DunnBC22/sentence-t5-base-FT-Quora_Sentence_Similarity-LG

This 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 more information on how it was created, check out the following link: https://github.com/DunnBC22/NLP_Projects/blob/main/Semantic_Similarity/Semantic%20Similarity-base.ipynb

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('DunnBC22/sentence-t5-base-FT-Quora_Sentence_Similarity-LG')
embeddings = model.encode(sentences)
print(embeddings)

Evaluation Results

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

Metric	Measure	Value	Notes
Accuracy	Cosine-Similarity	85.93	Threshold: 0.8320
F1	Cosine-Similarity	82.89	Threshold: 0.8178
Precision	Cosine-Similarity	77.43	-
Recall	Cosine-Similarity	89.18	-
Average Precision	Cosine-Similarity	87.13	-
Accuracy	Manhattan-Distance	85.95	Threshold: 12.7721
F1	Manhattan-Distance	82.89	Threshold: 13.5008
Precision	Manhattan-Distance	76.91	-
Recall	Manhattan-Distance	89.89	-
Average Precision	Manhattan-Distance	87.13	-
Accuracy	Euclidean-Distance	85.93	Threshold: 0.5797
F1	Euclidean-Distance	82.89	Threshold: 0.6037
Precision	Euclidean-Distance	77.43	-
Recall	Euclidean-Distance	89.18	-
Average Precision	Euclidean-Distance	87.13	-
Accuracy	Dot-Product	85.93	Threshold: 0.8320
F1	Dot-Product	82.89	Threshold: 0.8178
Precision	Dot-Product	77.43	-
Recall	Dot-Product	89.18	-
Average Precision	Dot-Product	87.14	-

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 4673 with parameters:

{'batch_size': 64, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.OnlineContrastiveLoss.OnlineContrastiveLoss

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 0,
    "evaluator": "sentence_transformers.evaluation.BinaryClassificationEvaluator.BinaryClassificationEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 2,
    "weight_decay": 0.01
}

Potential Improvements

One way to improve the results of this model is to use a larger checkpoint of T5. This was trained with the T5-base checkpoint.

The larger checkpoints are:

Checkpoint	# of Train Params
T5-Base	220 Million*
T5-Large	770 Million
T5-3B	3 Billion
T5-11B	11 Billion

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 34, 'do_lower_case': False}) with Transformer model: T5EncoderModel 
  (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})
  (2): Dense({'in_features': 768, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (3): Normalize()
)

Citing & Authors

Dataset Source: https://www.kaggle.com/datasets/quora/question-pairs-dataset