README.md

metadata

language:
  - en
license: unknown
tags:
  - frankenmerge
  - 103b
pipeline_tag: conversational
model-index:
  - name: BigWeave-v16-103b
    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: 65.87
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          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: 87.61
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          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: 73.22
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          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: 63.81
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          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: 80.43
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          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: 61.18
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v16-103b
          name: Open LLM Leaderboard

BigWeave v16 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. The layers are duplicated such that each group consists of consecutive layers with a two-layer overlap (i.e. larger groups than in v15).

Merge configuration:

slices:
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [0,11]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [9,13]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [11,15]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [13,17]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [15,23]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [21,25]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [23,49]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [47,51]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [49,53]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [51,55]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [53,57]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [55,59]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [57,61]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [59,63]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [61,65]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [63,67]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [65,69]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [67,71]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [69,73]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [71,75]
  - sources:
    - model: 152334H/miqu-1-70b-sf
      layer_range: [73,80]
merge_method: passthrough
dtype: float16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	72.02
AI2 Reasoning Challenge (25-Shot)	65.87
HellaSwag (10-Shot)	87.61
MMLU (5-Shot)	73.22
TruthfulQA (0-shot)	63.81
Winogrande (5-shot)	80.43
GSM8k (5-shot)	61.18