from transformers import AutoTokenizer, AutoModel
from modeling import BiEncoderModelRegression

# Load model components
tokenizer = AutoTokenizer.from_pretrained("minoosh/bert-reg-biencoder-mse")
base_model = AutoModel.from_pretrained("bert-base-uncased")
model = BiEncoderModelRegression(base_model, loss_fn="mse")

# Load weights
state_dict = torch.load("pytorch_model.bin")
model.load_state_dict(state_dict)

# Prepare inputs
texts1 = ["first text"]
texts2 = ["second text"]
inputs = tokenizer(
    texts1, texts2,
    padding=True,
    truncation=True,
    return_tensors="pt"
)

# Get similarity scores
outputs = model(**inputs)
similarity_scores = outputs["logits"]

Metrics

The model was trained using mse loss and evaluated using:

Mean Squared Error (MSE)
Mean Absolute Error (MAE)
Pearson Correlation
Spearman Correlation
Cosine Similarity