ED-Zephyria-48b / README.md
Steelskull's picture
Update README.md
b60d67b verified
metadata
base_model:
  - unsloth/Mistral-Small-Instruct-2409
library_name: transformers
tags:
  - mergekit
  - merge

ED-Zephyria-48b [EXPRIMENTAL]

Model Information

Base Model: unsloth/Mistral-Small-Instruct-2409

Strategy: Early Duplication

Total Layers: 55

Duplication Start: Layer 14 (25.5% of model)

Duplicated Layers: 35 (63.6% of model)

Unique Final Layers: 7 (12.7% of model)

Model Characteristics

  • Models down_proj and o_proj layers have been nulled and will require healing
  • Focuses on refining early features
  • Largest duplicated section among all strategies
  • Suitable for tasks requiring intensive low-level feature processing
  • May excel in tasks that benefit from extensive refinement of basic patterns

Configuration Visualization


[   Unique   ][        Duplicated        ][Unique]
0 --------- 13 14 ------------------- 48 49 --- 54
    25.5%              63.6%            10.9%