README.md

metadata

language:
  - en
license: apache-2.0
model-index:
  - name: flux-base-optimized
    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: 65.44
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          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: 81.74
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          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: 59.74
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          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: 50.02
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          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: 77.74
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          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: 44.66
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=chanwit/flux-base-optimized
          name: Open LLM Leaderboard

Flux-Base-Optimized

flux-base-optimized is the base model for finetuning the series of flux-7b models. It is hierarchical SLERP merged from the following models

• mistralai/Mistral-7B-v0.1 (Apache 2.0)
• teknium/OpenHermes-2.5-Mistral-7B (Apache 2.0)
• Intel/neural-chat-7b-v3-3 (Apache 2.0)
• meta-math/MetaMath-Mistral-7B (Apache 2.0)
• openchat/openchat-3.5-0106 was openchat/openchat-3.5-1210 (Apache 2.0)

Here's how we did the hierarchical SLERP merge.

                [flux-base-optimized]
                         ↑
                         |
               [stage-1]-+-[openchat]
                   ↑
                   |
         [stage-0]-+-[meta-math]
             ↑
             |
[openhermes]-+-[neural-chat]

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	63.22
AI2 Reasoning Challenge (25-Shot)	65.44
HellaSwag (10-Shot)	81.74
MMLU (5-Shot)	59.74
TruthfulQA (0-shot)	50.02
Winogrande (5-shot)	77.74
GSM8k (5-shot)	44.66