• It is not a legal, medical, financial, or religious authority.
• It does not represent every Yemeni region equally.

Ethical and Privacy Considerations

The training work used collected and generated Arabic text for dialect adaptation. Public or community-sourced social-media text can contain names, opinions, jokes, and local references. Cleaning passes were applied to reduce low-quality generations and unwanted assistant behavior, but users should not assume the data pipeline guarantees perfect anonymization or factual accuracy.

This model should not be used to impersonate private individuals, generate harassment, or make high-stakes decisions about people.

Citation

If you use this model, cite it as:

@misc{amin2026gemma4e2byemeniarabicassistantgguf,
  title={Gemma 4 E2B Yemeni Arabic Assistant GGUF},
  author={Amin, Loay},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/yemeni-ai-lab/gemma-4-e2b-yemeni-arabic-assistant-GGUF}}
}

Developer

Developed by Loay Amin under Yemeni AI Lab.

This project is part of an effort to improve AI support for Yemeni Arabic.