README.md

metadata

license: apache-2.0
tags:
  - merge
metrics:
  - accuracy
base_model: Mihaiii/Metis-0.3
inference: false
license_name: apache-2.0
model-index:
  - name: Metis-0.3-merged
    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: 62.2
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          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: 84
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          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: 62.65
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          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: 59.24
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          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: 78.14
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          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: 21.83
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Mihaiii/Metis-0.3-merged
          name: Open LLM Leaderboard

This is a merge between Metis-0.3 and Metis-0.1 having Metis-0.1 as base. It was done using mergekit.

It works well with long system prompts.

It isn't generic in a sense that it shouldn't be used for story telling, for example, but only for reasoning and text comprehension.

This model is trained on a private dataset.

Prompt Format:

<|system|>
{system_message} </s>
<|user|>
{prompt} </s>
<|assistant|>

Merge config:

slices:
  - sources:
      - model: Mihaiii/Metis-0.3
        layer_range: [0, 32]
      - model: Mihaiii/Metis-0.1
        layer_range: [0, 32]
merge_method: slerp
base_model: Mihaiii/Metis-0.1
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
dtype: bfloat16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	61.34
AI2 Reasoning Challenge (25-Shot)	62.20
HellaSwag (10-Shot)	84.00
MMLU (5-Shot)	62.65
TruthfulQA (0-shot)	59.24
Winogrande (5-shot)	78.14
GSM8k (5-shot)	21.83