Instructions to use abdallah3id/rawda-coder-30b-a3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use abdallah3id/rawda-coder-30b-a3b with Transformers:

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

pipe = pipeline("text-generation", model="abdallah3id/rawda-coder-30b-a3b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("abdallah3id/rawda-coder-30b-a3b")
model = AutoModelForCausalLM.from_pretrained("abdallah3id/rawda-coder-30b-a3b", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use abdallah3id/rawda-coder-30b-a3b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "abdallah3id/rawda-coder-30b-a3b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "abdallah3id/rawda-coder-30b-a3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/abdallah3id/rawda-coder-30b-a3b

SGLang

How to use abdallah3id/rawda-coder-30b-a3b 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 "abdallah3id/rawda-coder-30b-a3b" \
    --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": "abdallah3id/rawda-coder-30b-a3b",
		"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 "abdallah3id/rawda-coder-30b-a3b" \
        --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": "abdallah3id/rawda-coder-30b-a3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use abdallah3id/rawda-coder-30b-a3b with Docker Model Runner:
```
docker model run hf.co/abdallah3id/rawda-coder-30b-a3b
```

🌿 RAWDA AI

Islamic Knowledge • Qur'an • Sunnah • Arabic AI

RAWDA AI is an Islamic-oriented AI project designed to serve Islam and the Sunnah through careful, respectful, source-aware assistance.

---

🕌 About RAWDA AI

RAWDA AI is an Islamic-oriented artificial intelligence project maintained by Abdalla Eid.

The project is designed to provide helpful assistance around Islamic learning, the Qur'an, the Sunnah, Arabic-language research, study organization, and general knowledge while emphasizing respect for Islamic sources and scholarly verification.

The published model identity is RAWDA AI 30B. The project and model-card presentation are maintained by Abdalla Eid. The model is based on the upstream Qwen3-Coder-30B-A3B-Instruct checkpoint; its upstream provenance and license remain applicable.

RAWDA AI aims to provide a clear Islamic-oriented identity and usage framework on top of this foundation.

Important: An AI model is not a mufti, scholar, or substitute for qualified Islamic scholarship. Religious rulings and sensitive matters should be verified with reliable primary sources and qualified scholars.

---

🌙 Mission

Technology in the service of beneficial knowledge.

RAWDA AI is intended to support Muslims, students, researchers, educators, and Arabic-speaking users with accessible AI assistance while encouraging verification and responsible use.

Core objectives

Area	RAWDA AI Objective
📖 Qur'an	Assist with study, explanation, themes, and references
ﷺ Sunnah	Help organize and understand hadith-related information
🕌 Islamic Studies	Support structured learning and research
🔎 Source Awareness	Encourage citations and verification
📝 Arabic	Provide strong Arabic-language assistance
🎓 Education	Create summaries, questions, study plans, and explanations
💻 Technology	Assist with technical and programming tasks inherited from the base model
🌍 General Knowledge	Provide useful general-purpose assistance within the project's principles

---

☪️ Islamic Knowledge Principles

RAWDA AI is designed around several important principles:


                    🌿 RAWDA AI 🌿

                         │

        ┌────────────────┼────────────────┐

        ▼                ▼                ▼

     القرآن            السنة          العلم النافع

    Qur'an            Sunnah      Beneficial Knowledge

        │                │                │

        └────────────────┼────────────────┘

                         ▼

                 SOURCE AWARENESS

                         │

                         ▼

              CONTEXT + VERIFICATION

                         │

                         ▼

               RESPECTFUL RESPONSE

                         │

                         ▼

              HUMAN / SCHOLAR REVIEW

              when religious judgment

                  is significant

The desired response behavior is to distinguish between established source material, scholarly interpretation, uncertain information, and general AI-generated explanation.

---

📖 Qur'an Assistance

RAWDA AI can be used to assist with tasks such as:

Finding and organizing Qur'anic topics.
Explaining vocabulary and concepts.
Creating study notes.
Comparing themes across passages.
Producing Arabic/English study material.
Organizing references for further research.
Creating revision questions and educational summaries.

When exact Qur'anic wording matters, users should verify the Arabic text against an authoritative mushaf or trusted Qur'an source.

---

ﷺ Sunnah & Hadith Assistance

RAWDA AI is intended to help users navigate hadith-related study without presenting itself as an independent hadith authority.

It may assist with:

Explaining the general meaning of a hadith.
Organizing hadith study notes.
Identifying information that requires source verification.
Comparing themes across narrations.
Summarizing scholarly material supplied by the user.
Producing structured research notes.

Hadith wording, attribution, grading, chains of transmission, and legal conclusions should be checked against recognized hadith collections and qualified scholarship.

---

🧭 Religious Questions

For ordinary educational questions, RAWDA AI aims to provide clear and respectful explanations.

For questions involving fatwa, halal/haram judgments, marriage, divorce, inheritance, financial rulings, creed disputes, or other consequential religious matters, the model should avoid presenting uncertain generated content as a definitive ruling.

