metadata

language:
  - ar
license: apache-2.0
model-index:
  - name: AceGPT-7B
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 53.58
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 77.54
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 38.75
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 72.77
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 11.14
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=FreedomIntelligence/AceGPT-7B
          name: Open LLM Leaderboard

AceGPT

AceGPT is a fully fine-tuned generative text model collection based on LlaMA2, particularly in the
Arabic language domain. This is the repository for the 7B pretrained model.

Model Details

We have released the AceGPT family of large language models, which is a collection of fully fine-tuned generative text models based on LlaMA2, ranging from 7B to 13B parameters. Our models include two main categories: AceGPT and AceGPT-chat. AceGPT-chat is an optimized version specifically designed for dialogue applications. It is worth mentioning that our models have demonstrated superior performance compared to all currently available open-source Arabic dialogue models in multiple benchmark tests. Furthermore, in our human evaluations, our models have shown comparable satisfaction levels to some closed-source models, such as ChatGPT, in the Arabic language.

Model Developers

We are from the School of Data Science, the Chinese University of Hong Kong, Shenzhen (CUHKSZ), the Shenzhen Research Institute of Big Data (SRIBD), and the King Abdullah University of Science and Technology (KAUST).

Variations

AceGPT families come in a range of parameter sizes —— 7B and 13B, each size of model has a base category and a -chat category.

Input

Models input text only.

Output

Models output text only.

Model Evaluation Results

Experiments on Arabic MMLU and EXAMs. ' AverageBest ', ' STEM ', ' Humanities ', ' Social Sciences ' and ' Others (Business, Health, Misc)' belong to Arabic MMLU. Best performance is in bold and the second best is underlined.

Model	Average	STEM	Humanities	Social Sciences	Others (Business, Health, Misc)	EXAMs
Bloomz Muennighoff et al. (2022)	30.95	32.32	26.71	35.85	28.95	33.89
Llama2-7B	28.81	28.48	26.68	29.88	30.18	23.48
Llama2-13B	31.25	31.06	27.11	35.5	31.35	25.45
Jais-13B-base	30.01	27.85	25.42	39.7	27.06	35.67
AceGPT-7B-base	30.36	26.63	28.17	35.15	31.5	31.96
AceGPT-13B-base	37.26	35.16	30.3	47.34	36.25	36.63
ChatGPT	46.07	44.17	35.33	61.26	43.52	45.63

Samples

Arabic MMLU (5-shot)

فيما يلي أسئلة الاختيار من متعدد (مع الإجابات) حول جبر تجريدي
سؤال: العثور على جميع قيم c في Z_3 بحيث يكون Z_3 [x]/(x^2+c) حقلًا.
A. 0
B. 1
C. 2
D. 3
إجابة: B

سؤال: العبارة 1 | كل عنصر من مجموعة يولد مجموعة دورية من المجموعة. العبارة 2 | المجموعة المتناظرة S_10 لديها 10 عناصر.
A. صحيح، صحيح
B. خطأ، خطأ
C. صحيح، خطأ
D. خطأ، صحيح
إجابة: C

سؤال: البيان 1| كل وظيفة من مجموعة محدودة على نفسها يجب أن تكون واحدة لكل مجموعة. البيان 2 | كل فرع فرعي لمجموعة أبيلية هو أبيلي.
A. صحيح, صحيح
B. خاطئ, خاطئ
C. صحيح, خاطئ
D. خاطئ, صحيح\nإجابة: A

سؤال: اعثر على خاصية الحلقة 2Z.
A. 0
B. 3
C. 12
D. 30
إجابة: A

سؤال: ما هو الدرجة للامتداد الميداني الناتج من Q(sqrt(2), sqrt(3), sqrt(18)) على Q؟
A. 0
B. 4
C. 2
D. 6
إجابة:",

You can get more detail at https://github.com/FreedomIntelligence/AceGPT/tree/main

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	49.47
AI2 Reasoning Challenge (25-Shot)	53.58
HellaSwag (10-Shot)	77.54
MMLU (5-Shot)	43.00
TruthfulQA (0-shot)	38.75
Winogrande (5-shot)	72.77
GSM8k (5-shot)	11.14