README.md

metadata

language:
  - en
license: unknown
tags:
  - frankenmerge
  - 103b
pipeline_tag: conversational
model-index:
  - name: BigWeave-v15-103b
    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: 69.71
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          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: 86.41
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          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: 71.25
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          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: 66.1
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          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: 80.35
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          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: 56.18
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v15-103b
          name: Open LLM Leaderboard

BigWeave v15 103b

The BigWeave models aim to experimentally identify merge settings for increasing model performance. The version number merely tracks various attempts and is not a quality indicator. Only results demonstrating good performance are retained and shared.

Prompting Format

Mistral, Vicuna and Alpaca.

Merge process

This is a self-merge of 152334H/miqu-1-70b-sf. By conducting exl2 measurements, we identify the most relevant layers. These layers are then duplicated in pairs to ensure overlaps.

Merge configuration:

slices:
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [0,3]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [1,5]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [3,7]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [5,9]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [7,18]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [16,21]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [19,27]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [25,30]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [28,32]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [30,34]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [32,36]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [34,38]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [36,40]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [38,42]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [40,44]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [42,46]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [44,48]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [46,51]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [49,77]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [75,79]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [77,80]
merge_method: passthrough
dtype: float16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	71.67
AI2 Reasoning Challenge (25-Shot)	69.71
HellaSwag (10-Shot)	86.41
MMLU (5-Shot)	71.25
TruthfulQA (0-shot)	66.10
Winogrande (5-shot)	80.35
GSM8k (5-shot)	56.18