zjkarina/omniRecsysLLM_fasion

Multimodal Large Language Model for Amazon Fashion recommendations.

Description

This model is trained on a multitask Amazon Fashion dataset and can perform the following tasks:

• Product analysis
• Next-purchase recommendations
• Product comparison
• Personalized recommendations
• Review generation

Architecture

• Base model: Qwen2.5-7B-Instruct
• Visual encoder: CLIP-ViT-Large-Patch14
• Projection layer: MLP (1024 hidden units, 2 layers)
• Maximum sequence length: 4096 tokens

Usage

from transformers import AutoTokenizer, AutoModel
import torch

# Load model
model = AutoModel.from_pretrained("zjkarina/omniRecsysLLM_fasion")
tokenizer = AutoTokenizer.from_pretrained("zjkarina/omniRecsysLLM_fasion")

# Example usage
text = "Analyze this fashion product"
# Add code to handle both images and text

Training

The model was trained on the Amazon Fashion dataset using Any2Any Trainer.