README.md

metadata

license: other
library_name: transformers
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
base_model: wandb/gemma-7b-zephyr-sft
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
model-index:
  - name: gemma-7b-zephyr-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: 60.84
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-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: 80.44
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-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: 60.6
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-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: 42.48
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-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: 75.37
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-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: 49.96
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=tcapelle/gemma-7b-zephyr-dpo
          name: Open LLM Leaderboard

Gemma 7B Zephyr DPO

The Zephyr DPO recipe applied on top of SFT finetuned Gemma 7B

Model description

• Model type: A 8.5B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
• Language(s) (NLP): Primarily English
• Finetuned from model: wandb/gemma-7b-zephyr-sft

Recipe

We trained using the DPO script in alignment handbook recipe and logging to W&B

Visit the W&B workspace here

License

This model has the same license as the original Gemma model collection

Compute provided by Lambda Labs - 8xA100 80GB node

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	61.62
AI2 Reasoning Challenge (25-Shot)	60.84
HellaSwag (10-Shot)	80.44
MMLU (5-Shot)	60.60
TruthfulQA (0-shot)	42.48
Winogrande (5-shot)	75.37
GSM8k (5-shot)	49.96