README.md

metadata

license: apache-2.0
model-index:
  - name: mera-mix-4x7B
    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: 72.95
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          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: 89.17
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          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: 64.44
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          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: 77.17
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          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: 85.64
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          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: 66.11
            name: accuracy
        source:
          url: >-
            https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=meraGPT/mera-mix-4x7B
          name: Open LLM Leaderboard

Model mera-mix-4x7B

This is a mixture of experts (MoE) model that is half as large (4 experts instead of 8) as the Mixtral-8x7B while been comparable to it across different benchmarks. You can use it as a drop in replacement for your Mixtral-8x7B and get much faster inference.

mera-mix-4x7B achieves the score of 75.91 on the OpenLLM Eval and compares well with 72.7 by Mixtral-8x7B and 74.46 by Mixtral-8x22B.

You can try the model with the Mera Mixture Chat.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	75.91
AI2 Reasoning Challenge (25-Shot)	72.95
HellaSwag (10-Shot)	89.17
MMLU (5-Shot)	64.44
TruthfulQA (0-shot)	77.17
Winogrande (5-shot)	85.64
GSM8k (5-shot)	66.11

In addition, to the official Open LLM Leaderboard, the results on OpenLLM Eval have been validated by others as well (76.59).

Our own initial eval is available here (76.37).