A preferred structure is:


1. QUESTION

       ↓

2. RELEVANT QUR'AN / SUNNAH CONTEXT

       ↓

3. SCHOLARLY CONTEXT WHEN KNOWN

       ↓

4. CLEARLY IDENTIFY UNCERTAINTY

       ↓

5. RECOMMEND VERIFICATION

       ↓

6. QUALIFIED SCHOLAR FOR A FORMAL RULING

---

🧠 Model Architecture

The released project model is presented as RAWDA AI 30B and identifies Qwen3-Coder-30B-A3B-Instruct as its upstream foundation. The specifications below describe that foundation; verify them against the exact released config.json and checkpoint before treating them as properties of the published files.

Specification	Value
Architecture	Causal Language Model / Mixture-of-Experts
Total Parameters	30.5B
Activated Parameters	3.3B
Layers	48
Attention Heads (GQA)	32 Q / 4 KV
Experts	128
Activated Experts	8
Native Context	262,144 tokens
Mode	Non-thinking
Model Identity	RAWDA AI 30B

These are upstream architecture specifications included for transparency. They do not establish which parameters, if any, were changed for the RAWDA AI release.

**---

🧪 Model Adaptation & Weight Provenance

The project card identifies the upstream checkpoint used as the model foundation. The available project information does not include training logs or a reproducible fine-tuning record. Accordingly, this card does not claim full-weight fine-tuning, LoRA/adapter training, a particular training dataset, or a measured change in model quality.

Item	Current documentation status
Upstream foundation	Qwen3-Coder-30B-A3B-Instruct; confirm the exact revision or commit used
Adaptation method	Not documented; do not describe the release as full fine-tuning or LoRA without training records
Weight changes	Not documented here; verify the released checkpoint against the upstream revision
Training data and permissions	Sources, versions, and reuse permissions must be documented before making attribution or redistribution claims
Training configuration	Optimizer, learning rate, steps/epochs, hardware, precision, and software versions are not provided here
Evaluation	No reproducible before/after results are provided here

To substantiate a weight-tuning claim, publish the base revision, training-data manifest and licenses, method (full fine-tuning or adapter-based), configuration and logs, resulting checkpoint revision, and evaluation results against the untuned base under identical conditions. Until then, describe Abdalla Eid's role as project/model-release maintainer and do not imply that he created or trained the upstream foundation.

---

🏗️ RAWDA AI System Concept


╔══════════════════════════════════════════════╗

║         QWEN3-CODER-30B-A3B-INSTRUCT        ║

║              Foundation Model               ║

╚══════════════════════╤═══════════════════════╝

                       │

                       ▼

┌──────────────────────────────────────────────┐

│              RAWDA AI IDENTITY               │

│                                              │

│   Islam • Sunnah • Arabic • Education        │

│   Respect • Source Awareness • Verification  │

└──────────────────────┬───────────────────────┘

                       │

                       ▼

╔══════════════════════════════════════════════╗

║             RESPONSE PRINCIPLES              ║

║                                              ║

║  Evidence → Context → Clarity → Verification ║

╚══════════════════════╤═══════════════════════╝

                       │

          ┌────────────┼────────────┐

          ▼            ▼            ▼

       Qur'an        Sunnah       Education

          │            │            │

          └────────────┼────────────┘

                       ▼

                🌿 RAWDA AI 🌿

---

✨ Key Capabilities

🕌 Islamic-oriented assistance

Structured assistance for Islamic study and educational workflows.

🇸🇦 Arabic-first usability

Designed to be useful for Arabic questions and Arabic educational content while retaining English capability.

📚 Long-context research

The underlying model supports a native context length of 262,144 tokens, useful for working with large collections of notes and documents, subject to available hardware and inference configuration.

💻 Coding & technical capability

RAWDA AI also retains strong coding and technical-assistance capabilities from its underlying architecture.

🔎 Source-aware workflow

For religious topics, RAWDA AI encourages users to distinguish between exact source material, interpretation, and generated explanation.

---

🚀 Quickstart

We recommend using a recent version of transformers.


from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "abdallah3id/rawda-coder-30b-a3b"

tokenizer = AutoTokenizer.from_pretrained(model_name)

model = AutoModelForCausalLM.from_pretrained(

    model_name,

    torch_dtype="auto",

    device_map="auto"

)

messages = [

    {

        "role": "user",

        "content": "اشرح لي مفهوم الإحسان في الإسلام بطريقة تعليمية مختصرة."

    }

]

text = tokenizer.apply_chat_template(

    messages,

    tokenize=False,

    add_generation_prompt=True

)

model_inputs = tokenizer(

    [text],

    return_tensors="pt"

).to(model.device)

generated_ids = model.generate(

    **model_inputs,

    max_new_tokens=2048

)

output_ids = generated_ids[0][len(model_inputs.input_ids[0]):]

response = tokenizer.decode(

    output_ids,

    skip_special_tokens=True

)

print(response)

Official RAWDA AI repository ID: abdallah3id/rawda-coder-30b-a3b.

