README.md

metadata

license: apache-2.0
tags:
  - merge
  - mergekit
  - dare_ties
  - 7B
  - eren23/OGNO-7b-dpo-truthful
  - Kquant03/NeuralTrix-7B-dpo-laser

MergedAI-7B-DARE-0

A merged model using Dynamic Adaptive Region Ensemble (DARE) with TIES sign election step using mergekit.

Model Details

• Base Models:
  • eren23/OGNO-7b-dpo-truthful
  • Kquant03/NeuralTrix-7B-dpo-laser
• Merge Method: dare_ties

Configuration

models:
  - model: eren23/OGNO-7b-dpo-truthful
  - model: Kquant03/NeuralTrix-7B-dpo-laser
    parameters:
      density: 0.53
      weight: 0.95
merge_method: dare_ties
base_model: eren23/OGNO-7b-dpo-truthful
parameters:
  int8_mask: true
dtype: bfloat16

Usage

This model can be used with the standard transformers library:

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Ro-xe/MergedAI-7B-DARE-0")
tokenizer = AutoTokenizer.from_pretrained("Ro-xe/MergedAI-7B-DARE-0")