sentinal-ai / src /models /bert_model.py
furged's picture
Sentinel AI - clean HF deployment
1d6bb40
Raw
History Blame Contribute Delete
914 Bytes
from transformers import pipeline
print("Loading DistilBERT sentiment model (first run downloads ~260MB)...")
classifier = pipeline(
"sentiment-analysis",
model="distilbert-base-uncased-finetuned-sst-2-english",
device=-1 # force CPU; avoids trying to find a GPU that isn't there
# on most deploy hosts, which can otherwise raise or stall
)
print("DistilBERT model loaded.")
def get_bert_prediction(text):
result = classifier(text)[0]
label = result["label"]
confidence = result["score"]
prediction = 1 if label == "POSITIVE" else 0
entropy = 1 - confidence
return prediction, confidence, entropy
if __name__ == "__main__":
text = "this movie was amazing"
prediction, confidence, entropy = (
get_bert_prediction(text)
)
print("Prediction:", prediction)
print("Confidence:", confidence)
print("Entropy:", entropy)