sentence-transformers/all-MiniLM-L6-v2 · Hugging Face

Source: https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2 This is asentence-transformersmodel: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#usage-sentence-transformersUsage (Sentence-Transformers)

Using this model becomes easy when you havesentence-transformersinstalled:

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('sentence-transformers/all-MiniLM-L6-v2')
embeddings = model.encode(sentences)
print(embeddings)

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#usage-huggingface-transformersUsage (HuggingFace Transformers)

Withoutsentence-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
import torch.nn.functional as F

#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('sentence-transformers/all-MiniLM-L6-v2')
model = AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')

# 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
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

# Normalize embeddings
sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)

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

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#backgroundBackground

The project aims to train sentence embedding models on very large sentence level datasets using a self-supervised contrastive learning objective. We used the pretrainednreimers/MiniLM\-L6\-H384\-uncasedmodel and fine-tuned in on a 1B sentence pairs dataset. We use a contrastive learning objective: given a sentence from the pair, the model should predict which out of a set of randomly sampled other sentences, was actually paired with it in our dataset.

We developed this model during theCommunity week using JAX/Flax for NLP & CV, organized by Hugging Face. We developed this model as part of the project:Train the Best Sentence Embedding Model Ever with 1B Training Pairs. We benefited from efficient hardware infrastructure to run the project: 7 TPUs v3-8, as well as intervention from Googles Flax, JAX, and Cloud team member about efficient deep learning frameworks.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#intended-usesIntended uses

Our model is intended to be used as a sentence and short paragraph encoder. Given an input text, it outputs a vector which captures the semantic information. The sentence vector may be used for information retrieval, clustering or sentence similarity tasks.

By default, input text longer than 256 word pieces is truncated.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#training-procedureTraining procedure

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#pre-trainingPre-training

We use the pretrainednreimers/MiniLM\-L6\-H384\-uncasedmodel. Please refer to the model card for more detailed information about the pre-training procedure.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#fine-tuningFine-tuning

We fine-tune the model using a contrastive objective. Formally, we compute the cosine similarity from each possible sentence pairs from the batch. We then apply the cross entropy loss by comparing with true pairs.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#hyper-parametersHyper parameters

We trained our model on a TPU v3-8. We train the model during 100k steps using a batch size of 1024 (128 per TPU core). We use a learning rate warm up of 500. The sequence length was limited to 128 tokens. We used the AdamW optimizer with a 2e-5 learning rate. The full training script is accessible in this current repository:train\_script\.py.

https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2#training-dataTraining data

We use the concatenation from multiple datasets to fine-tune our model. The total number of sentence pairs is above 1 billion sentences. We sampled each dataset given a weighted probability which configuration is detailed in thedata\_config\.jsonfile.

DatasetPaperNumber of training tuplesReddit comments (2015-2018)paper726,484,430S2ORCCitation pairs (Abstracts)paper116,288,806WikiAnswersDuplicate question pairspaper77,427,422PAQ(Question, Answer) pairspaper64,371,441S2ORCCitation pairs (Titles)paper52,603,982S2ORC(Title, Abstract)paper41,769,185Stack Exchange(Title, Body) pairs-25,316,456Stack Exchange(Title+Body, Answer) pairs-21,396,559Stack Exchange(Title, Answer) pairs-21,396,559MS MARCOtripletspaper9,144,553GOOAQ: Open Question Answering with Diverse Answer Typespaper3,012,496Yahoo Answers(Title, Answer)paper1,198,260Code Search-1,151,414COCOImage captionspaper828,395SPECTERcitation tripletspaper684,100Yahoo Answers(Question, Answer)paper681,164Yahoo Answers(Title, Question)paper659,896SearchQApaper582,261Eli5paper325,475Flickr 30kpaper317,695Stack ExchangeDuplicate questions (titles)304,525AllNLI (SNLIandMultiNLIpaper SNLI,paper MultiNLI277,230Stack ExchangeDuplicate questions (bodies)250,519Stack ExchangeDuplicate questions (titles+bodies)250,460Sentence Compressionpaper180,000Wikihowpaper128,542Altlexpaper112,696Quora Question Triplets-103,663Simple Wikipediapaper102,225Natural Questions (NQ)paper100,231SQuAD2.0paper87,599TriviaQA-73,346Total****1,170,060,424