Model reference · open weights
multi-qa-MiniLM-L6-cls_dot is an open-weight embedding model from flax-sentence-embeddings. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.
Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.
What it is
| Released by | flax-sentence-embeddings |
|---|---|
| Type | Embedding models |
| Task | Embeddings |
| Context | 512 tokens |
| Runs with | sentence-transformers |
| Released | 2022-03-02 |
| Popularity | 5k downloads / month |
| Licence | Unknown |
About
SentenceTransformers is a set of models and frameworks that enable training and generating sentence embeddings from given data. The generated sentence embeddings can be utilized for Clustering, Semantic Search and other tasks. We used a pretrained nreimers/MiniLM-L6-H384-uncased model and trained it using Siamese Network setup and contrastive learning objective. Question and answer pairs from StackExchange was used as training data to make the model robust to Question / Answer embedding similarity. For this model, cls output was used instead of mean pooling as sentence embeddings. Dot product was used to calculate similarity for learning objective.
We developed this model during the Community 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 assistance from Google’s Flax, JAX, and Cloud team members about efficient deep learning frameworks.
Our model is intended to be used as a sentence encoder for a search engine. Given an input sentence, it outputs a vector which captures the sentence semantic information. The sentence vector may be used for semantic-search, clustering or sentence similarity tasks.
Here is how to use this model to get the features of a given text using SentenceTransformers library:
from sentence_transformers import SentenceTransformer
model = SentenceTransformer('flax-sentence-embeddings/multi-qa_v1-MiniLM-L6-cls_dot')
text = "Replace me by any question / answer you'd like."
text_embbedding = model.encode(text)
# array([-0.01559514, 0.04046123, 0.1317083 , 0.00085931, 0.04585106,
# -0.05607086, 0.0138078 , 0.03569756, 0.01420381, 0.04266302 ...],
# dtype=float32)
We use the pretrained nreimers/MiniLM-L6-H384-uncased. Please refer to the model card for more detailed information about the pre-training procedure.
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.
We trained on model on a TPU v3-8. We train the model during 80k 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.
We used the concatenation from multiple Stackexchange Question-Answer datasets to fine-tune our model. MSMARCO, NQ & other question-answer datasets were also used.
| Dataset | Paper | Number of training tuples |
|---|---|---|
| Stack Exchange QA - Title & Answer | - | 4,750,619 |
| Stack Exchange | - | 364,001 |
| TriviaqQA | - | 73,346 |
| SQuAD2.0 | paper | 87,599 |
| Quora Question Pairs | - | 103,663 |
| Eli5 | paper | 325,475 |
| PAQ | paper | 64,371,441 |
| WikiAnswers | paper | 77,427,422 |
| MS MARCO | paper | 9,144,553 |
| GOOAQ: Open Question Answering with Diverse Answer Types | paper | 3,012,496 |
| Yahoo Answers Question/Answer | paper | 681,164 |
| SearchQA | - | 582,261 |
| Natural Questions (NQ) | paper | 100,231 |
@misc{multiqa-embeddings,
author = {Flax Sentence Embeddings Team},
title = {Multi QA: Sentence Embeddings},
year = {2021},
url = {https://huggingface.co/flax-sentence-embeddings/multi-qa_v1-MiniLM-L6-cls_dot},
}
From the published model card. Full card on the HuggingFace links in the sidebar.
Using it via the API
Once AxForge deploys multi-qa-minilm-l6-cls-dot for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (multi-qa-minilm-l6-cls-dot below is illustrative; you get the exact model name on deployment.)
$ curl -sS https://api.axforge.ai/v1/embeddings \
-H "Authorization: Bearer $AXFORGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"multi-qa-minilm-l6-cls-dot","input":"text to embed"}'
Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.