FreedomIntelligence/AceGPT-v2-70B-Chat

Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:8kPublished:Jun 17, 2024License:apache-2.0Architecture:Transformer Open Weights Warm

AceGPT-v2-70B-Chat is a 70 billion parameter, fully fine-tuned generative text model developed by King Abdullah University of Science and Technology (KAUST), the Chinese University of Hong Kong, Shenzhen (CUHKSZ), and the Shenzhen Research Institute of Big Data (SRIBD). Optimized for dialogue applications, it demonstrates superior performance among open-source Arabic dialogue models and comparable satisfaction levels to closed-source models like ChatGPT in Arabic. This model excels in Arabic language understanding and generation, making it ideal for conversational AI in Arabic-speaking contexts.

Loading preview...

AceGPT-v2-70B-Chat: Advanced Arabic Dialogue Model

AceGPT-v2-70B-Chat is a 70 billion parameter, fully fine-tuned generative text model, part of the AceGPT family developed by researchers from KAUST, CUHKSZ, and SRIB. This version is specifically optimized for dialogue applications, building upon the AceGPT-v2-70B pre-trained model.

Key Capabilities & Differentiators

  • Arabic Language Specialization: Designed with a primary focus on the Arabic language domain, offering robust performance in Arabic dialogue.
  • Superior Arabic Benchmarks: Outperforms all currently available open-source Arabic dialogue models in multiple benchmark tests, including ArabicMMLU, EXAMS, ACVA, Arabic BoolQ, and Arabic ARC-C.
  • Comparable to Closed-Source Models: Achieves satisfaction levels comparable to models like ChatGPT in human evaluations for Arabic language tasks.
  • Dialogue Optimization: "-chat" variants are specifically engineered for conversational AI and dialogue applications.
  • Model Family: Part of a larger family ranging from 7B to 70B parameters, with both base and chat-optimized categories.

Performance Highlights

AceGPT-v2-70B-Chat demonstrates strong performance across various benchmarks, notably achieving 72.50% on ArabicMMLU, 82.66% on Arabic BoolQ, and 85.53% on Arabic ARC-C. Its overall average score of 73.99% positions it competitively, even against models like GPT-4 in specific Arabic metrics.

Ideal Use Cases

  • Arabic Conversational AI: Building chatbots, virtual assistants, and dialogue systems for Arabic-speaking users.
  • Arabic Content Generation: Generating high-quality, contextually relevant text in Arabic.
  • Research & Development: As a strong baseline or component for further research in Arabic NLP and LLMs.

Finetunes
3 models