Update app.py

app.py

@@ -1,8 +1,7 @@
 """
 Amazon Trailer Inspector
-HuggingFace Spaces + Gradio pipeline
+HuggingFace Spaces + Gradio 5.x pipeline
 Gemma-3 (primary) → Llama-3.2-Vision → Qwen2.5-VL (fallbacks)
-Parallel multi-image inference, clean results UI.
 """
 
 import gradio as gr
@@ -13,15 +12,15 @@ import re
 import os
 from PIL import Image
 import io
-from huggingface_hub import InferenceClient
+from huggingface_hub import InferenceClient, HfApi
 
 # ──────────────────────────────────────────────────────────────
-# Model chain  (Gemma first, automatic fallback)
+# Model chain  (tried in order, first success wins)
 # ──────────────────────────────────────────────────────────────
 MODELS = [
-    "google/gemma-3-4b-it",                        # Primary  – Gemma 3 multimodal (free)
-    "meta-llama/Llama-3.2-11B-Vision-Instruct",    # Fallback 1
-    "Qwen/Qwen2.5-VL-7B-Instruct",                 # Fallback 2
+    "meta-llama/Llama-3.2-11B-Vision-Instruct",  # Most reliable free vision model
+    "Qwen/Qwen2.5-VL-7B-Instruct",               # Fallback 1
+    "google/gemma-3-4b-it",                       # Fallback 2
 ]
 
 DETECTION_PROMPT = """You are a precise visual inspector for Amazon trailer fleets.
@@ -32,7 +31,7 @@ Carefully examine the trailer image and locate these 4 components:
 3. PRIME_LOGO — The Amazon Prime logo: blue swooping arrow/checkmark. Can be full or partial, on rear or side.
 4. TRAILER_ID — A vertical fluorescent green or yellow-green ID label strip on the corner post (shows a number like SV2602705).
 
-Reply ONLY with valid JSON — absolutely no extra text, no markdown code fences:
+Reply ONLY with valid JSON — no extra text, no markdown fences:
 {
   "sensors":    {"found": true,  "confidence": "high",   "notes": "two diamond plates visible lower-left"},
   "gps_device": {"found": false, "confidence": "medium", "notes": "top corner obscured"},
@@ -42,52 +41,111 @@ Reply ONLY with valid JSON — absolutely no extra text, no markdown code fences
 
 KEYS = ["sensors", "gps_device", "prime_logo", "trailer_id"]
 
+
 # ──────────────────────────────────────────────────────────────
-# Vision helpers
+# Token validation (runs once at startup)
 # ──────────────────────────────────────────────────────────────
 
-def pil_to_b64(img: Image.Image, max_side: int = 1120) -> str:
+def check_token() -> tuple[bool, str]:
+    token = os.environ.get("HF_TOKEN", "").strip()
+    if not token:
+        return False, "HF_TOKEN secret is not set. Go to Space Settings → Repository Secrets → add HF_TOKEN."
+    try:
+        api = HfApi(token=token)
+        api.whoami()
+        return True, "Token OK"
+    except Exception as e:
+        return False, f"HF_TOKEN is invalid or expired: {e}"
+
+TOKEN_OK, TOKEN_MSG = check_token()
+
+
+# ──────────────────────────────────────────────────────────────
+# Image helpers
+# ──────────────────────────────────────────────────────────────
+
+def pil_to_b64(img: Image.Image, max_side: int = 1024) -> str:
+    """Resize and encode to base64 JPEG."""
     if max(img.size) > max_side:
         img = img.copy()
         img.thumbnail((max_side, max_side), Image.LANCZOS)
     buf = io.BytesIO()
-    img.save(buf, format="JPEG", quality=88)
+    img.save(buf, format="JPEG", quality=85)
     return base64.b64encode(buf.getvalue()).decode()
 
 
+# ──────────────────────────────────────────────────────────────
+# LLM call — with detailed error capture
+# ──────────────────────────────────────────────────────────────
+
 def call_model(img: Image.Image, model: str) -> dict:
-    """One LLM call — raises on failure."""
-    token  = os.environ.get("HF_TOKEN")
-    client = InferenceClient(model=model, token=token)
-    b64    = pil_to_b64(img)
+    """Call one vision LLM. Raises ValueError with a descriptive message on failure."""
+    token = os.environ.get("HF_TOKEN", "").strip() or None
+    client = InferenceClient(token=token)
+    b64 = pil_to_b64(img)
 
     messages = [{
         "role": "user",
         "content": [
-            {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
-            {"type": "text", "text": DETECTION_PROMPT},
+            {
+                "type": "image_url",
+                "image_url": {"url": f"data:image/jpeg;base64,{b64}"},
+            },
+            {
+                "type": "text",
+                "text": DETECTION_PROMPT,
+            },
         ],
     }]
 
