Datasets:

nirantk

/

triplets

like 12

Hello!

Pull Request overview

• Add the sentence-transformers tag.

Details

The upcoming Sentence Transformers v3 update will introduce training directly with Dataset instances, e.g. like so:

from datasets import load_dataset
from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer
from sentence_transformers.losses import MultipleNegativesRankingLoss

# 1. Load a model to finetune
model = SentenceTransformer("microsoft/mpnet-base")

# 2. Load a dataset to finetune on
dataset = load_dataset("nirantk/triplets", split="train").select_columns(["query", "pos", "neg"])
train_dataset = dataset[:-5000]
eval_dataset = dataset[-5000:]

# 3. Define a loss function (https://sbert.net/docs/training/loss_overview.html)
loss = MultipleNegativesRankingLoss(model)

# 4. Create a trainer & train
trainer = SentenceTransformerTrainer(
    model=model,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    loss=loss,
)
trainer.train()

# 5. Save the trained model
model.save_pretrained("models/mpnet-base-nirantk-triplets")

In preparation for the release, I'm going through and tagging some excellent datasets that immediately match one of the dataset formats required for one of the loss functions as sentence-transformers. Then I can link to datasets with this tag in the Sentence Transformers documentation.

This dataset in particular matches the (anchor, positive, negative) triplets without any label, allowing this dataset to be used out of the box for CachedMultipleNegativesRankingLoss, MultipleNegativesRankingLoss, TripletLoss, CachedGISTEmbedLoss, and GISTEmbedLoss.

For context, I have some example models trained with v3 on my profile: https://huggingface.co/models?author=tomaarsen&sort=created

• Tom Aarsen