Upload folder using huggingface_hub

f2427b6 verified 3 months ago

985 Bytes

license: llama3.2
base_model:
  - ValiantLabs/Llama3.2-3B-Enigma
  - deep-div/MediLlama-3.2
  - prithivMLmods/Llama-3.2-3B-Math-Oct
  - Saxo/Linkbricks-Llama3.2-Korean-cpt-3b
  - GetSoloTech/Llama-3.2-3B-Reasoning
tags:
  - merge
  - llama
  - llama-3.2

Llama3.2-3B-Linear-5way

A merge of 5 specialist Llama 3.2 3B models using Linear merge method.

Merged Models

ValiantLabs/Llama3.2-3B-Enigma - Code specialist
deep-div/MediLlama-3.2 - Medical specialist
prithivMLmods/Llama-3.2-3B-Math-Oct - Math specialist
Saxo/Linkbricks-Llama3.2-Korean-cpt-3b - Korean language specialist
GetSoloTech/Llama-3.2-3B-Reasoning - Reasoning specialist

Evaluation

ARC-Challenge: 47.0%

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Bhavkumar21/Llama3.2-3B-Linear-5way")
tokenizer = AutoTokenizer.from_pretrained("Bhavkumar21/Llama3.2-3B-Linear-5way")