Introduction
We introduce NV-Embed, a generalist embedding model that ranks No. 1 on the Massive Text Embedding Benchmark (MTEB benchmark)(as of May 24, 2024), with 56 tasks, encompassing retrieval, reranking, classification, clustering, and semantic textual similarity tasks. Notably, our model also achieves the highest score of 59.36 on 15 retrieval tasks within this benchmark.
NV-Embed presents several new designs, including having the LLM attend to latent vectors for better pooled embedding output, and demonstrating a two-stage instruction tuning method to enhance the accuracy of both retrieval and non-retrieval tasks.
For more technical details, refer to our paper: NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models.
For more benchmark results (other than MTEB), please find the AIR-Bench for QA (English only) and Long-Doc.
Model Details
- Base Decoder-only LLM: Mistral-7B-v0.1
- Pooling Type: Latent-Attention
- Embedding Dimension: 4096
How to use
Here is an example of how to encode queries and passages using Huggingface-transformer and Sentence-transformer. Please find the required package version here.
Usage (HuggingFace Transformers)
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, AutoModel
# Each query needs to be accompanied by an corresponding instruction describing the task.
task_name_to_instruct = {"example": "Given a question, retrieve passages that answer the question",}
query_prefix = "Instruct: "+task_name_to_instruct["example"]+"\nQuery: "
queries = [
'are judo throws allowed in wrestling?',
'how to become a radiology technician in michigan?'
]
# No instruction needed for retrieval passages
passage_prefix = ""
passages = [
"Since you're reading this, you are probably someone from a judo background or someone who is just wondering how judo techniques can be applied under wrestling rules. So without further ado, let's get to the question. Are Judo throws allowed in wrestling? Yes, judo throws are allowed in freestyle and folkstyle wrestling. You only need to be careful to follow the slam rules when executing judo throws. In wrestling, a slam is lifting and returning an opponent to the mat with unnecessary force.",
"Below are the basic steps to becoming a radiologic technologist in Michigan:Earn a high school diploma. As with most careers in health care, a high school education is the first step to finding entry-level employment. Taking classes in math and science, such as anatomy, biology, chemistry, physiology, and physics, can help prepare students for their college studies and future careers.Earn an associate degree. Entry-level radiologic positions typically require at least an Associate of Applied Science. Before enrolling in one of these degree programs, students should make sure it has been properly accredited by the Joint Review Committee on Education in Radiologic Technology (JRCERT).Get licensed or certified in the state of Michigan."
]
# load model with tokenizer
model = AutoModel.from_pretrained('nvidia/NV-Embed-v1', trust_remote_code=True)
# get the embeddings
max_length = 4096
query_embeddings = model.encode(queries, instruction=query_prefix, max_length=max_length)
passage_embeddings = model.encode(passages, instruction=passage_prefix, max_length=max_length)
# normalize embeddings
query_embeddings = F.normalize(query_embeddings, p=2, dim=1)
passage_embeddings = F.normalize(passage_embeddings, p=2, dim=1)
# get the embeddings with DataLoader (spliting the datasets into multiple mini-batches)
# batch_size=2
# query_embeddings = model._do_encode(queries, batch_size=batch_size, instruction=query_prefix, max_length=max_length, num_workers=32, return_numpy=True)
# passage_embeddings = model._do_encode(passages, batch_size=batch_size, instruction=passage_prefix, max_length=max_length, num_workers=32, return_numpy=True)
scores = (query_embeddings @ passage_embeddings.T) * 100
print(scores.tolist())
#[[77.9402084350586, 0.4248958230018616], [3.757718086242676, 79.60113525390625]]
Usage (Sentence-Transformers)
import torch
from sentence_transformers import SentenceTransformer
# Each query needs to be accompanied by an corresponding instruction describing the task.
task_name_to_instruct = {"example": "Given a question, retrieve passages that answer the question",}
query_prefix = "Instruct: "+task_name_to_instruct["example"]+"\nQuery: "
queries = [
'are judo throws allowed in wrestling?',
'how to become a radiology technician in michigan?'
