Update app.py

app.py

@@ -1,47 +1,97 @@
+import os
+import re
+from functools import lru_cache
+
 import gradio as gr
-from sentence_transformers import SentenceTransformer, util
-from transformers import pipeline
 
-# --- MODELS ---
-embedder = SentenceTransformer("all-MiniLM-L6-v2")
-llm = pipeline("text-generation", model="gpt2", max_new_tokens=60)
+# --- Space-friendly settings ---
+# Keep caches persistent if you enabled Space storage
+os.environ.setdefault("HF_HOME", "/data/huggingface")
+os.environ.setdefault("GRADIO_ANALYTICS_ENABLED", "false")
+
+# ---------- Lazy model loaders (no heavy work at import time) ----------
+@lru_cache(maxsize=1)
+def get_embedder():
+    # lazy import to avoid blocking the frontend
+    from sentence_transformers import SentenceTransformer
+    return SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+
+@lru_cache(maxsize=1)
+def get_ref_emb():
+    from sentence_transformers import util  # imported where needed
+    embedder = get_embedder()
+    reference_text = "Patient diagnosed with anaemia and low haemoglobin."
+    return embedder.encode(reference_text, convert_to_tensor=True)
 
-# --- RULE-BASED FUNCTION ---
-def rule_based(text):
-    if "anaemia" in text.lower() or "anemia" in text.lower():
-        return "✅ Anaemia detected (keyword match)"
-    else:
-        return "❌ Anaemia not detected (no keyword match)"
+@lru_cache(maxsize=1)
+def get_llm():
+    from transformers import pipeline
+    # keep it light; GPT-2 works on CPU
+    return pipeline("text-generation", model="gpt2", max_new_tokens=80)
 
-# --- MACHINE LEARNING FUNCTION ---
-reference_text = "Patient diagnosed with anaemia and low haemoglobin."
-ref_emb = embedder.encode(reference_text, convert_to_tensor=True)
+# ------------------------ Functions per approach ------------------------
+def rule_based(text: str) -> str:
+    if re.search(r"\banae?mia\b", text, flags=re.IGNORECASE):
+        return "Anaemia detected (keyword match)"
+    return "Anaemia not detected (no keyword match)"
 
-def ml_based(text):
+def ml_based(text: str) -> str:
+    # load lazily (first call)
+    from sentence_transformers import util
+    embedder = get_embedder()
+    ref_emb = get_ref_emb()
     emb = embedder.encode(text, convert_to_tensor=True)
-    score = util.cos_sim(emb, ref_emb).item()
-    return f"Similarity to anaemia reference: {score:.2f}\n" + \
-           ("✅ Anaemia likely" if score > 0.45 else "❌ Anaemia unlikely")
+    score = float(util.cos_sim(emb, ref_emb).item())
+    verdict = "Anaemia likely" if score > 0.45 else "Anaemia unlikely"
+    return f"Similarity: {score:.2f}\n{verdict}"
+
+def ai_based(text: str) -> str:
+    llm = get_llm()
+    prompt = (
+        "Determine if the following clinical note suggests anaemia. "
+        "Answer clearly in one short paragraph with reasoning.\n\n"
+        f"{text}\n"
+    )
+    out = llm(prompt)[0]["generated_text"]
+    return out
+
+# ----------------------------- UI -----------------------------
+def build_ui():
+    with gr.Blocks(title="RB vs ML vs AI — Anaemia Demo", theme=gr.themes.Soft()) as demo:
+        gr.Markdown("## Clinical Text Understanding — Three Approaches")
+        with gr.Row():
+            default_note = "The patient presents with fatigue and very low haemoglobin."
+            input_box = gr.Textbox(value=default_note, label="Clinical note", lines=4)
 
-# --- AI / FOUNDATION MODEL FUNCTION ---
-def ai_based(text):
-    prompt = f"Determine if the following note suggests anaemia:\n\n{text}\n\nAnswer clearly with reasoning."
-    result = llm(prompt)[0]["generated_text"]
-    return result
+        with gr.Tabs():
+            with gr.Tab("Rule-Based"):
+                rb_btn = gr.Button("Run Rule-Based")
+                rb_out = gr.Textbox(label="Result", lines=3)
+                rb_btn.click(fn=rule_based, inputs=input_box, outputs=rb_out)
 
-# --- GRADIO UI ---
-with gr.Blocks(title="Rule-Based vs Machine Learning vs AI") as demo:
-    gr.Markdown("# 🔍 Clinical Text Understanding Demo\nCompare rule-based, ML, and AI approaches for detecting **anaemia**.")
-    text_input = gr.Textbox(label="Enter clinical note", lines=4)
-    rb_output = gr.Textbox(label="Rule-Based Result")
-    ml_output = gr.Textbox(label="Machine Learning Result")
-    ai_output = gr.Textbox(label="AI (LLM) Result")
+            with gr.Tab("Machine Learning (Embeddings)"):
+                ml_btn = gr.Button("Run ML")
+                ml_out = gr.Textbox(label="Result", lines=4)
+                ml_btn.click(fn=ml_based, inputs=input_box, outputs=ml_out)
 
-    def process(text):
-        return rule_based(text), ml_based(text), ai_based(text)
+            with gr.Tab("AI / Foundation Model"):
+                ai_btn = gr.Button("Run AI")
+                ai_out = gr.Textbox(label="Result", lines=8)
+                ai_btn.click(fn=ai_based, inputs=input_box, outputs=ai_out)
 
-    btn = gr.Button("Run All")
-    btn.click(process, inputs=text_input, outputs=[rb_output, ml_output, ai_output])
+        gr.Markdown(
+            "Notes: models are loaded lazily on first run to keep the UI responsive in Spaces."
+        )
+    return demo
 
-demo.launch(server_name="0.0.0.0", server_port=7860, ssr_mode=False, show_error=True)
+demo = build_ui()
 
+# Queue + launch: SSR off is important for Spaces that show a blank/broken view.
+demo.queue(concurrency_count=1, max_size=10).launch(
+    server_name="0.0.0.0",
+    server_port=7860,
+    ssr_mode=False,
+    show_error=True,
+    inbrowser=False,
+    share=False,
+)