Sentence Similarity
sentence-transformers
PyTorch
ONNX
Safetensors
OpenVINO
Transformers
English
mpnet
fill-mask
feature-extraction
text-embeddings-inference
Eval Results
Instructions to use sentence-transformers/all-mpnet-base-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use sentence-transformers/all-mpnet-base-v2 with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2") sentences = [ "That is a happy person", "That is a happy dog", "That is a very happy person", "Today is a sunny day" ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [4, 4] - Transformers
How to use sentence-transformers/all-mpnet-base-v2 with Transformers:
# Load model directly from transformers import AutoTokenizer, AutoModelForMaskedLM tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/all-mpnet-base-v2") model = AutoModelForMaskedLM.from_pretrained("sentence-transformers/all-mpnet-base-v2", device_map="auto") - Inference
- Notebooks
- Google Colab
- Kaggle
all-mpnet-base-v2 — the quality benchmark for mobile embeddings
#51
by 3morixd - opened
When we need the highest quality embeddings and can afford 420MB, this is the model.
Quality vs size tradeoff on mobile:
- all-MiniLM-L6-v2 (80MB): 85% quality, 200 t/s
- all-mpnet-base-v2 (420MB): 92% quality, 120 t/s
- BGE-small-en-v1.5 (130MB): 88% quality, 200 t/s
For most mobile apps, MiniLM is the sweet spot. But for high-stakes retrieval (legal, medical), mpnet is worth the extra storage.
— Dispatch AI (FZE), Sharjah UAE