| --- |
| license: cc-by-nd-4.0 |
| language: |
| - fa |
| - en |
| - ar |
| tags: |
| - 500x |
| - chat |
| - reasoning |
| - JibayAi |
| - evaluation |
| - benchmark |
| - persian |
| - english |
| - arabic |
| - artificial intelligence |
| task_categories: |
| - text-generation |
| pretty_name: PRCC-BENCH |
| size_categories: |
| - n<1K |
| paperswithcode_id: pprc-bench |
| configs: |
|
|
| - config_name: fa |
| data_files: |
| - split: test |
| path: PPRC-Bench-FA-JibayAi-500x.json |
|
|
| - config_name: en |
| data_files: |
| - split: test |
| path: PPRC-Bench-EN-JibayAi-500x.json |
|
|
| - config_name: ar |
| data_files: |
| - split: test |
| path: PPRC-Bench-AR-JibayAi-500x.json |
|
|
| - config_name: all |
| data_files: |
| - split: test |
| path: PRCC-BENCH-NO.csv |
| --- |
| |
| # PRCC-BENCH |
|
|
| <p align="center"> |
| <img src="https://jibay.ir/uploads/news/2c7b341670a38bf338f35bd224660958.png" width="100%"> |
| </p> |
|
|
| <div align="center"> |
|
|
| # Persian Reasoning, Comprehension, Conversation Benchmark |
|
|
| ### Developed by Jibay Ai |
|
|
| Persian Benchmark for Evaluating Understanding, Reasoning, Question Answering, Information Extraction, Instruction Following and Analysis. |
|
|
| </div> |
|
|
| --- |
|
|
| # About |
|
|
| PRCC-BENCH is a large-scale Persian benchmark developed by **Jibay Ai** to evaluate real-world capabilities of language models. |
|
|
| Unlike traditional benchmarks that mainly focus on memorization or factual recall, PRCC-BENCH evaluates whether a model can truly: |
|
|
| - Understand |
| - Reason |
| - Answer questions |
| - Extract information |
| - Follow instructions |
| - Analyze complex situations |
|
|
| The benchmark aims to provide a practical and realistic measurement of Persian language intelligence. |
|
|
| --- |
|
|
| # Benchmark Information |
|
|
| | Property | Value | |
| |---|---| |
| | Name | PRCC-BENCH | |
| | Organization | Jibay Ai | |
| | Language | Persian | |
| | Version | 1.0.0 | |
| | Total Questions | 500 | |
| | Scoring | +5 Correct | |
| | Incorrect | 0 | |
| | Blank | 0 | |
| | Evaluation | Exact + Semantic | |
|
|
| --- |
|
|
| # Evaluation Categories |
|
|
| ## 1. Comprehension |
|
|
| Measures whether a model truly understands language. |
|
|
| Capabilities: |
|
|
| - Context understanding |
| - Event tracking |
| - Entity understanding |
| - Reference resolution |
| - Passage comprehension |
| - Multi-step understanding |
|
|
| Examples: |
|
|
| - Identify actors |
| - Understand actions |
| - Detect context relationships |
| - Interpret meaning |
|
|
| --- |
|
|
| ## 2. Reasoning |
|
|
| Measures logical thinking ability. |
|
|
| Capabilities: |
|
|
| - Deduction |
| - Cause and effect |
| - Multi-step reasoning |
| - Temporal reasoning |
| - Consistency |
|
|
| Examples: |
|
|
| - Infer hidden conclusions |
| - Connect multiple facts |
| - Solve contextual tasks |
|
|
| --- |
|
|
| ## 3. Question Answering |
|
|
| Measures: |
|
|
| - Accuracy |
| - Completeness |
| - Precision |
| - Context awareness |
|
|
| Task types: |
|
|
| - Direct QA |
| - Contextual QA |
| - Implicit answers |
| - Multi-part questions |
|
|
| --- |
|
|
| ## 4. Information Extraction |
|
|
| Measures: |
|
|
| - Entity extraction |
| - Number extraction |
| - Fact retrieval |
| - Relationship understanding |
| - Context filtering |
|
|
| Examples: |
|
|
| - Extract important details |
| - Ignore irrelevant information |
|
|
| --- |
|
|
| ## 5. Instruction Following |
|
|
| Measures: |
|
|
| - Obeying requirements |
| - Exact formatting |
| - Rule execution |
| - Constraint handling |
|
|
| Examples: |
|
|
| - Structured output |
| - Following conditions |
| - Response discipline |
|
|
| --- |
|
|
| ## 6. Analysis |
|
|
| Measures: |
|
|
| - Deep understanding |
| - Comparison |
| - Problem decomposition |
| - Conclusion generation |
|
|
| Examples: |
|
|
| - Analytical reasoning |
| - Evidence comparison |
|
|
| --- |
|
|
| # Scoring System |
