README.md

---
license: apache-2.0
base_model:
- Qwen/Qwen2.5-7B
pipeline_tag: text-generation
tags:
- not-for-all-audiences
language:
- en
library_name: transformers
---

## Model Description

Model created by analyzing and selecting the optimal layers from other Qwen2.5-7B models based on their dimensional utilization efficiency, measured by the Normalized Effective Rank (NER). Computed like:

   - Input: Weight matrix for each model layer
   - Compute singular values σᵢ where σᵢ ≥ 0 # σᵢ represents the importance of each dimension
   - Filter values above numerical threshold (>1e-12) 
   - Sum all singular values: S = Σσᵢ # S acts as normalization factor
   - Create probability distribution: pᵢ = σᵢ/S # converts singular values to probabilities summing to 1
   - Compute Shannon entropy: H = -Σ(pᵢ * log₂(pᵢ)) # measures information content 
   - Calculate maximum possible entropy: H_max = log₂(n)
   - Final NER score = H/H_max # normalizes score to [0,1] range
   - Results in value between 0 and 1
     # 0 = single dimension dominance, 1 = uniform dimensional utilization

## Creating Composite Model

Code here: https://huggingface.co/jeffmeloy/Qwen2.5-7B-nerd-uncensored-v1.0/blob/main/ner_merge.py

Layer Analysis:
   - Download base and fine-tuned models from Hugging Face Hub 
   - Calculate Normalized Effective Rank (NER) for each layer within each model 

Layer Selection:
   - Identify common layer structures across models 
   - Define model and layer name pairs that have highest NER for each layer based on their NER scores 

Model Composition:
   - Incrementally build a composite model using layer with highest NER from model pool. 

Output Generation:
   - Save merge reports documenting layer sources 
   - Copy config and tokenizer files from base model
   - Save the composite model with complete weights # model ready to use

Configfile:

base_model: "Qwen/Qwen2.5-7B"

fine_tuned_models: # uncomment the models you want to merge

#- "Qwen/Qwen2.5-7B"

#- "Qwen/Qwen2.5-7B-Instruct"

#- "EVA-UNIT-01/EVA-Qwen2.5-7B-v0.1"

#- "FourOhFour/Vapor_v2_7B"

#- "Goekdeniz-Guelmez/Josiefied-Qwen2.5-7B-Instruct-abliterated-v2"

#- "happzy2633/qwen2.5-7b-ins-v3"

#- "huihui-ai/Qwen2.5-7B-Instruct-abliterated-v2"

#- "HumanLLMs/Humanish-Qwen2.5-7B-Instruct"

#- "Orion-zhen/Qwen2.5-7B-Instruct-Uncensored"

#- "Orion-zhen/Meissa-Qwen2.5-7B-Instruct"

#- "jeffmeloy/Qwen2.5-7B-nerd-uncensored-v0.9"

#- "jeffmeloy/Qwen2.5-7B-nerd-uncensored-v1.0"

#- "jeffmeloy/Qwen2.5-7B-nerd-uncensored-v1.1"

#- "jeffmeloy/Qwen2.5-7B-nerd-uncensored-v1.2"

#- "AmberYifan/Qwen2.5-7B-dpo-2k"

#- "sethuiyer/Qwen2.5-7B-Anvita"

#- "rombodawg/Rombos-LLM-V2.5-Qwen-7b"

#- "Cran-May/T.E-8.1"

#- "beomi/Qwen2.5-7B-Instruct-kowiki-qa"

#- "Orion-zhen/Qwen2.5-7B-Gutenberg-KTO"

#- "fblgit/cybertron-v4-qw7B-MGS"

#- "nguyentd/FinancialAdvice-Qwen2.5-7B"

#- "WhiteRabbitNeo/WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B"

#- "edgerunner-ai/EdgeRunner-Command-Nested"

#- "katanemo/Arch-Function-7B"

#- "DeepGlint-AI/llava-mlcd-qwen2.5-7b"

#- "mergekit-community/mergekit-slerp-aflqaqy"

#- "mergekit-community/mergekit-ties-inxwsfo"

#- "Qwen/Qwen2.5-Coder-7B-Instruct"

#- "Qwen/Qwen2.5-Math-7B-Instruct" 

#- "Qwen/Qwen2.5-Coder-7B"

#- "Qwen/Qwen2.5-Math-7B"

#- "thomas-yanxin/XinYuan-Qwen2.5-7B-0917"

#- "jbjeong91/Qwen2.5_7B_IST_StoryGen_vanilla"

#- "AmberYifan/Qwen2.5-7B-dpo-2k-hhrlhf"

#- "jbjeong91/Qwen2.5_7B_IST_StoryGen_test2" 

models_dir: "./input_models/"

output_dir: "./merged_model/"

metric_dir: "./metrics/"