README.md

metadata

license: mit
library_name: transformers
model-index:
  - name: Arcanum-12b
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 29.07
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 31.88
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 10.27
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 9.4
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 13.53
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 28.74
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Xclbr7/Arcanum-12b
          name: Open LLM Leaderboard

Arcanum-12b 🧙‍♂️

Arcanum-12b is a merged large language model created by combining TheDrummer/Rocinante-12B-v1.1 and MarinaraSpaghetti/NemoMix-Unleashed-12B using a novel merging technique.

Model Details 📊

• Developed by: Xclbr7
• Model type: Causal Language Model
• Language(s): English (primarily), may support other languages
• License: MIT
• Repository: https://huggingface.co/Xclbr7/Arcanum-12b

Model Architecture 🏗️

• Base model: MarinaraSpaghetti/NemoMix-Unleashed-12B
• Parameter count: ~12 billion
• Architecture specifics: Transformer-based language model

Training & Merging 🔄

Arcanum-12b was created by merging two existing 12B models:

1. TheDrummer/Rocinante-12B-v1.1

   • Density parameters: [1, 0.8, 0.6]
   • Weight: 0.7

2. MarinaraSpaghetti/NemoMix-Unleashed-12B

   • Density parameters: [0.5, 0.7, 0.9]
   • Weight: 0.8

Merging method: Ties Additional parameters:

• Normalization: True
• Int8 mask: True
• Data type: float16

Intended Use 🎯

Conversation with different personas.

Performance and Limitations ⚖️

Not tested yet.

Ethical Considerations 🤔

As a merged model based on existing language models, Arcanum-12b may inherit biases and limitations from its parent models. Users should be aware of potential biases in generated content and use the model responsibly.

Acknowledgments 🙏

We acknowledge the contributions of the original model creators:

• TheDrummer for Rocinante-12B-v1.1
• MarinaraSpaghetti for NemoMix-Unleashed-12B

Their work formed the foundation for Arcanum-12b.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	20.48
IFEval (0-Shot)	29.07
BBH (3-Shot)	31.88
MATH Lvl 5 (4-Shot)	10.27
GPQA (0-shot)	9.40
MuSR (0-shot)	13.53
MMLU-PRO (5-shot)	28.74