|
|
| | Result | Score | |
| |---|---| |
| | Correct | +5 | |
| | Incorrect | 0 | |
| | Blank | 0 | |
| | Near Match | 0 | |
|
|
| Rule: |
|
|
| Only fully correct responses receive points. |
|
|
| --- |
|
|
| # Judging Modes |
|
|
| ## Exact Match |
|
|
| Normalization includes: |
|
|
| - Spaces |
| - Persian digits |
| - Punctuation |
|
|
| Comparison occurs after normalization. |
|
|
| --- |
|
|
| ## Semantic Match |
|
|
| Answers are accepted only if meaning remains fully equivalent. |
|
|
| --- |
|
|
| ## List Exact |
|
|
| All required information must exist. |
|
|
| --- |
|
|
| ## Order Exact |
|
|
| Required order must remain preserved. |
|
|
| --- |
|
|
| # Intended Use Cases |
|
|
| PRCC-BENCH may be used for: |
|
|
| - AI Evaluation |
| - LLM Benchmarking |
| - Academic Research |
| - Product Testing |
| - Regression Testing |
| - Persian Model Comparison |
| - Quality Assurance |
|
|
| --- |
|
|
| # Philosophy |
|
|
| Understanding > Memorization |
|
|
| Reasoning > Pattern Matching |
|
|
| Execution > Generation |
|
|
| --- |
|
|
| # License |
|
|
| This project is licensed under: |
|
|
| ## CC BY-ND |
|
|
| Creative Commons Attribution — No Derivatives |
|
|
| Permission granted: |
|
|
| ✅ Use |
| ✅ Evaluate |
| ✅ Publish Results |
| ✅ Compare Models |
| ✅ Share Benchmark |
|
|
| Conditions: |
|
|
| - Attribution must remain visible. |
| - Original benchmark identity must remain unchanged. |
|
|
| Additional Project Policy: |
|
|
| - Modifying PRCC-BENCH is not authorized for official distribution. |
| - Publishing altered versions as official releases is not permitted. |
| - Publishing under another benchmark identity is not recognized as an official release. |
| - Official upgrades and official benchmark publications are reserved exclusively for Jibay Ai. |
|
|
| --- |
|
|
| # Citation |
|
|
| If you use PRCC-BENCH: |
|
|
| PRCC-BENCH |
|
|
| Developed by Jibay Ai |
|
|
| --- |
|
|
| # فارسی |
|
|
| ## معرفی |
|
|
| PRCC-BENCH یک بنچمارک زبان فارسی توسعهیافته توسط **Jibay Ai** است. |
|
|
| هدف این پروژه ارزیابی توانایی واقعی مدلهای هوش مصنوعی در انجام وظایف زبانی و تحلیلی است. |
|
|
| این بنچمارک بر موارد زیر تمرکز دارد: |
|
|
| - درک |
| - استدلال |
| - سوال و جواب |
| - استخراج اطلاعات |
| - پیروی از دستورات |
| - تحلیل |
|
|
| --- |
|
|
| # مشخصات |
|
|
| | ویژگی | مقدار | |
| |---|---| |
| | نام | PRCC-BENCH | |
| | توسعهدهنده | Jibay Ai | |
| | زبان | فارسی | |
| | نسخه | 1.0.0 | |
| | تعداد سوال | 500 | |
| | امتیاز پاسخ صحیح | 5 | |
| | پاسخ غلط | 0 | |
| | ارزیابی | دقیق + معنایی | |
|
|
| --- |
|
|
| # بخشهای ارزیابی |
|
|
| ## درک |
|
|
| سنجش توانایی مدل در: |
|
|
| - فهم متن |
| - دنبال کردن رویداد |
| - تشخیص ارجاع |
| - درک رابطهها |
|
|
| --- |
|
|
| ## استدلال |
|
|
| بررسی: |
|
|
| - منطق |
| - علت و معلول |
| - تحلیل چندمرحلهای |
| - نتیجهگیری |
|
|
| --- |
|
|
| ## سوال و جواب |
|
|
| ارزیابی: |
|
|
| - دقت |
| - کامل بودن |
| - وابستگی به متن |
|
|
| --- |
|
|
| ## استخراج اطلاعات |
|
|
| بررسی: |
|
|
| - استخراج موجودیت |
| - استخراج عدد |
| - تشخیص اطلاعات مهم |
|
|
| --- |
|
|
| ## پیروی از دستورات |
|
|
| بررسی: |
|
|
| - رعایت محدودیت |
| - اجرای دقیق |
| - حفظ ساختار |
|
|
| --- |
|
|
| ## تحلیل |
|
|
| بررسی: |
|
|
| - تجزیه مسئله |
| - مقایسه |
| - استنتاج |
|
|
| --- |
|
|
| # سیستم امتیازدهی |
|
|
| پاسخ صحیح: ۵ |
|
|
| سایر حالتها: ۰ |
|
|
| فقط پاسخ کاملاً صحیح امتیاز دریافت میکند. |
|
|
| --- |
|
|
| # مجوز |
|
|
| این پروژه تحت مجوز: |
|
|
| ## CC BY-ND |
|
|
| منتشر شده است. |
|
|
| همه افراد و سازمانها مجاز هستند: |
|
|
| ✅ استفاده کنند |
| ✅ ارزیابی انجام دهند |
| ✅ نتایج منتشر کنند |
| ✅ مقایسه انجام دهند |
|
|
| اما: |
|
|
| ❌ هیچ شخص یا سازمانی اجازه تغییر یا ارتقاء نسخه رسمی را ندارد. |
|
|
| ❌ انتشار نسخه تغییر یافته به عنوان نسخه رسمی مجاز نیست. |
|
|
| ❌ انتشار با نام دیگر به عنوان نسخه رسمی مجاز نیست. |
|
|
| ✅ حق توسعه نسخه رسمی، ارتقاء و انتشارات رسمی فقط متعلق به Jibay Ai است. |
|
|
| --- |
|
|
| PRCC-BENCH |
|
|
| by jibay.ir |
