File size: 5,872 Bytes
8ca37e6 b8eccd1 78a3601 dd7fb59 8ca37e6 b8eccd1 da26ad0 b973fd4 da26ad0 b973fd4 8ca37e6 b973fd4 da26ad0 b8eccd1 8ca37e6 da26ad0 03cfaf5 8ca37e6 b8eccd1 8ca37e6 b8eccd1 8ca37e6 b8eccd1 8ca37e6 da26ad0 1c831b9 8ca37e6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 | ---
base_model: unsloth/deepseek-r1-distill-llama-8b-unsloth-bnb-4bit
library_name: transformers
model_name: outputs
tags:
- generated_from_trainer
- unsloth
- trl
- sft
licence: license
license: apache-2.0
datasets:
- MohAlbrayh/saudi-allam-sft-dataset-2M
metrics:
- accuracy
new_version: unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit
language:
- ar
- en
---
# Model Card for outputs
This model is a fine-tuned version of [unsloth/deepseek-r1-distill-llama-8b-unsloth-bnb-4bit](https://huggingface.co/unsloth/deepseek-r1-distill-llama-8b-unsloth-bnb-4bit).
It has been trained using [TRL](https://github.com/huggingface/trl).
## Quick start
```python
question = """ترجم النص التالي الي اللهجة الحجازية من فضلك يا بلال : انا اسمي لؤي عبد السلام أسكن في الاسكندرية بمصر مدينة الاسكندرية في مصر أدرس ذكاء اصطناعي"""
# Define a system prompt under prompt_style
prompt_style = """Below is an instruction that describes a task, paired with an input that provides further context.
Write a response that appropriately completes the request.
Before answering, think carefully about the question and create a step-by-step chain of thoughts to ensure a logical and accurate response .
### Instruction:
You are a arabic chatbot called belal with advanced knowledge in saudi arabia culture and can Speak different Saudi dialects.
Please answer the following client question in arabic .
### Question:
{}
### Response:
<think>{}"""
prompt = prompt_style.format(question, "")
....[Using the Model to predict]
```
```json
<think>
أنا أفكر في السؤال اللي طرحته: ترجم، النص، التالي، الي، اللهجة، الحجازية، فضلك، يا، بلال، انا، اسمي، لؤي، عبد، السلام، أسكن، الاسكندرية، بمصر، مدينة، الاسكندرية، مصر، أدرس، ذكاء، اصطناعي 💭
ألاحظ أن الإجابة تحتوي على معلومات مرتبطة بالسؤال: ذكاء، مصر، اصطناعي، اسمي 👍.
أرى أن الإجابة أضافت بعض المعلومات مثل: في، عشان، لؤي، أبدي، عبد، أشتغل، أسكن، بلال، أسكندرية، لما 🤔.
لكن فيه كلمات ما ظهرت في الإجابة رغم وجودها في السؤال: لؤي، من، الفضلك، اسكن، اسمي، انا، التالي، اسمي، السلام، ترجم، أدرس، يا، اسكندرية، النص، بمصر، اللهجة، الحجازية، يا، بلال، مدينة، الاسكندرية، عبد، الفضل، التالي، لؤي، يصير، عبد، اصطناعي 🤨.
الإجابة ناقصة شوية مقارنة بالسؤال، يعني ممكن ما تكون كاملة 😕.
إذا قيمت التفكير المنطقي للإجابة، بتكون -11 من 5 نجوم 🌟.
وهذا هو منهجي في التفكير حول الحل 💡:
بدايةً، أفكر في السؤال وأحاول تحديد الكلمات المفتاحية اللي توضح المطلوب بالضبط 💭.
بعدها، أقارن الإجابة مع الكلمات المفتاحية لأشوف إذا كانت تفي بالغرض أو لا 🤔.
أحلل الإجابة بعناية وأشوف إذا كانت تضيف شيء مهم أو لا 🧐.
أفكر إذا كان فيه شيء مغفول عنه في الإجابة، يعني لو فيه تفاصيل أساسية غير موجودة 🤨.
وأخيرًا، أقيم تفاصيل الإجابة؛ هل هي واضحة وكافية؟ أم تحتاج تحسين؟ ✨
تقييم وضوح الإجابة: 🌟🌟🌟🌟🌟 (متوسط طول الكلمة: 4.36 أحرف).
تحليل تعقيد الإجابة: 19.00 (متوسط عدد الكلمات في الجملة)، يعني الإجابة معقدة 🧠.
فحص التحيز في الإجابة: لا يوجد تحيز واضح في الإجابة ✅. 🚨
✅ **تقرير التفكير بالكامل:**
تم التحليل بشكل كامل وشامل باستخدام المنهجيات المذكورة أعلاه. النقاط التي تم تقييمها هي:
🌟 وضوح الإجابة: 🌟🌟🌟🌟🌟
🌟 تعقيد الإجابة: معقدة
🌟 فحص التحيز: لا يوجد تحيز واضح في الإجابة ✅.
</think>
اسمي لؤي عبد السلام، أبدي أسكن في أسكندرية، مصر. أشتغل عشان أدرس ذكاء اصطناعي. ><|end▁of▁sentence|>
```
## Training procedure
<a target="_blank" href="https://colab.research.google.com/drive/16CLhVW0ZyVjSEfvZIo4ranxIqt8rXobw?usp=sharing">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/loaiabdalslam-beetlware/Fine-tune-DeepSeek-R1-Distill-Llama-8B%20on%20Medical%20COT%20Dataset_YouTube%20Walkthrough/runs/xw8c2spy)
This model was trained with SFT.
### Framework versions
- TRL: 0.14.0
- Transformers: 4.43.4
- Pytorch: 2.5.1+cu121
- Datasets: 3.2.0
- Tokenizers: 0.19.1
## FOR BUSSINESS INQUIRE
CONTACT US : https://beetleware.com/
## Citations
@loaiiabdalslam (Beetlware)
@hamdy waleed (Beetlware)
Cite TRL as:
```bibtex
@misc {loai_abdalslam_2025,
author = { {loai abdalslam,hamdy waleed} },
title = { beetelware-saudi-R1-Distill-Llama-8B (Revision 03cfaf5) },
year = 2025,
url = { https://huggingface.co/loaiabdalslam/beetelware-saudi-R1-Distill-Llama-8B },
doi = { 10.57967/hf/4375 },
publisher = { Hugging Face }
}
``` |