thebajajra/RexReranker-base

Reranker Model

This model was exported from checkpoint: rexbert-reranker-base/checkpoint-92220

Model Details

• Base Model: thebajajra/RexBERT-base
• Task: Document Reranking
• Output: Relevance score between 0 and 1

Usage

import torch
from transformers import AutoTokenizer
from train_modernbert_reranker import ModernBERTReranker

# Load model and tokenizer
model_path = "rexreranker-base"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = ModernBERTReranker.from_pretrained(model_path)
model.eval()

# Example inference
query = "wireless bluetooth headphones"
document = "Title: Sony WH-1000XM5\nDescription: Premium wireless headphones with noise cancellation"

inputs = tokenizer(query, document, return_tensors="pt", truncation=True, max_length=2048)

with torch.no_grad():
    outputs = model(**inputs)
    score = outputs.logits.squeeze().item()
    
print(f"Relevance score: {score:.4f}")

Input Format

The model expects query-document pairs formatted as:

Query: <query text>
[SEP]
Title: <title>
Description: <description>

Training

This model was trained on the Amazebay reranker dataset with:

• Max sequence length: 2048
• BF16 precision
• Flash Attention 2
• Multi-GPU training (4 GPUs)