README.md

metadata

language:
  - en
license: llama3.1
tags:
  - moe
model-index:
  - name: L3.1-ClaudeMaid-4x8B
    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: 66.96
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          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: 29.44
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          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: 12.84
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          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: 5.48
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          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: 13.75
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          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: 28.67
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=xxx777xxxASD/L3.1-ClaudeMaid-4x8B
          name: Open LLM Leaderboard

GGUF

Experimental RP-oriented MoE

Llama 3 ClaudeMaid v1.0 4x8B

base_model: NeverSleep_Lumimaid-v0.2-8B
gate_mode: random
dtype: bfloat16
experts_per_token: 2
experts:
  - source_model: aifeifei798_DarkIdol-Llama-3.1-8B-Instruct-1.0-Uncensored
  - source_model: Nitral-AI_Sekhmet_Bet-L3.1-8B-v0.2
  - source_model: NeverSleep_Lumimaid-v0.2-8B
  - source_model: Undi95_Meta-Llama-3.1-8B-Claude-bf16

Usage

It seems like 1.71 koboldcpp can't run GGUFs of llama-3.1 MoE models yet, or perhaps im just dumb and messed something up. If anyone has similar problem - run the model directly from llama.cpp, here's simple open source GUI(Windows) you can use if the console is your worst enemy

UPDATE 28.07.2024 Try this koboldcpp version

Models used

• NeverSleep/Lumimaid-v0.2-8B
• Undi95/Meta-Llama-3.1-8B-Claude
• Nitral-AI/Sekhmet_Bet-L3.1-8B-v0.2
• aifeifei798/DarkIdol-Llama-3.1-8B-Instruct-1.0-Uncensored

Prompt format: Llama 3

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	26.19
IFEval (0-Shot)	66.96
BBH (3-Shot)	29.44
MATH Lvl 5 (4-Shot)	12.84
GPQA (0-shot)	5.48
MuSR (0-shot)	13.75
MMLU-PRO (5-shot)	28.67