| --- |
| language: |
| - en |
| - ar |
| pipeline_tag: text-generation |
| tags: |
| - operator |
| - hajerilm |
| - cultureyes |
| - arabic |
| - english |
| --- |
| |
| # Operator 4B |
|
|
| Operator is an Arabic-English model from the HajeriLM family, published by Cultureyes. |
|
|
| ## العربية |
|
|
| **Operator 4B** نموذج لغوي عربي وإنجليزي من عائلة **HajeriLM**، تنشره **Cultureyes**. أشارك أوزانه ليتمكن الباحثون والمطورون من فحصها، ومواصلة التدريب، وتخصيص النموذج، والبناء عليه. طُوّر هذا العمل المستقل في الإمارات مع التركيز على الاستفادة الجيدة من وقت الحوسبة والموارد المتاحة. |
|
|
| I’m sharing this work in the spirit of the UAE’s commitment to innovation. I built it by working with the compute I had and repeatedly asking how to make each training step count. I’m sharing the weights so other people can inspect them, continue training, specialize them, and build their own work on top. |
|
|
| The model is an open starting point from an independent effort in the UAE, shared for people who want to explore what they can build from it. |
|
|
| ## A personal note |
|
|
| I’m sharing Operator during [UAE Innovates 2026](https://uaeinnovates.gov.ae/), as my own contribution from the UAE. This is independent work, not a government project or a claim that one model can match the resources of a major lab. I built with the compute I had, and I’m releasing the weights so others can inspect them, continue training, specialize them, and build something of their own. |
|
|
| I keep coming back to [President Sheikh Mohamed bin Zayed’s description](https://www.thenationalnews.com/news/uae/2024/05/29/president-sheikh-mohamed-concludes-important-visit-to-south-korea/) of an Emirati who talks about AI and still gathers firewood and makes coffee in the desert: “This is the true Emirati.” At the Union Parade, an Emirati participant recalled the President’s words: “all tribes in the UAE are one tribe.” [Sheikh Khaled bin Mohamed bin Zayed put the scale of our ambition simply](https://www.govinfo.gov/content/pkg/DCPD-202500610/pdf/DCPD-202500610.pdf): “But we punch above our weight.” That is the spirit I hope this small independent effort carries. |
|
|
| ## ملاحظة شخصية |
|
|
| أشارك Operator خلال [الإمارات تبتكر 2026](https://uaeinnovates.gov.ae/) بوصفه مساهمة شخصية من الإمارات. هذا عمل مستقل، وليس مشروعًا حكوميًا أو ادعاءً بأن نموذجًا واحدًا يوازي إمكانات المختبرات الكبرى. بنيته بالحوسبة المتاحة لي، وأنشر أوزانه كي يتمكن الآخرون من فحصها ومواصلة تدريبها وتخصيصها والبناء عليها. |
|
|
| أعود كثيرًا إلى [وصف صاحب السمو الشيخ محمد بن زايد](https://www.thenationalnews.com/news/uae/2024/05/29/president-sheikh-mohamed-concludes-important-visit-to-south-korea/) للمواطن الإماراتي الذي يتحدث عن الذكاء الاصطناعي، ويحافظ في الوقت نفسه على تقاليد البر والقهوة: «هذا هو الإماراتي الحقيقي». وفي مسيرة الاتحاد، نقل أحد المشاركين عن سموه قوله إن «كل القبائل في الإمارات قبيلة واحدة». وقال سمو الشيخ خالد بن محمد بن زايد عن طموح الدولة: «لكننا نحقق أثرًا يتجاوز حجمنا» (ترجمة للعبارة الإنجليزية). هذه هي الروح التي أتمنى أن يحملها هذا العمل المستقل الصغير. |
|
|
| ## Languages |
|
|
| Operator was trained for use in **English and Arabic**. The model card and examples should be read with that scope in mind; this statement does not imply equal performance across languages or tasks. |
|
|
| دُرّب Operator للاستخدام باللغتين **العربية والإنجليزية**. ولا يعني ذلك تكافؤ الأداء في اللغتين أو في جميع المهام. |
|
|
| ## Model |
|
|
| - **Model family:** HajeriLM |
| - **Publisher:** Cultureyes |
| - **Format:** BF16 safetensors |
| - **Parameter scale:** approximately 4B |
| - **Architecture:** 32 layers, hidden size 2048, 16 attention heads, 2 key/value heads |
| - **Position encoding:** RoPE with theta 500,000; maximum configured context 32,768 tokens |
| - **Attention gates:** enabled |
| - **License:** not specified; reuse terms have not been declared |
|
|
| يتكوّن النموذج من 32 طبقة، وبُعد مخفي قدره 2048، ويستخدم 16 رأس انتباه ورأسي مفتاح/قيمة. يعتمد RoPE بقيمة theta تساوي 500,000، مع سياق أقصى مضبوط على 32,768 رمزًا، وبوابات انتباه مفعّلة. |
|
|
| ## Local inference |
|
|
| GGUF conversion and llama.cpp support are still being reviewed. No GGUF file is part of this release yet. I’ll add local-inference files after the conversion and runtime checks are complete. |
|
|
| ## الاستدلال المحلي |
|
|
| لا يزال تحويل GGUF ودعم llama.cpp قيد المراجعة. لا يتضمن هذا الإصدار ملفات GGUF حاليًا. سأضيف ملفات الاستدلال المحلي بعد اكتمال التحقق من التحويل والتشغيل. |
|
|
| ## Training approach |
|
|
| The training workflow was built around a practical question: how can limited compute be used more carefully? The approach tracks cross-entropy per example, can stop spending updates on examples below a chosen loss threshold, saves resumable checkpoints, and adapts the learning rate from training loss. The aim is to focus training effort where it is still useful and make model building more accessible beyond large clusters. |
|
|
| ## Intended use |
|
|
| Use this checkpoint for research, experimentation, continued pretraining or fine-tuning, and local model development. Test it on your own tasks before relying on it in a product or consequential setting. |
|
|
| The weights are publicly downloadable, but no license grant is declared in this release. Reuse and redistribution terms remain unresolved. |
|
|
| ## Citation and attribution |
|
|
| Please refer to this model as **Operator**, from the **HajeriLM family**, published by **Cultureyes**. Preserve the model name and source link when sharing derivatives. |
|
|
| ## Release notes |
|
|
| Further development notes and the journal will document how the work evolves. |
|
|
