README.md

metadata

language:
  - en
license: llama2
tags:
  - Xwin
  - Euryale 1.3
  - frankenmerge
  - 90b
pipeline_tag: conversational
model-index:
  - name: BigWeave-v6-90b
    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: 65.36
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          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: 87.21
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          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: 68.04
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          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: 57.96
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          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: 81.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          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: 44.58
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=llmixer/BigWeave-v6-90b
          name: Open LLM Leaderboard

BigWeave v6 90B

A Goliath-120b style frankenmerge of Xwin-LM-70b-v0.1 and Euryale-1.3-70b. The goal is to find other merge combinations that work well.

The version number is for me to keep track of the merges, only results that seem to work reasonably well are kept/published.

Prompting Format

Vicuna and Alpaca.

Merge process

The models used in the merge are Xwin-LM-70b-v0.1 and Euryale-1.3-70b.

The layer mix:

- range 0, 12
  Xwin
- range 9, 14
  Euryale
- range 12, 62
  Xwin
- range 54, 71
  Euryale
- range 62, 80
  Xwin

Acknowledgements

@Xwin-LM For creating Xwin

@Sao10K For creating Euryale

@alpindale For creating the original Goliath

@chargoddard For developing mergekit.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	67.47
AI2 Reasoning Challenge (25-Shot)	65.36
HellaSwag (10-Shot)	87.21
MMLU (5-Shot)	68.04
TruthfulQA (0-shot)	57.96
Winogrande (5-shot)	81.69
GSM8k (5-shot)	44.58