]
# No instruction needed for retrieval passages
passages = [
"Since you're reading this, you are probably someone from a judo background or someone who is just wondering how judo techniques can be applied under wrestling rules. So without further ado, let's get to the question. Are Judo throws allowed in wrestling? Yes, judo throws are allowed in freestyle and folkstyle wrestling. You only need to be careful to follow the slam rules when executing judo throws. In wrestling, a slam is lifting and returning an opponent to the mat with unnecessary force.",
"Below are the basic steps to becoming a radiologic technologist in Michigan:Earn a high school diploma. As with most careers in health care, a high school education is the first step to finding entry-level employment. Taking classes in math and science, such as anatomy, biology, chemistry, physiology, and physics, can help prepare students for their college studies and future careers.Earn an associate degree. Entry-level radiologic positions typically require at least an Associate of Applied Science. Before enrolling in one of these degree programs, students should make sure it has been properly accredited by the Joint Review Committee on Education in Radiologic Technology (JRCERT).Get licensed or certified in the state of Michigan."
]
# load model with tokenizer
model = SentenceTransformer('nvidia/NV-Embed-v1', trust_remote_code=True)
model.max_seq_length = 4096
model.tokenizer.padding_side="right"
def add_eos(input_examples):
input_examples = [input_example + model.tokenizer.eos_token for input_example in input_examples]
return input_examples
# get the embeddings
batch_size = 2
query_embeddings = model.encode(add_eos(queries), batch_size=batch_size, prompt=query_prefix, normalize_embeddings=True)
passage_embeddings = model.encode(add_eos(passages), batch_size=batch_size, normalize_embeddings=True)
scores = (query_embeddings @ passage_embeddings.T) * 100
print(scores.tolist())
Correspondence to
Chankyu Lee (chankyul@nvidia.com), Rajarshi Roy (rajarshir@nvidia.com), Wei Ping (wping@nvidia.com)
Citation
If you find this code useful in your research, please consider citing:
@misc{lee2024nvembed,
title={NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models},
author={Chankyu Lee and Rajarshi Roy and Mengyao Xu and Jonathan Raiman and Mohammad Shoeybi and Bryan Catanzaro and Wei Ping},
year={2024},
eprint={2405.17428},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
License
This model should not be used for any commercial purpose. Refer the license for the detailed terms.
For commercial purpose, we recommend you to use the models of NeMo Retriever Microservices (NIMs).
Troubleshooting
1. How to enable Multi-GPU (Note, this is the case for HuggingFace Transformers)
from transformers import AutoModel
from torch.nn import DataParallel
embedding_model = AutoModel.from_pretrained("nvidia/NV-Embed-v1")
for module_key, module in embedding_model._modules.items():
embedding_model._modules[module_key] = DataParallel(module)
2. Required Packages
If you have trouble, try installing the python packages as below
pip uninstall -y transformer-engine
pip install torch==2.2.0
pip install transformers==4.42.4
pip install flash-attn==2.2.0
pip install sentence-transformers==2.7.0
3. Fixing "nvidia/NV-Embed-v1 is not the path to a directory containing a file named config.json"
Switch to your local model path，and open config.json and change the value of "_name_or_path" and replace it with your local model path.
4. Access to model nvidia/NV-Embed-v1 is restricted. You must be authenticated to access it
Use your huggingface access token to execute "huggingface-cli login".
5. How to resolve slight mismatch in Sentence transformer results.
A slight mismatch in the Sentence Transformer implementation is caused by a discrepancy in the calculation of the instruction prefix length within the Sentence Transformer package.
To fix this issue, you need to build the Sentence Transformer package from source, making the necessary modification in this line as below.
git clone https://github.com/UKPLab/sentence-transformers.git
cd sentence-transformers
git checkout v2.7-release
# Modify L353 in SentenceTransformer.py to **'extra_features["prompt_length"] = tokenized_prompt["input_ids"].shape[-1]'**.
pip install -e .