Create app.py

app.py

@@ -0,0 +1,28 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForMaskedLM
+import torch, numpy as np
+from scipy.fft import fft, fftfreq
+
+phi = (1 + np.sqrt(5)) / 2
+tok = AutoTokenizer.from_pretrained("antonypamo/ProSavantEngine_Phi9_3")
+model = AutoModelForMaskedLM.from_pretrained("antonypamo/ProSavantEngine_Phi9_3")
+
+def phi_score(text):
+    inputs = tok(text, return_tensors="pt")
+    with torch.no_grad():
+        outs = model(**inputs, output_hidden_states=True)
+    h = torch.stack(outs.hidden_states).mean(dim=0).squeeze(0).cpu().numpy()
+    s = np.abs(fft(h.mean(axis=1)))
+    f = fftfreq(len(s), d=1.0)[:len(s)//2]
+    a = s[:len(f)]
+    w = np.cos(f*np.pi/phi)**2
+    sc = np.dot(a,w)/(np.linalg.norm(a)*np.linalg.norm(w))
+    return float((sc+1)/2)
+
+demo = gr.Interface(
+    fn=phi_score,
+    inputs=gr.Textbox(label="Enter text"),
+    outputs=gr.Number(label="Φ-weighted coherence (0–1)"),
+    title="ProSavantEngine Φ9.3 — Resonance Analyzer"
+)
+demo.launch()