Fikr-7B-Reasoning
is an Arabic reasoning model fine-tuned from
Qwen/Qwen2.5-7B-Instruct
using supervised fine-tuning (SFT) to enforce structured, multi-step Chain-of-Thought reasoning inside native
<think>...</think>
tokens.
📊 Benchmark Results
Both models were evaluated on the full test split (
1,319 questions
) of the
Arabic-GSM8K
benchmark using exact numerical match via greedy decoding (
temperature=0.0
) on an NVIDIA A100 GPU:
Model
Architecture
Arabic Reasoning Format
Arabic-GSM8K Accuracy
Delta (Δ)
Qwen/Qwen2.5-7B-Instruct
(Base)
7B Dense
Standard CoT
72.40%
[cite: 3]
Baseline
Fikr-7B-Reasoning
(Ours)
7B Dense
<think>
Native CoT
70.81%
[cite: 4]
-1.59%
Key Takeaways
Accuracy Trade-off:
While the model strictly enforces the
<think>
tag reasoning structure, it experienced a minor
-1.59%
regression in raw exact-match accuracy compared to the strong Qwen 2.5 7B baseline[cite: 3, 4]. This is a common phenomenon in first-pass SFT reasoning models and typically requires a follow-up Reinforcement Learning (RL) phase (like DPO or PPO) to align the structured thoughts with correct final answers.
Clean Trace Separation:
The fine-tuned model strictly separates intermediate logical steps inside
<think>...</think>
from the final parsed response[cite: 4].
Arabic Mathematical Fluency:
The model demonstrates improved formatting stability across multi-step arithmetic, fraction reductions, and currency conversions in native Arabic phrasing.
🔍 Sample Comparison
Problem:
ذهب كايلار إلى المتجر لشراء أكواب لشقة الجديدة. تكلفة الكوب الواحد 5 دولارات، لكن كل كوب ثاني يكلف فقط 60٪ من السعر. يريد كايلار شراء 16 كوبًا. كم يجب أن يدفع مقابلها؟
[cite: 3]
Fikr-7B-Reasoning (Ours)
<think>
1. سعر الكوب الأول هو 5 دولارات.
2. سعر الكوب الثاني هو 60٪ من 5 دولارات، أي 5 * 0.60 = 3 دولارات.
3. لحساب التكلفة لـ 16 كوبًا:
- الكؤوس الأولى (8 كؤوس): 8 * 5 = 40 دولارًا.
- الكؤوس المخفضة (8 كؤوس): 8 * 3 = 24 دولارًا.
- التكلفة الإجمالية: 40 + 24 = 64 دولارًا.
</think>
لذا، يجب على كايلار أن يدفع 64 دولارًا مقابل 16 كوبًا.
Prompt Format
To trigger full reasoning behavior, format inputs using the Qwen ChatML template with the system instruction:
prompt = """<|im_start|>systemأنت مساعد رياضيات بارع. فكّر خطوة بخطوة داخل <think> قبل إعطاء الإجابة النهائية.<|im_end|><|im_start|>user[سؤالك هنا]<|im_end|><|im_start|>assistant<think>"""
أمثلة على الاستدلال والمنطق (Reasoning Showcases)
المثال الأول: مسألة حسابية متعددة الخطوات
السؤال:
اشترى أحمد 3 صناديق من التفاح، يحتوي كل صندوق على 24 تفاحة. إذا تبرع بنصف التفاح لمسجد الحي، وباع ثلث ما تبقى بسعر 1.5 ريال للتفاحة الواحدة، كم ريالاً جنى أحمد؟
في سباق جري، أنهى خالد السباق قبل سعد ولكن بعد فيصل. وأنهى ماجد السباق قبل فيصل. ما هو الترتيب التنازلي للمتسابقين من المركز الأول إلى الأخير؟
استجابة النموذج:
<think>
الخطوة 1: خالد قبل سعد ولكنه بعد فيصل: فيصل > خالد > سعد
الخطوة 2: ماجد قبل فيصل: فيصل > ماجد > خالد > سعد
الخطوة 3: ترتيب التنازلي: فيصل، ماجد، خالد، سعد
</think>
فيصل، ماجد، خالد، سعد
⏳ المثال الثالث: الجبر ومسائل الأعمار
السؤال:
عمر سارة الآن ثلاثة أضعاف عمر أخيها الصغير. بعد 5 سنوات، سيكون مجموع عمريهما 38 سنة. كم عمر سارة الآن؟
استجابة النموذج:
<think>
الخطوة 1: ليكن عمر أخي سارة الصغير هو x. إذن عمر سارة هي 3x.
الخطوة 2: بعد 5 سنوات، سيكون عمر أخي سارة الصغير x + 5، وسارة ستكون 3x + 5.
الخطوة 3: مجموع أعمارهما بعد 5 سنوات هو (x + 5) + (3x + 5) = 4x + 10.
الخطوة 4: نعلم أن 4x + 10 = 38. إذن 4x = 28، و x = 7.
الخطوة 5: عمر سارة الآن هو 3x = 3 * 7 = 21.
</think>
21
💰 المثال الرابع: حساب الخصومات والضرائب
السؤال:
لابتوب سعره الأصلي 4000 ريال، عليه خصم 20%. ثم أضيف عليه ضريبة قيمة مضافة بنسبة 15% بعد الخصم. ما هو السعر النهائي للابتوب؟
استجابة النموذج:
Plaintext
<think>
الخطوة 1: حساب قيمة الخصم: 4000 * 0.20 = 800
الخطوة 2: حساب السعر بعد الخصم: 4000 - 800 = 3200
الخطوة 3: حساب ضريبة القيمة المضافة: 3200 * 0.15 = 480
الخطوة 4: حساب السعر النهائي: 3200 + 480 = 3680
</think>
3680 ريال
Runs of Hatim2221 Fikr-7B-Reasoning on huggingface.co
730
Total runs
-31
24-hour runs
-1.5K
3-day runs
-1.4K
7-day runs
-774
30-day runs
More Information About Fikr-7B-Reasoning huggingface.co Model
Fikr-7B-Reasoning huggingface.co is an AI model on huggingface.co that provides Fikr-7B-Reasoning's model effect (), which can be used instantly with this Hatim2221 Fikr-7B-Reasoning model. huggingface.co supports a free trial of the Fikr-7B-Reasoning model, and also provides paid use of the Fikr-7B-Reasoning. Support call Fikr-7B-Reasoning model through api, including Node.js, Python, http.
Fikr-7B-Reasoning huggingface.co is an online trial and call api platform, which integrates Fikr-7B-Reasoning's modeling effects, including api services, and provides a free online trial of Fikr-7B-Reasoning, you can try Fikr-7B-Reasoning online for free by clicking the link below.
Hatim2221 Fikr-7B-Reasoning online free url in huggingface.co:
Fikr-7B-Reasoning is an open source model from GitHub that offers a free installation service, and any user can find Fikr-7B-Reasoning on GitHub to install. At the same time, huggingface.co provides the effect of Fikr-7B-Reasoning install, users can directly use Fikr-7B-Reasoning installed effect in huggingface.co for debugging and trial. It also supports api for free installation.