metadata

base_model:
  - TheHierophant/Underground-Mind-V0.9
  - TheHierophant/Underground-Cognitive-V0.3-test
  - TheHierophant/Fimbulvetr-11B-Attention-V0.1-test
  - TheHierophant/Underground-Mind-V0.3-test-finetuning
library_name: transformers
tags:
  - mergekit
  - merge

merge

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

Merge Details

Merge Method

This model was merged using the passthrough merge method using TheHierophant/Fimbulvetr-11B-Attention-V0.1-test as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: TheHierophant/Fimbulvetr-11B-Attention-V0.1-test
        layer_range: [0, 16]
        parameters:
          scale:
            - filter: o_proj
              value: 1.25
            - filter: down_proj
              value: 1.25
          attention_heads: 32
          long_term_attention: true
  - sources:
      - model: TheHierophant/Underground-Mind-V0.9
        layer_range: [16, 32]
        parameters:
          scale:
            - filter: o_proj
              value: 1.5
            - filter: down_proj
              value: 1.5
          significance: 0.8
          semantic_linking: true
  - sources:
      - model: TheHierophant/Underground-Mind-V0.3-test-finetuning
        layer_range: [32, 40]
        parameters:
          scale:
            - filter: o_proj
              value: 1.75
            - filter: down_proj
              value: 1.75
          task_specialization: true
          enhanced_attention: true
  - sources:
      - model: TheHierophant/Underground-Cognitive-V0.3-test
        layer_range: [40, 47]
        parameters:
          scale:
            - filter: o_proj
              value: 2.0
            - filter: down_proj
              value: 2.0
          attention_heads: 18
          abstract_attention: true
          deep_cognitive_focus: true

merge_method: passthrough
base_model: TheHierophant/Fimbulvetr-11B-Attention-V0.1-test
dtype: bfloat16