metadata

language:
  - en
license: llama2
tags:
  - dare
  - super mario merge
  - pytorch
  - llama
  - llama-2
  - merge
pipeline_tag: text-generation
inference: false
model-index:
  - name: llama2-megamerge-dare-13b-v2
    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: 59.39
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          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: 80.93
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          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: 55.26
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          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: 47.27
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          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: 75.53
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          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: 29.26
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=martyn/llama2-megamerge-dare-13b-v2
          name: Open LLM Leaderboard

llama 13b mega merge v2

The following were merged with DARE using https://github.com/martyn/safetensors-merge-supermario

Mergelist

meta-llama/Llama-2-13b-hf
ajibawa-2023/Code-13B
ajibawa-2023/Python-Code-13B
meta-math/MetaMath-13B-V1.0
rombodawg/LosslessMegaCoder-llama2-13b-mini
NousResearch/Nous-Hermes-Llama2-13b
allenai/digital-socrates-13b
migtissera/Synthia-13B
Gryphe/MythoLogic-L2-13b
allenai/tulu-2-dpo-13b
FPHam/Free_Sydney_13b_HF
FPHam/Free_Sydney_V2_13b_HF
FPHam/Sydney_Overthinker_13b_HF
KoboldAI/LLaMA2-13B-Psyfighter2
Undi95/Unholy-v1-12L-13B
athirdpath/Eileithyia-13B
athirdpath/Orca-2-13b-Alpaca-Uncensored

Merge command

python hf_merge.py mergelist.txt 13b-merge-v2 -p 0.11 -lambda 2.1

Notes

seems to generalize instruct styles
p and lambda are still a guess

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	57.94
AI2 Reasoning Challenge (25-Shot)	59.39
HellaSwag (10-Shot)	80.93
MMLU (5-Shot)	55.26
TruthfulQA (0-shot)	47.27
Winogrande (5-shot)	75.53
GSM8k (5-shot)	29.26