PubMedBERT Embeddings Matryoshka - ONNX - O4

O4 optimized weights of NeuML/pubmedbert-base-embeddings-matryoshka.
Usage

from optimum.onnxruntime import ORTModelForFeatureExtraction
from transformers import AutoTokenizer
import torch

# Mean Pooling - Take attention mask into account for correct averaging
def meanpooling(output, mask):
    embeddings = output[0] # First element of model_output contains all token embeddings
    mask = mask.unsqueeze(-1).expand(embeddings.size()).float()
    return torch.sum(embeddings * mask, 1) / torch.clamp(mask.sum(1), min=1e-9)

# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

model = ORTModelForFeatureExtraction.from_pretrained("hooman650/pubmedbert-base-embeddings-matryoshka-onnx-04",provider="CUDAExecutionProvider")
tokenizer = AutoTokenizer.from_pretrained("hooman650/pubmedbert-base-embeddings-matryoshka-onnx-04")

# Tokenize sentences
inputs = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt').to("cude") # if on GPU

# Compute token embeddings
with torch.no_grad():
    output = model(**inputs)

# Perform pooling. In this case, mean pooling.
embeddings = meanpooling(output, inputs['attention_mask'])

# Requested matryoshka dimensions
dimensions = 256

print("Sentence embeddings:")
print(embeddings[:, :dimensions])