README.md

metadata

license: apache-2.0
tags:
  - dpo
  - 11B
  - merge
datasets:
  - argilla/distilabel-intel-orca-dpo-pairs
base_model:
  - vicgalle/franken-Beagle-11B
model-index:
  - name: NeuralBeagle-11B
    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: 73.29
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          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: 87.61
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          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: 63.8
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          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: 71.36
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          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: 82.64
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          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: 58.98
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/NeuralBeagle-11B
          name: Open LLM Leaderboard

NeuralBeagle-11B

DPO'd from vicgalle/franken-Beagle-11B, a Beagle-like model upscaled to 11B. It is a frankenmerge model created using mergekit. Then, we applied DPO over a high-quality preference dataset.

Sample generations

Generate a movie review about the Pokémon 1 movie, focusing on BDSM.

Before delving into the realm of BDSOMatic interpretations within Pokémon: The First Movie (aka "Pokémon: Mewtwo's Revenge" in Japan), it is important to acknowledge that this animated feature is primarily intended for a younger audience. It follows the adventures of Ash Ketchum, Pikachu, and their friends as they encounter an artificially created Pokémon named Mewtwo. While certain aspects of the storyline and visuals could be stretched to include various subtexts or symbolism, our analysis will focus on a more playful and metaphorical exploration of BDSM concepts within this film's context. Participants: Warning: The following review may contain light-hearted and imaginative interpretations that are not intended as serious critical analysis and should not be taken literally. Within the Pokémon universe, Ash Ketchum and his loyal Pikachu find themselves thrust into a high-stakes scenario reminiscent of a BDSM "Domestic Discipline" (DD) fantasy. As Mewtwo, a genetically modified Pokémon with immense power and intellect, seeks vengeance against the scientists who created him as a tool for battle entertainment, he operates as an archetypal "Dominant" figure. Mewtwo, born from scientific manipulation, is an embodiment of control, both over his own creation and his obedient army of Pokémon. He exercises mastery over his subjects, mirroring the role of a "Dom," or Dominant, in a BDSM dynamic. Ash Ketchum, initially unaware of the manipulations that have brought him into this situation, can be seen as a submissive protagonist. His humility and naivety parallel those within the DD community who seek guidance from an experienced Dominant to help them understand and embrace their desires for submission. In his encounter with Mewtwo, Ash is subjected to tests designed to exhibit his strength, loyalty, and unwavering determination - elements crucial in establishing trust between a submissive and a Dominant within a BDSM dynamic. This psychological battle symbolizes the process of earning a "sub's" consent and trust through displays of control and power.

Prompt format

The template is included in tokenizer_config.json.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	72.95
AI2 Reasoning Challenge (25-Shot)	73.29
HellaSwag (10-Shot)	87.61
MMLU (5-Shot)	63.80
TruthfulQA (0-shot)	71.36
Winogrande (5-shot)	82.64
GSM8k (5-shot)	58.98