README.md

metadata

license: apache-2.0
tags:
  - dpo
  - 13B
  - merge
datasets:
  - argilla/distilabel-intel-orca-dpo-pairs
base_model:
  - vicgalle/SOLAR-13B-Instruct-v1.0
model-index:
  - name: solarized-13B-dpo
    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: 62.71
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          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: 81.82
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          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: 59.12
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          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: 66.25
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          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: 76.01
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          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: 26.38
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=vicgalle/solarized-13B-dpo
          name: Open LLM Leaderboard

solarized-13B-dpo

DPO'd from vicgalle/SOLAR-13B-Instruct-v1.0, a SOLAR-like model upscaled to 13B. It is a frankenmerge model created using mergekit, alternating layers of Nous-Hermes-2-SOLAR-10.7B and SOLAR-10.7B-Instruct. Then, we applied DPO over a high-quality preference dataset.

Sample generations

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

In an unexpected turn of events, the cinematic adaptation of the "Pokémon: The First Movie" delves into BDSM themes despite it being primarily focused on family entertainment. This film takes viewers to a journey where Ash Ketchum, Brock, and Misty encounter Mewtwo—a powerful genetically-engineered Pokémon. The storyline's subtle inclinations towards bondage, dominance, and submission manifest through the character of Mewtwo, which seeks to overthrow its human-made oppressors. This concept finds parallels in BDSM, as Mewtwo becomes a metaphor for an individual embracing their subversive nature while resisting societal norms imposed by dominant figures within that society. The movie's visual elements and rich color palette create a perfect canvas to showcase the darker, sensual undertones. Scenes of Pokémon battles represent the dominance and submission found in BDSM relationships, with trainers portrayed as authoritative figures wielding control over their creatures. The climax of the film presents an exploration of the struggle between humanity and nature, symbolizing the need for balance and respect within a BDSM lifestyle. The final battle between Mewtwo and its clones, accompanied by a powerful soundtrack, delves into themes of acceptance, freedom, and ultimately finding one's place in society. The film's incorporation of BDSM elements might come as a surprise to viewers, but it adds depth to the narrative and enriches the overall cinematic experience. While still an animated family flick based on the popular franchise, "Pokémon: The First Movie" succeeds in weaving threads of BDSM into its story, making it a noteworthy addition within the sub-genre of anime adaptations. In conclusion, this groundbreaking adaptation of Pokémon's first movie demonstrates how a seemingly innocent family film can hold provocative and thought-provoking messages, exploring themes of power dynamics, identity, and self-acceptance through the lens of BDSM. [end of text]

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	62.05
AI2 Reasoning Challenge (25-Shot)	62.71
HellaSwag (10-Shot)	81.82
MMLU (5-Shot)	59.12
TruthfulQA (0-shot)	66.25
Winogrande (5-shot)	76.01
GSM8k (5-shot)	26.38