zjkarina/omniRecsysLLM_semanticIDsmodality

Recommendation model with semantic IDs for Amazon Fashion.

Description

This model uses VQ-VAE to create semantic item IDs, enabling a more accurate understanding of semantic relationships between products.

Architecture

• Base model: Qwen2.5-Omni-7B
• Item vocabulary size: 709,036
• ID embedding dimension: 512
• VQ-VAE codebook size: 10,000
• VQ-VAE codebook dimension: 256
• Dataset: Amazon Fashion 2023 Full

Usage

from any2any_trainer.models.recommendation import SemanticIDRecommendationModel

# Load model
model = SemanticIDRecommendationModel.from_pretrained("zjkarina/omniRecsysLLM_semanticIDsmodality")

# Generate recommendations with semantic IDs
recommendations = model.predict_next_item(
    text="The user bought jeans and a t-shirt",
    id_ids=[12345, 67890],  # Item IDs from purchase history
    top_k=5,
    use_semantic_ids=True
)

Training

The model was trained on the Amazon Fashion 2023 dataset using semantic IDs generated via VQ-VAE.