-    resp = client.chat_completion(messages=messages, max_tokens=512, temperature=0.05)
-    raw  = resp.choices[0].message.content.strip()
+    resp = client.chat_completion(
+        model=model,
+        messages=messages,
+        max_tokens=512,
+        temperature=0.05,
+    )
+    raw = resp.choices[0].message.content.strip()
+
+    # Strip accidental markdown fences
+    raw = re.sub(r"^```(?:json)?", "", raw).strip()
+    raw = re.sub(r"```$", "", raw).strip()
 
-    m = re.search(r'\{[\s\S]*\}', raw)
+    m = re.search(r"\{[\s\S]*\}", raw)
     if not m:
-        raise ValueError(f"No JSON in response: {raw[:200]}")
-    return json.loads(m.group())
+        raise ValueError(f"Model returned no JSON.\nRaw output: {raw[:300]}")
+
+    try:
+        return json.loads(m.group())
+    except json.JSONDecodeError as e:
+        raise ValueError(f"JSON parse error: {e}\nRaw: {m.group()[:300]}")
 
 
 def analyze_one(img: Image.Image) -> tuple:
-    """Try models in order. Returns (result_dict_or_None, model_name_or_error)."""
-    last_err = "no models tried"
+    """
+    Try each model in MODELS order.
+    Returns (result_dict, model_short_name) on success,
+            (None, error_summary_string)     on total failure.
+    """
+    attempt_log = []
     for model in MODELS:
+        short = model.split("/")[-1]
         try:
             result = call_model(img, model)
-            return result, model.split("/")[-1]
+            return result, short
         except Exception as e:
-            last_err = f"{model.split('/')[-1]}: {e}"
-    return None, last_err
+            msg = str(e)
+            # Shorten common HTTP error noise
+            if "429" in msg:
+                msg = "rate-limited (429)"
+            elif "401" in msg or "403" in msg:
+                msg = "auth error — check HF_TOKEN"
+            elif "404" in msg:
+                msg = "model not found (404)"
+            elif "503" in msg or "502" in msg:
+                msg = "model loading / unavailable"
+            attempt_log.append(f"{short}: {msg}")
+
+    return None, " | ".join(attempt_log)
 
 
 # ──────────────────────────────────────────────────────────────
@@ -95,15 +153,8 @@ def analyze_one(img: Image.Image) -> tuple:
 # ──────────────────────────────────────────────────────────────
 
 def merge(results: list) -> dict:
-    """
-    Union across all images:
-    - component is FOUND if any image found it
-    - highest confidence wins
-    - first non-empty notes kept
-    """
     RANK = {"high": 3, "medium": 2, "low": 1, "": 0}
     merged = {k: {"found": False, "confidence": "low", "notes": ""} for k in KEYS}
-
     for res in results:
         if not res:
             continue
@@ -119,14 +170,10 @@ def merge(results: list) -> dict:
 
 
 # ──────────────────────────────────────────────────────────────
-# Main pipeline function (called by Gradio)
+# Load images from Gradio 5.x file paths
 # ──────────────────────────────────────────────────────────────
 
-def load_images(file_paths):
-    """
-    HF Spaces Gradio 5.x: gr.File(type='filepath') returns list[str].
-    Handles string paths and legacy file-object fallback.
-    """
+def load_images(file_paths) -> list:
     imgs = []
     if not file_paths:
         return imgs
@@ -134,28 +181,31 @@ def load_images(file_paths):
         file_paths = [file_paths]
     for p in file_paths:
         try:
-            path = p if isinstance(p, str) else (getattr(p, "name", None) or str(p))
+            path = p if isinstance(p, str) else getattr(p, "name", str(p))
             imgs.append(Image.open(path).convert("RGB"))
         except Exception as e:
             print(f"[load] skipped {p}: {e}")
     return imgs
 
 
-def analyze(file_paths):
-    """
-    Main Gradio callback.
-    Returns: (result_html: str, status_html: str)
-    """
-    images = load_images(file_paths)
+# ──────────────────────────────────────────────────────────────
+# Main Gradio callback
+# ──────────────────────────────────────────────────────────────
 