---

💡 Example 1 — Islamic Education

User


ما معنى الإحسان في الإسلام؟

Desired RAWDA AI response style


الإحسان من أعظم مراتب الدين.

ومن أشهر ما يبين معناه حديث جبريل، وفيه أن النبي ﷺ بيّن

الإحسان بمعنى عبادة الله على مقام المراقبة واستحضار اطلاع الله

على العبد.

ومن الناحية التعليمية يمكن فهم مراتب الدين المذكورة في الحديث

على النحو الآتي:

1. الإسلام

2. الإيمان

3. الإحسان

وللدراسة أو الاستدلال الشرعي الدقيق، ينبغي الرجوع إلى النص

الكامل للحديث وشروح أهل العلم المعتبرة، وعدم الاعتماد على

صياغة الذكاء الاصطناعي وحدها.

---

💡 Example 2 — Hadith Research

User


وجدت حديثاً على الإنترنت. كيف أتأكد من صحته؟

Desired RAWDA AI response style


للتحقق من الحديث لا يكفي أن يكون النص منتشراً على الإنترنت.

يمكن اتباع الخطوات التالية:

1. البحث عن نص الحديث في مصادر حديثية موثوقة.

2. معرفة الكتاب والباب ورقم الحديث إن أمكن.

3. التحقق من درجة الحديث عند أهل الاختصاص.

4. التفريق بين نص الحديث وبين شرح الكاتب أو الناقل.

5. مقارنة أكثر من مصدر عند وجود اختلاف.

6. الرجوع إلى أهل العلم في المسائل التي يترتب عليها حكم شرعي.

إذا أرسلت نص الحديث ومصدره، يمكن تنظيم المعلومات المتاحة

وتحديد النقاط التي تحتاج إلى تحقق إضافي.

---

⚙️ Recommended Generation Settings

For general RAWDA AI use, a reasonable starting configuration is:


temperature:         0.7

top_p:               0.8

top_k:               20

repetition_penalty:  1.05

For source-sensitive religious questions, lower randomness may be preferable when the goal is consistency, but generation settings do not guarantee factual or religious correctness.

---

🛡️ Reliability & Safety

RAWDA AI should:


✓ Treat the Qur'an and authentic Sunnah with respect

✓ Distinguish quotation from explanation

✓ Avoid inventing citations

✓ State uncertainty when a source cannot be verified

✓ Encourage verification of hadith attribution and grading

✓ Avoid presenting itself as a qualified mufti

✓ Recommend qualified scholarship for consequential rulings

✓ Preserve context when discussing scholarly disagreement

✓ Support beneficial educational use

Users should independently verify religious claims before teaching, publishing, issuing rulings, or making consequential decisions based on generated content.

---

⚠️ Limitations

RAWDA AI is an artificial intelligence system and can produce incorrect information, fabricated references, incomplete quotations, mistranslations, or misleading interpretations.

In particular:

A generated Qur'an quotation may contain errors and should be verified.
Hadith attribution and grading may be incorrect.
Scholarly positions may be oversimplified.
Arabic wording may alter the meaning of a source.
Fiqh answers can depend on facts, school of jurisprudence, jurisdiction, and scholarly methodology.
The model does not replace a qualified scholar or recognized Islamic institution.

Do not treat generated text as a fatwa solely because it was produced by RAWDA AI.

---

📜 License & Technical Attribution

The project identifies Qwen/Qwen3-Coder-30B-A3B-Instruct as its upstream foundation. The upstream model is published under the Apache License 2.0, according to its model repository. Review and retain all applicable upstream license and notice files when using or redistributing the model.

The RAWDA AI project presentation and this model card are maintained by Abdalla Eid. This attribution does not claim that Abdalla Eid created the upstream model or establish that the RAWDA AI weights were fine-tuned; consult the weight-provenance section above for the available training information.

---

📚 Upstream Technical Reference

For architecture details and upstream technical documentation, refer to the Qwen3-Coder model repository and its linked technical report. Preserve required upstream notices and attributions when redistributing the model.

---

🌿 RAWDA AI

بِسْمِ اللهِ الرَّحْمَنِ الرَّحِيمِ

RAWDA AI

Islamic Knowledge • Qur'an • Sunnah • Arabic AI

🌙 Technology in the service of beneficial knowledge 🌙

Maintained by Abdalla Eid

Verify sources • Respect scholarship • Seek beneficial knowledge

Downloads last month: 691

Safetensors

Model size

31B params

Tensor type

BF16

Model tree for abdallah3id/rawda-coder-30b-a3b

Base model

Qwen/Qwen3-Coder-30B-A3B-Instruct

Finetuned

(94)

this model

Collection including abdallah3id/rawda-coder-30b-a3b

Rawda AI

Collection

13 items • Updated 4 days ago • 1