fix: remove blocking warmup to pass HF health check

streamlit/app.py

@@ -1,4 +1,4 @@
-import sys, os, time
+import sys, os, time  # time kept for perf logging in inference calls
 sys.path.append(os.path.dirname(os.path.dirname(__file__)))   # import inference/
 
 import streamlit as st
@@ -9,68 +9,18 @@ log = get_logger("app")
 
 st.set_page_config(page_title="Yugensys Vision AI — Manufacturing", layout="wide")
 
-# ---------------------------------------------------------------------------
-# Imports — deferred so the page renders before any heavy loading starts
-# ---------------------------------------------------------------------------
 from inference.detect import detect_assets, _model as get_poc1_model
 from inference.classify import classify_qc, get_efficientnet, get_yolo
 
 # ---------------------------------------------------------------------------
-# Model warm-up — runs AFTER the first render so HF health check passes
-# ---------------------------------------------------------------------------
-STEPS = [
-    (get_poc1_model,    "Loading POC1 · Asset Detection model…",   0.00, 0.34),
-    (get_efficientnet,  "Loading POC2 · EfficientNet classifier…", 0.34, 0.67),
-    (get_yolo,          "Loading POC2 · YOLOv11 fallback…",        0.67, 1.00),
-]
-
-def models_ready() -> bool:
-    """True once all three @st.cache_resource loaders have been called."""
-    return st.session_state.get("models_loaded", False)
-
-def run_warmup():
-    """Blocking warm-up with progress bar — only called when models aren't cached yet."""
-    log.info("=== warm-up started ===")
-    t_total = time.perf_counter()
-
-    container = st.empty()
-    with container.container():
-        st.markdown("### ⚙️ Starting Yugensys Vision AI…")
-        bar   = st.progress(0.0)
-        label = st.empty()
-
-        for loader, msg, start_pct, end_pct in STEPS:
-            label.caption(msg)
-            bar.progress(start_pct)
-            t0 = time.perf_counter()
-            loader()
-            elapsed = time.perf_counter() - t0
-            log.info("warm-up step done: %s  (%.2fs)", msg, elapsed)
-            bar.progress(end_pct)
-
-        label.caption("✅  All models loaded — ready!")
-        time.sleep(0.5)
-
-    container.empty()
-    st.session_state["models_loaded"] = True
-    log.info("=== warm-up complete  total=%.2fs ===", time.perf_counter() - t_total)
-
-# ---------------------------------------------------------------------------
-# Sidebar + main UI — rendered on EVERY run (including the very first one)
-# so HF health check gets a valid HTTP response immediately
+# Sidebar + main UI — models load lazily on-demand via @st.cache_resource
 # ---------------------------------------------------------------------------
 st.sidebar.title("Yugensys Vision AI")
 screen = st.sidebar.radio(
     "Select POC",
     ["Asset Identification", "Quality Inspection (Auto Fallback)"],
-    disabled=not models_ready(),   # grey out nav until models are ready
 )
 
-# Now do the blocking warm-up (only on cache miss — fast no-op afterwards)
-if not models_ready():
-    run_warmup()
-    st.rerun()   # re-render with nav enabled and warm-up UI gone
-
 # ---------------------------------------------------------------------------
 # POC screens
 # ---------------------------------------------------------------------------