NegMPNet

This is a negation-aware version of all-mpnet-base-v2. It is a sentence-transformers model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search. For further information, see our paper This is not correct! Negation-aware Evaluation of Language Generation Systems.

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 sentence", "Each sentence is converted"]

model = SentenceTransformer("tum-nlp/NegMPNet")
embeddings = model.encode(sentences)
print(embeddings)

Negation-awareness

This model has a better sensitivity towards negations compared to its base model. You can try it yourself:

from sentence_transformers import SentenceTransformer, util
import torch

base_model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")
finetuned_model = SentenceTransformer("tum-nlp/NegMPNet")

def cos_similarities(references: list, candidates: list, model: SentenceTransformer, batch_size=8) -> torch.Tensor:
    assert len(references) == len(candidates), "Number of references and candidates must be equal"
    emb_ref = model.encode(references, batch_size=batch_size)
    emb_cand = model.encode(candidates, batch_size=batch_size)
    return torch.diag(util.cos_sim(emb_ref, emb_cand))

references = ["Ray charles is legendary.", "Ray charles is legendary"]
candidates = ["Ray charles is a legend.", "Ray charles isn't legendary."]
print(cos_similarities(references, candidates, base_model)) # prints tensor([0.9453, 0.8683]) -> no negation-awareness
print(cos_similarities(references, candidates, finetuned_model)) # prints tensor([0.9585, 0.4263]) -> sensitive to negation

Usage (HuggingFace Transformers)

Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.

from transformers import AutoTokenizer, AutoModel
import torch

#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)

# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained("tum-nlp/NegMPNet")
model = AutoModel.from_pretrained("tum-nlp/NegMPNet")

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

Evaluation Results

For an automated evaluation of this model, see the Sentence Embeddings Benchmark: https://seb.sbert.net

Training

The model was trained with the parameters:

DataLoader:

sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader of length 358 with parameters:

{'batch_size': 64}

Loss:

__main__.MultipleNegativesRankingLoss with parameters:

{'scale': 20.0, 'similarity_fct': 'cos_sim'}

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 35,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 36,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 75, 'do_lower_case': False}) with Transformer model: MPNetModel
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)

Citation

Please cite our INLG 2023 paper, if you use our model. BibTeX:

@misc{anschütz2023correct,
      title={This is not correct! Negation-aware Evaluation of Language Generation Systems},
      author={Miriam Anschütz and Diego Miguel Lozano and Georg Groh},
      year={2023},
      eprint={2307.13989},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}