inf retriever v1

Introduction

INF-Retriever-v1 is a dense retrieval model built on the gte-Qwen2-7B-instruct model, fine-tuned for optimizing retrieval tasks in both Chinese and English. It is developed by INF TECH and excels in heterogeneous information retrieval tasks, ranking No.1 on the AIR-Bench 24.04 as of December 23, 2024.

Architecture

INF-Retriever-v1 leverages the capabilities of the gte-Qwen2-7B-instruct model, optimized specifically for retrieval tasks. It supports sentence-transformers and transformers frameworks, allowing efficient processing of both English and Chinese text data.

Training

The model has been fine-tuned using retrieval-focused datasets, ensuring high performance and accuracy in retrieving relevant information from large corpora.

Guide: Running Locally

Basic Steps

1. Install Dependencies: Ensure Python and libraries such as transformers, torch, and sentence_transformers are installed.
2. Load the Model:

   from sentence_transformers import SentenceTransformer
   model = SentenceTransformer("infly/inf-retriever-v1", trust_remote_code=True)
   

3. Process Queries and Documents:
   • Use the model's encode method for embeddings.
   • Compute similarity scores using dot products of embeddings.
4. Alternative with Transformers:

   import torch
   from transformers import AutoTokenizer, AutoModel
   
   tokenizer = AutoTokenizer.from_pretrained('infly/inf-retriever-v1', trust_remote_code=True)
   model = AutoModel.from_pretrained('infly/inf-retriever-v1', trust_remote_code=True)
   

Suggest Cloud GPUs

Consider using cloud GPUs from providers like AWS, Google Cloud, or Azure for efficient processing, especially when handling large datasets.

License

The usage license for INF-Retriever-v1 is not specified in the provided documentation. Please refer to the Hugging Face model card for detailed licensing information.