-    if not images:
+def analyze(file_paths):
+    # ── Token guard ──
+    if not TOKEN_OK:
         return (
-            _placeholder(),
-            _status("idle"),
+            _error(f"<b>Setup required:</b> {TOKEN_MSG}"),
+            _status("error"),
         )
 
+    images = load_images(file_paths)
+    if not images:
+        return _placeholder(), _status("idle")
+
     n = len(images)
-    all_results, errors, models_used = [], [], set()
+    all_results, all_errors, models_used = [], [], set()
 
     with concurrent.futures.ThreadPoolExecutor(max_workers=min(n, 4)) as pool:
         futs = [pool.submit(analyze_one, img) for img in images]
@@ -165,23 +215,29 @@ def analyze(file_paths):
                 all_results.append(res)
                 models_used.add(meta)
             else:
-                errors.append(meta)
+                all_errors.append(meta)
 
     if not all_results:
+        # Show the REAL error from each model attempt
+        err_detail = "<br>".join(all_errors) if all_errors else "Unknown error"
         return (
-            _error("Analysis failed — all models returned errors.<br>"
-                   "Make sure <b>HF_TOKEN</b> is set in Space Secrets."),
+            _error(
+                f"<b>All models failed.</b><br><br>"
+                f"<code style='font-size:12px;line-height:1.8;'>{err_detail}</code><br><br>"
+                f"Common causes:<br>"
+                f"• HF_TOKEN missing/expired → Space Settings → Secrets<br>"
+                f"• Models overloaded (rate limit 429) → retry in a minute<br>"
+                f"• Image too large → try a smaller/compressed photo"
+            ),
             _status("error"),
         )
 
     merged    = merge(all_results)
     model_str = " · ".join(sorted(models_used)) or "AI"
-    warn      = (f"<br><small style='color:#d97706;'>⚠️ {len(errors)} image(s) failed</small>"
-                 if errors else "")
-    result_h  = build_cards(merged, n, model_str, warn)
-    status_h  = _status("done", n, len(all_results))
+    warn      = (f"<br><small style='color:#d97706;'>⚠️ {len(all_errors)} image(s) failed: "
+                 f"{all_errors[0][:80]}</small>" if all_errors else "")
 
-    return result_h, status_h
+    return build_cards(merged, n, model_str, warn), _status("done", n, len(all_results))
 
 
 # ──────────────────────────────────────────────────────────────
@@ -214,12 +270,12 @@ def build_cards(merged: dict, img_n: int, model_str: str, warn: str) -> str:
         conf  = d.get("confidence", "low")
         notes = d.get("notes", "")
 
-        rbg  = "#f0fdf4" if found else "#fef2f2"
-        rbd  = "#bbf7d0" if found else "#fecaca"
-        stc  = "#15803d" if found else "#b91c1c"
-        stx  = "✅ Found"  if found else "❌ Missing"
-        cdc  = {"high":"#16a34a","medium":"#d97706","low":"#dc2626"}.get(conf,"#9ca3af")
-        note = (
+        rbg = "#f0fdf4" if found else "#fef2f2"
+        rbd = "#bbf7d0" if found else "#fecaca"
+        stc = "#15803d" if found else "#b91c1c"
+        stx = "✅ Found" if found else "❌ Missing"
+        cdc = {"high": "#16a34a", "medium": "#d97706", "low": "#dc2626"}.get(conf, "#9ca3af")
+        note_html = (
             f'<div style="margin-top:8px;padding-top:8px;border-top:1px solid {rbd};'
             f'font-size:12px;color:#4b5563;font-style:italic;line-height:1.5;">"{notes}"</div>'
             if notes else ""
@@ -234,7 +290,7 @@ def build_cards(merged: dict, img_n: int, model_str: str, warn: str) -> str:
     <div style="flex:1;min-width:0;">
       <div style="font-weight:700;font-size:14px;color:#111827;">{name}</div>
       <div style="font-size:11px;color:#9ca3af;margin-top:1px;">{desc}</div>
-      {note}
+      {note_html}
     </div>
     <div style="text-align:right;flex-shrink:0;padding-left:8px;">
       <div style="font-weight:700;color:{stc};font-size:13px;white-space:nowrap;">{stx}</div>
@@ -253,7 +309,7 @@ def build_cards(merged: dict, img_n: int, model_str: str, warn: str) -> str:
         {si} {found_n}/{total} — {sl}
       </div>
       <div style="font-size:12px;color:#6b7280;margin-top:3px;">
-        {img_n} image{'s' if img_n>1 else ''} · {model_str}{warn}
+        {img_n} image{'s' if img_n > 1 else ''} · {model_str}{warn}
       </div>
     </div>
     <div style="font-size:36px;">🚛</div>
@@ -264,8 +320,7 @@ def build_cards(merged: dict, img_n: int, model_str: str, warn: str) -> str:
 
