README.md

metadata

language:
  - en
license: mit
base_model:
  - princeton-nlp/Llama-3-Base-8B-SFT
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
model-index:
  - name: Llama3
    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: 33.21
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          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: 26.71
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          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: 5.51
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          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: 8.05
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          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: 7.1
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          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: 24.02
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=sabersaleh/Llama3
          name: Open LLM Leaderboard

This is an aligned model based on princeton-nlp/Llama-3-Base-8B-SFT. The alignment loss has two parts, preference loss and a regularization term. The alignment loss does not require the assumption of an existing underlying reward model. Moreover, the distribution's ``softness" is adjustable via the softmax exponent, an algorithm parameter.

More information will be added later.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	17.43
IFEval (0-Shot)	33.21
BBH (3-Shot)	26.71
MATH Lvl 5 (4-Shot)	5.51
GPQA (0-shot)	8.05
MuSR (0-shot)	7.10
MMLU-PRO (5-shot)	24.02