README.md

metadata

license: apache-2.0
library_name: transformers
tags:
  - mergekit
  - merge
  - not-for-all-audiences
base_model:
  - bamec66557/VICIOUS_MESH-12B-BETA
  - bamec66557/VICIOUS_MESH-12B-OMEGA
model-index:
  - name: Mistral-Nemo-VICIOUS_MESH-12B-2407
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 67.21
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 31.36
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 12.08
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 8.84
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 14.34
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 29.76
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=bamec66557/Mistral-Nemo-VICIOUS_MESH-12B-2407
          name: Open LLM Leaderboard

merge

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• bamec66557/VICIOUS_MESH-12B-BETA
• bamec66557/VICIOUS_MESH-12B-OMEGA

Configuration

The following YAML configuration was used to produce this model:

base_model: bamec66557/VICIOUS_MESH-12B-OMEGA
dtype: bfloat16
merge_method: slerp
tokenizer_source: base

# Slices Configuration (Layer-Specific Merging)
slices:
  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [0, 5]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [0, 5]
    parameters:
      t:
        - name: self_attn
          value: [0.85, 0.88, 0.91, 0.94, 0.97]
        - name: mlp
          value: [0.9, 0.92, 0.95, 0.98, 1.0]
        - name: layer_norm
          value: [0.75, 0.78, 0.81, 0.84, 0.87]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [5, 10]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [5, 10]
    parameters:
      t:
        - name: self_attn
          value: [0.8, 0.83, 0.86, 0.89, 0.92]
        - name: mlp
          value: [0.88, 0.91, 0.94, 0.97, 1.0]
        - name: layer_norm
          value: [0.7, 0.73, 0.76, 0.79, 0.82]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [10, 15]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [10, 15]
    parameters:
      t:
        - name: self_attn
          value: [0.75, 0.78, 0.81, 0.84, 0.87]
        - name: mlp
          value: [0.85, 0.88, 0.91, 0.94, 0.97]
        - name: layer_norm
          value: [0.65, 0.68, 0.71, 0.74, 0.77]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [15, 20]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [15, 20]
    parameters:
      t:
        - name: self_attn
          value: [0.72, 0.75, 0.78, 0.81, 0.84]
        - name: mlp
          value: [0.8, 0.83, 0.86, 0.89, 0.92]
        - name: layer_norm
          value: [0.6, 0.63, 0.66, 0.69, 0.72]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [20, 25]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [20, 25]
    parameters:
      t:
        - name: self_attn
          value: [0.7, 0.73, 0.76, 0.79, 0.82]
        - name: mlp
          value: [0.75, 0.78, 0.81, 0.84, 0.87]
        - name: layer_norm
          value: [0.55, 0.58, 0.61, 0.64, 0.67]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [25, 30]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [25, 30]
    parameters:
      t:
        - name: self_attn
          value: [0.68, 0.71, 0.74, 0.77, 0.8]
        - name: mlp
          value: [0.7, 0.73, 0.76, 0.79, 0.82]
        - name: layer_norm
          value: [0.5, 0.53, 0.56, 0.59, 0.62]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [30, 35]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [30, 35]
    parameters:
      t:
        - name: self_attn
          value: [0.65, 0.68, 0.71, 0.74, 0.77]
        - name: mlp
          value: [0.68, 0.71, 0.74, 0.77, 0.8]
        - name: layer_norm
          value: [0.45, 0.48, 0.51, 0.54, 0.57]
        - name: embed_tokens
          value: [1.0]

  - sources:
      - model: bamec66557/VICIOUS_MESH-12B-OMEGA
        layer_range: [35, 40]
      - model: bamec66557/VICIOUS_MESH-12B-BETA
        layer_range: [35, 40]
    parameters:
      t:
        - name: self_attn
          value: [0.6, 0.63, 0.66, 0.69, 0.72]
        - name: mlp
          value: [0.65, 0.68, 0.71, 0.74, 0.77]
        - name: layer_norm
          value: [0.4, 0.43, 0.46, 0.49, 0.52]
        - name: embed_tokens
          value: [1.0]

# Regularization
regularization:
  - method: weight_clipping
    clip_range: [-0.1, 0.1]
  - method: random_noise
    scale: 0.003
  - method: attention_dropout
    scale: 0.05
  - method: gradient_clipping
    clip_norm: 1.0

# Postprocessing
postprocessing:
  - operation: non_linear_scaling
    parameters:
      function: tanh
  - operation: sharpening
    intensity: 0.4
  - operation: gaussian_smoothing
    sigma: 1.0
  - operation: normalize
  - operation: dynamic_scaling
    scale_range: [0.85, 1.15]
  - operation: smoothing
    parameters:
      adaptive: true
      range: [0.9, 1.1]
      kernel_size: 3

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	27.26
IFEval (0-Shot)	67.21
BBH (3-Shot)	31.36
MATH Lvl 5 (4-Shot)	12.08
GPQA (0-shot)	8.84
MuSR (0-shot)	14.34
MMLU-PRO (5-shot)	29.76