 def _placeholder() -> str:
     return """
-<div style="text-align:center;padding:60px 20px;color:#94a3b8;
-            font-family:-apple-system,sans-serif;">
+<div style="text-align:center;padding:60px 20px;color:#94a3b8;font-family:sans-serif;">
   <div style="font-size:48px;margin-bottom:14px;">📷</div>
   <div style="font-size:15px;font-weight:600;color:#64748b;">Upload trailer images to begin</div>
   <div style="font-size:13px;margin-top:6px;">Front view, rear view, or both — all work</div>
@@ -275,31 +330,50 @@ def _placeholder() -> str:
 def _status(state: str, total: int = 0, ok: int = 0) -> str:
     msgs = {
         "idle":  ("🟡", "#d97706", "Waiting for images"),
-        "done":  ("🟢", "#16a34a", f"{ok}/{total} image{'s' if total>1 else ''} processed"),
-        "error": ("🔴", "#dc2626", "Analysis failed — check HF_TOKEN secret"),
+        "done":  ("🟢", "#16a34a", f"{ok}/{total} image{'s' if total > 1 else ''} processed"),
+        "error": ("🔴", "#dc2626", "See error details →"),
     }
     icon, color, text = msgs.get(state, msgs["idle"])
     return (
-        f'<div style="font-size:12px;color:{color};text-align:center;'
-        f'padding:6px 0 2px;">{icon} {text}</div>'
+        f'<div style="font-size:12px;color:{color};text-align:center;padding:6px 0 2px;">'
+        f'{icon} {text}</div>'
     )
 
 
 def _error(msg: str) -> str:
     return (
         f'<div style="background:#fef2f2;border:1.5px solid #fca5a5;border-radius:12px;'
-        f'padding:20px;color:#b91c1c;font-family:sans-serif;font-size:14px;">'
-        f'⚠️ {msg}</div>'
+        f'padding:18px 20px;color:#b91c1c;font-family:sans-serif;font-size:13px;line-height:1.7;">'
+        f'{msg}</div>'
     )
 
 
+# ──────────────────────────────────────────────────────────────
+# Startup banner (shown in Space logs)
+# ──────────────────────────────────────────────────────────────
+
+print("=" * 55)
+print("  Amazon Trailer Inspector — starting up")
+print(f"  Token status : {TOKEN_MSG}")
+print(f"  Models       : {[m.split('/')[-1] for m in MODELS]}")
+print("=" * 55)
+
+
 # ──────────────────────────────────────────────────────────────
 # Gradio UI
 # ──────────────────────────────────────────────────────────────
 
+TOKEN_BANNER = "" if TOKEN_OK else (
+    '<div style="background:#fef3c7;border:1.5px solid #fde68a;border-radius:10px;'
+    'padding:12px 16px;margin-bottom:14px;font-size:13px;color:#92400e;font-family:sans-serif;">'
+    '⚠️ <b>HF_TOKEN not set.</b> Go to Space <b>Settings → Repository Secrets</b> '
+    'and add <code>HF_TOKEN</code> with your HuggingFace Read token. '
+    'Get one free at <a href="https://huggingface.co/settings/tokens" target="_blank">'
+    'huggingface.co/settings/tokens</a></div>'
+)
+
 CSS = """
 .gradio-container { max-width: 980px !important; margin: auto !important; }
-#upload-box .wrap { border-radius: 12px !important; min-height: 120px; }
 #analyze-btn { font-size: 15px !important; font-weight: 700 !important;
                letter-spacing: .02em; border-radius: 10px !important; }
 footer { display: none !important; }
@@ -313,8 +387,7 @@ THEME = gr.themes.Soft(
 
 with gr.Blocks(title="🚛 Amazon Trailer Inspector", theme=THEME, css=CSS) as demo:
 
