Monarch Mixer-BERT

An 80M checkpoint of M2-BERT, pretrained with sequence length 32768, and it has been fine-tuned for long-context retrieval.

Check out the paper Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture and our blog post on retrieval for more on how we trained this model for long sequence.

This model was trained by Jon Saad-Falcon, Dan Fu, and Simran Arora.

Check out our GitHub for instructions on how to download and fine-tune it!

How to use

You can load this model using Hugging Face AutoModel:

from transformers import AutoModelForSequenceClassification
model = AutoModelForSequenceClassification.from_pretrained(
  "togethercomputer/m2-bert-80M-32k-retrieval",
  trust_remote_code=True
)

You should expect to see a large error message about unused parameters for FlashFFTConv. If you'd like to load the model with FlashFFTConv, you can check out our GitHub.

This model generates embeddings for retrieval. The embeddings have a dimensionality of 768:

from transformers import AutoTokenizer, AutoModelForSequenceClassification

max_seq_length = 32768
testing_string = "Every morning, I make a cup of coffee to start my day."
model = AutoModelForSequenceClassification.from_pretrained(
  "togethercomputer/m2-bert-80M-32k-retrieval",
  trust_remote_code=True
)

tokenizer = AutoTokenizer.from_pretrained(
  "bert-base-uncased",
  model_max_length=max_seq_length
)
input_ids = tokenizer(
  [testing_string],
  return_tensors="pt",
  padding="max_length",
  return_token_type_ids=False,
  truncation=True,
  max_length=max_seq_length
)

outputs = model(**input_ids)
embeddings = outputs['sentence_embedding']

You can also get embeddings from this model using the Together API as follows (you can find your API key here):

import os
import requests

def generate_together_embeddings(text: str, model_api_string: str, api_key: str):
    url = "https://api.together.xyz/api/v1/embeddings"
    headers = {
        "accept": "application/json",
        "content-type": "application/json",
        "Authorization": f"Bearer {api_key}"
    }
    session = requests.Session()
    response = session.post(
        url,
        headers=headers,
        json={
            "input": text,
            "model": model_api_string
        }
    )
    if response.status_code != 200:
        raise ValueError(f"Request failed with status code {response.status_code}: {response.text}")
    return response.json()['data'][0]['embedding']

print(generate_together_embeddings(
  'Hello world',
  'togethercomputer/m2-bert-80M-32k-retrieval',
  os.environ['TOGETHER_API_KEY'])[:10]
)

Acknowledgments

Alycia Lee helped with AutoModel support.

Citation

If you use this model, or otherwise found our work valuable, you can cite us as follows:

@inproceedings{fu2023monarch,
  title={Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture},
  author={Fu, Daniel Y and Arora, Simran and Grogan, Jessica and Johnson, Isys and Eyuboglu, Sabri and Thomas, Armin W and Spector, Benjamin and Poli, Michael and Rudra, Atri and R{\'e}, Christopher},
  booktitle={Advances in Neural Information Processing Systems},
  year={2023}
}