README.md

metadata

language:
  - en
license: other
library_name: transformers
tags:
  - mergekit
  - merge
  - OG_finetune_merge
base_model:
  - Qwen/Qwen2-72B-Instruct
  - alpindale/magnum-72b-v1
model-index:
  - name: MG-FinalMix-72B
    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: 80.14
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          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: 57.5
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          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: 33.61
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          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: 18.01
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          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: 21.22
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          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: 49.19
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Undi95/MG-FinalMix-72B
          name: Open LLM Leaderboard

WIP of retouched alpindale/magnum-72b-v1 but I will not use "Magnum" in the name. Call it FinalMix!

Found some issues, trying to fix them for my own usage and adding more RP data with merging.

You can do your own quantized files with the imatrix.dat file done with "wiki.train.raw".

Credits to Alpin and the gang for magnum-72b-v1, and Ikari for his datasets.

Prompt template ChatML

<|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
{output}<|im_end|>

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	43.28
IFEval (0-Shot)	80.14
BBH (3-Shot)	57.50
MATH Lvl 5 (4-Shot)	33.61
GPQA (0-shot)	18.01
MuSR (0-shot)	21.22
MMLU-PRO (5-shot)	49.19