-    # ── Header ────────────────────────────────────────────────
-    gr.HTML("""
+    gr.HTML(f"""
 <div style="text-align:center;padding:30px 0 18px;font-family:sans-serif;">
   <div style="font-size:46px;margin-bottom:10px;">🚛</div>
   <h1 style="font-size:26px;font-weight:800;color:#0f172a;margin:0 0 6px;">
@@ -323,14 +396,12 @@ with gr.Blocks(title="🚛 Amazon Trailer Inspector", theme=THEME, css=CSS) as d
   <p style="color:#64748b;font-size:14px;margin:0;">
     AI-powered verification of required trailer components from photos
   </p>
-</div>""")
+</div>
+{TOKEN_BANNER}""")
 
-    # ── Two-column layout ─────────────────────────────────────
     with gr.Row(equal_height=False):
 
-        # Left – upload + checklist
         with gr.Column(scale=1, min_width=280):
-
             gr.HTML("""
 <div style="background:#f8fafc;border:1px solid #e2e8f0;border-radius:14px;
             padding:16px 18px;margin-bottom:14px;">
@@ -340,19 +411,19 @@ with gr.Blocks(title="🚛 Amazon Trailer Inspector", theme=THEME, css=CSS) as d
   </div>
   <div style="display:grid;gap:9px;font-size:13px;color:#334155;">
     <div style="display:flex;align-items:center;gap:10px;">
-      <span style="background:#fef3c7;border-radius:7px;padding:4px 9px;font-size:15px;">🔷</span>
+      <span style="background:#fef3c7;border-radius:7px;padding:4px 9px;">🔷</span>
       <span><b>Sensors</b> — two diamond-shaped plates</span>
     </div>
     <div style="display:flex;align-items:center;gap:10px;">
-      <span style="background:#dbeafe;border-radius:7px;padding:4px 9px;font-size:15px;">📡</span>
+      <span style="background:#dbeafe;border-radius:7px;padding:4px 9px;">📡</span>
       <span><b>GPS Device</b> — white box, top corner</span>
     </div>
     <div style="display:flex;align-items:center;gap:10px;">
-      <span style="background:#ede9fe;border-radius:7px;padding:4px 9px;font-size:15px;">🔵</span>
+      <span style="background:#ede9fe;border-radius:7px;padding:4px 9px;">🔵</span>
       <span><b>Prime Logo</b> — Amazon Prime mark</span>
     </div>
     <div style="display:flex;align-items:center;gap:10px;">
-      <span style="background:#d1fae5;border-radius:7px;padding:4px 9px;font-size:15px;">🏷️</span>
+      <span style="background:#d1fae5;border-radius:7px;padding:4px 9px;">🏷️</span>
       <span><b>Trailer ID</b> — corner post label strip</span>
     </div>
   </div>
@@ -362,8 +433,7 @@ with gr.Blocks(title="🚛 Amazon Trailer Inspector", theme=THEME, css=CSS) as d
                 label="Upload Trailer Image(s)",
                 file_count="multiple",
                 file_types=["image"],
-                type="filepath",          # HF Spaces: returns plain string paths
-                elem_id="upload-box",
+                type="filepath",
             )
 
             gr.HTML("""
@@ -380,25 +450,20 @@ with gr.Blocks(title="🚛 Amazon Trailer Inspector", theme=THEME, css=CSS) as d
 
             status_html = gr.HTML(_status("idle"))
 
-        # Right – results panel
         with gr.Column(scale=1, min_width=320):
             result_html = gr.HTML(_placeholder())
 
-    # ── Footer ────────────────────────────────────────────────
     gr.HTML("""
 <div style="text-align:center;padding:20px 0 10px;color:#94a3b8;
             font-size:12px;font-family:sans-serif;">
-  Gemma 3 · Llama 3.2 Vision · Qwen2.5-VL &nbsp;|&nbsp;
-  Images processed in parallel &nbsp;|&nbsp; No data is stored
+  Llama 3.2 Vision · Qwen2.5-VL · Gemma 3 &nbsp;|&nbsp;
+  Images processed in parallel &nbsp;|&nbsp; No data stored
 </div>""")
 
-    # ── Wiring ────────────────────────────────────────────────
     analyze_btn.click(
         fn=analyze,
         inputs=[file_input],
         outputs=[result_html, status_html],
     )
 
-
-# HF Spaces handles host/port — no arguments needed
 demo.launch()