metadata

language:
  - en
  - zh
license: other
datasets:
  - Locutusque/UltraTextbooks-2.0
license_name: tongyi-qianwen-research
license_link: https://huggingface.co/Qwen/Qwen1.5-0.5B/blob/main/LICENSE
inference:
  parameters:
    do_sample: true
    temperature: 0.8
    top_p: 0.95
    top_k: 40
    max_new_tokens: 250
    repetition_penalty: 1.1
model-index:
  - name: tau-1.8B
    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: 37.2
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          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: 60.26
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          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: 45.96
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          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: 39.72
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          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: 61.09
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          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: 30.17
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=M4-ai/tau-1.8B
          name: Open LLM Leaderboard

tau-1.8B

Model Details

Model Name: tau-1.8B
Base Model: Qwen1.5-1.8B
Dataset: UltraTextbooks-2.0
Model Size: 1.8B parameters
Model Type: Language Model
Training Procedure: Further pre-training of Qwen1.5-1.8B on UltraTextbooks-2.0.

Model Use

tau-1.8B is designed to be a general-purpose language model with enhanced capabilities in the domains of machine learning, mathematics, and coding. It can be used for a wide range of natural language processing tasks, such as:

Educational question answering
Text summarization
Content generation for educational purposes
Code understanding and generation
Mathematical problem solving

The model's exposure to the diverse content in the UltraTextbooks-2.0 dataset makes it particularly well-suited for applications in educational technology and research.

Training Data

tau-1.8B was further pre-trained on the UltraTextbooks-2.0 dataset, which is an expanded version of the original UltraTextbooks dataset. UltraTextbooks-2.0 incorporates additional high-quality synthetic and human-written textbooks from various sources on the Hugging Face platform, with a focus on increasing the diversity of content in the domains of machine learning, mathematics, and coding.

For more details on the dataset, please refer to the UltraTextbooks-2.0 Dataset Card.

Performance and Limitations

Refer to Evaluation for evaluations. It is essential to note that the model may still exhibit biases or inaccuracies present in the training data. Users are encouraged to critically evaluate the model's outputs and report any issues to facilitate continuous improvement.

Environmental Impact

The training of tau-1.8B required computational resources that contribute to the model's overall environmental impact. However, efforts were made to optimize the training process and minimize the carbon footprint.

Ethical Considerations

tau-1.8B was trained on a diverse dataset that may contain biases and inaccuracies. Users should be aware of these potential limitations and use the model responsibly. The model should not be used for tasks that could cause harm or discriminate against individuals or groups.

Evaluation

Coming Soon

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	45.73
AI2 Reasoning Challenge (25-Shot)	37.20
HellaSwag (10-Shot)	60.26
MMLU (5-Shot)	45.96
TruthfulQA (0-shot)	39.72
Winogrande (5-shot)	61.09
GSM8k (5-shot)	30.17