Llama-goodlens-mpnet

This is a sentence-transformers model: It maps sentences & paragraphs to a 768 dimensional dense vector space and it is designed to use in recommender systems for content-base filtering and as a side information for cold-start recommendation.

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["This is an example product description", "Each product description is converted"]
model = SentenceTransformer('beeformer/Llama-goodlens-mpnet')
embeddings = model.encode(sentences)
print(embeddings)

Training procedure

Pre-training

We use the pretrained sentence-transformers/all-mpnet-base-v2 model. Please refer to the model card for more detailed information about the pre-training procedure.

Fine-tuning

We use the initial model without modifying its architecture or pre-trained model parameters. However, we reduce the processed sequence length to 384 to reduce the training time of the model.

Dataset

We finetuned our model on the combination of the Goodbooks-10k and the MovieLens20M datasets with item descriptions generated with meta-llama/Meta-Llama-3.1-8B-Instruct model. For details please see the dataset pages: beeformer/recsys-movielens-20m and beeformer/recsys-goodbooks-10k.

Evaluation Results

Table with results TBA.

Intended uses

This model was trained as a demonstration of capabilities of the beeFormer training framework (link and details TBA) and is intended for research purposes only.

Citation

Preprint available here

TBA