Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from sentence_transformers import SentenceTransformer
|
2 |
+
import torch
|
3 |
+
import gradio as gr
|
4 |
+
from scipy.spatial.distance import cosine
|
5 |
+
|
6 |
+
# Disable CUDA
|
7 |
+
torch.backends.cudnn.enabled = False
|
8 |
+
torch.cuda.is_available = lambda : False
|
9 |
+
|
10 |
+
# Load model and tokenizer
|
11 |
+
modelname = "algolia/algolia-large-en-generic-v2410"
|
12 |
+
model = SentenceTransformer(modelname)
|
13 |
+
def get_embedding(text):
|
14 |
+
embedding = model.encode([text])
|
15 |
+
return embedding[0]
|
16 |
+
|
17 |
+
def compute_similarity(query, documents):
|
18 |
+
query_emb = get_embedding(query)
|
19 |
+
doc_embeddings = [get_embedding(doc) for doc in documents]
|
20 |
+
|
21 |
+
# Calculate cosine similarity
|
22 |
+
similarities = [1 - cosine(query_emb, doc_emb) for doc_emb in doc_embeddings]
|
23 |
+
ranked_docs = sorted(zip(documents, similarities), key=lambda x: x[1], reverse=True)
|
24 |
+
|
25 |
+
# Format output
|
26 |
+
return [{"document": doc, "similarity_score": round(sim, 4)} for doc, sim in ranked_docs]
|
27 |
+
|
28 |
+
# Gradio interface function
|
29 |
+
def gradio_compute_similarity(query, documents):
|
30 |
+
# Prefix the query string
|
31 |
+
query = "query: " + query
|
32 |
+
# Split documents by lines for the Gradio input
|
33 |
+
documents_list = documents.split("\n")
|
34 |
+
results = compute_similarity(query, documents_list)
|
35 |
+
return results
|
36 |
+
|
37 |
+
# Gradio Interface
|
38 |
+
iface = gr.Interface(
|
39 |
+
fn=gradio_compute_similarity,
|
40 |
+
inputs=[
|
41 |
+
gr.Textbox(label="Query", placeholder="Enter your query here"),
|
42 |
+
gr.Textbox(lines=5, label="Documents", placeholder="Enter a list of documents, one per line")
|
43 |
+
],
|
44 |
+
outputs=gr.JSON(label="Ranked Results"),
|
45 |
+
allow_flagging="never",
|
46 |
+
title="Document Similarity",
|
47 |
+
description="Provide a query and a list of documents. See the ranked similarity scores."
|
48 |
+
)
|
49 |
+
|
50 |
+
iface.launch()
|