Add example usage script
Browse files- example_usage.py +40 -0
example_usage.py
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
import torch
|
| 3 |
+
from transformers import AutoModel, AutoTokenizer
|
| 4 |
+
|
| 5 |
+
# Load model and tokenizer
|
| 6 |
+
model_name = "SURIYA-KP/nomic-embed-text-v2-moe-fine-tuned-depression-symptoms"
|
| 7 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 8 |
+
model = AutoModel.from_pretrained(model_name)
|
| 9 |
+
|
| 10 |
+
# Function to get embeddings
|
| 11 |
+
def get_embedding(text):
|
| 12 |
+
# Tokenize
|
| 13 |
+
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
|
| 14 |
+
|
| 15 |
+
# Get model output
|
| 16 |
+
with torch.no_grad():
|
| 17 |
+
outputs = model(**inputs)
|
| 18 |
+
|
| 19 |
+
# Mean pooling
|
| 20 |
+
token_embeddings = outputs.last_hidden_state
|
| 21 |
+
attention_mask = inputs['attention_mask']
|
| 22 |
+
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
|
| 23 |
+
embedding = torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
|
| 24 |
+
|
| 25 |
+
return embedding.numpy()[0]
|
| 26 |
+
|
| 27 |
+
# Example usage
|
| 28 |
+
text1 = "I feel worthless and useless."
|
| 29 |
+
text2 = "I am feeling happy and content today."
|
| 30 |
+
|
| 31 |
+
emb1 = get_embedding(text1)
|
| 32 |
+
emb2 = get_embedding(text2)
|
| 33 |
+
|
| 34 |
+
# Calculate cosine similarity
|
| 35 |
+
cos_sim = torch.nn.functional.cosine_similarity(
|
| 36 |
+
torch.tensor(emb1).unsqueeze(0),
|
| 37 |
+
torch.tensor(emb2).unsqueeze(0)
|
| 38 |
+
).item()
|
| 39 |
+
|
| 40 |
+
print(f"Cosine similarity between texts: {cos_sim:.4f}")
|