Create app.py

app.py

@@ -0,0 +1,52 @@
+# app.py
+import streamlit as st
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+import numpy as np
+import json
+
+@st.cache_resource
+def load_model():
+    model = AutoModelForSequenceClassification.from_pretrained("your-model-path")
+    tokenizer = AutoTokenizer.from_pretrained("your-model-path")
+    return tokenizer, model
+
+def get_top95(labels, probs):
+    sorted_indices = torch.argsort(probs, descending=True)
+    sorted_probs = probs[sorted_indices]
+    sorted_labels = labels[sorted_indices]
+
+    cumulative = torch.cumsum(sorted_probs, dim=0)
+    cutoff = torch.where(cumulative >= 0.95)[0]
+    last_idx = cutoff[0].item() + 1 if len(cutoff) > 0 else len(sorted_probs)
+
+    return list(zip(sorted_labels[:last_idx], sorted_probs[:last_idx].tolist()))
+
+# UI
+st.set_page_config(page_title="Article Topic Classifier")
+st.title("🔬 Article Topic Classifier")
+st.markdown("Enter the **title** and optionally **abstract** of the article.")
+
+title = st.text_input("Title", placeholder="e.g. Neural Networks for Quantum Physics")
+abstract = st.text_area("Abstract (optional)", placeholder="e.g. We explore the application of neural nets...")
+
+if st.button("Classify"):
+    if not title and not abstract:
+        st.warning("Please enter at least the title.")
+    else:
+        tokenizer, model = load_model()
+
+        text = title + ". " + abstract if abstract else title
+        inputs = tokenizer(text, return_tensors="pt", truncation=True)
+        with torch.no_grad():
+            outputs = model(**inputs)
+            probs = torch.nn.functional.softmax(outputs.logits[0], dim=-1)
+
+        with open("labels.json") as f:
+            id2label = json.load(f)
+
+        top_labels = get_top95(id2label, probs)
+
+        st.subheader("📚 Top topics (95% confidence)")
+        for label, prob in top_labels:
+            st.markdown(f"- **{label}**: {prob:.3f}")