Update app.py

app.py

@@ -1,7 +1,5 @@
 # ==============================================================================
-# 🦅 GarudaCoder-27B — Indonesian Coding Intelligence
-# Precision Engineered Interface (Hugging Face ZeroGPU Edition)
-# Fully Compatible with Gradio 6.x & ZeroGPU A100
+# 🦅 GarudaCoder-27B — Workspace (Native Gradio 6 + ZeroGPU A100)
 # ==============================================================================
 
 import os
@@ -28,6 +26,7 @@ from transformers import (
     TextIteratorStreamer,
 )
 from peft import PeftModel
+from huggingface_hub import snapshot_download
 
 try:
     from transformers import AutoModelForMultimodalLM
@@ -37,9 +36,6 @@ except ImportError:
     except ImportError:
         from transformers import AutoModelForVision2Seq as AutoModelForMultimodalLM
 
-# ==============================================================================
-# 1. KONFIGURASI MODEL & PROMPT
-# ==============================================================================
 BASE_ID = "unsloth/Qwen3.8-27B-unsloth-bnb-4bit"
 ADAPTER_ID = "TheCoderScientist/GarudaCoder-27B-ID-lora"
 EOS_IDS = [248046, 248044]
@@ -53,15 +49,26 @@ Prioritas:
 - Jangan mengarang fakta, API, versi, angka, dokumentasi, atau hasil eksekusi.
 - Bila informasi tidak cukup, katakan apa yang kurang.
 - Gunakan konteks gambar, video, file, atau hasil web search yang diberikan.
-- Bedakan fakta dari inferensi.
-- Untuk informasi yang berubah atau bersifat lokal, gunakan sumber yang dapat diverifikasi.
-- Jangan menampilkan chain-of-thought privat.
-- Jangan mengklaim telah menjalankan kode jika memang belum dijalankan.
 - Identitas kamu adalah GarudaCoder.
 - Jika ditanya basis model, jelaskan secara ringkas bahwa GarudaCoder menggunakan Qwen3.8-27B sebagai base model dengan adapter GarudaCoder.
 """
 
-print("[Init] Menyiapkan Processor & Tokenizer...")
+# ==============================================================================
+# 1. PRE-DOWNLOAD BOBOT MODEL KE DISK (DI STARTUP, BUKAN PAS USER CHAT)
+# ==============================================================================
+print("\n" + "=" * 80)
+print("[Startup] Memastikan bobot model 22.3 GB terunduh di disk...")
+print("=" * 80)
+
+# Unduh Base Model dan Adapter langsung saat container boot
+snapshot_download(
+    repo_id=BASE_ID,
+    allow_patterns=["*.json", "*.safetensors", "*.model", "*.jinja"],
+)
+snapshot_download(repo_id=ADAPTER_ID)
+print("[Startup] Semua bobot model siap di disk lokal.\n")
+
+print("[Startup] Memuat Processor & Tokenizer...")
 processor = AutoProcessor.from_pretrained(ADAPTER_ID, trust_remote_code=True)
 tokenizer = getattr(processor, "tokenizer", processor)
 
@@ -71,10 +78,10 @@ if getattr(tokenizer, "pad_token", None) is None:
 _model = None
 
 def get_model():
-    """ZeroGPU Lazy Loader: Muat model hanya saat inferensi diminta."""
+    """Memindahkan bobot dari disk lokal ke NVIDIA A100 ZeroGPU."""
     global _model
     if _model is None:
-        print("[ZeroGPU] Memuat bobot 4-bit ke NVIDIA A100...")
+        print("[ZeroGPU] Memuat bobot 4-bit ke VRAM A100...")
         bnb_config = BitsAndBytesConfig(
             load_in_4bit=True,
             bnb_4bit_quant_type="nf4",
@@ -87,7 +94,7 @@ def get_model():
             attn_implementation="sdpa",
             trust_remote_code=True,
         )
-        print("[ZeroGPU] Mengaktifkan adapter LoRA GarudaCoder...")
+        print("[ZeroGPU] Menempelkan adapter LoRA...")
         _model = PeftModel.from_pretrained(
             base_model,
             ADAPTER_ID,
@@ -98,7 +105,7 @@ def get_model():
     return _model
 
 # ==============================================================================
-# 2. FILE & MULTIMODAL PARSERS
+# 2. FILE EXTRACTOR & WEB SEARCH
 # ==============================================================================
 IMAGE_EXTENSIONS = {".png", ".jpg", ".jpeg", ".webp", ".bmp"}
 VIDEO_EXTENSIONS = {".mp4", ".mov", ".mkv", ".webm"}
@@ -117,92 +124,54 @@ def normalize_path(file_obj):
         return file_obj.get("path") or file_obj.get("name")
     return getattr(file_obj, "path", None) or getattr(file_obj, "name", None)
 
-def file_uri(path):
-    return Path(os.path.abspath(path)).as_uri()
-
-def extract_text_file(path, max_chars=40000):
-    try:
-        with open(path, "r", encoding="utf-8", errors="ignore") as f:
-            return f.read(max_chars)
-    except Exception as e:
-        return f"[Gagal membaca file teks: {e}]"
-
-def extract_pdf(path, max_chars=40000):
+def extract_text(path, max_chars=40000):
     try:
-        from pypdf import PdfReader
-        reader = PdfReader(path)
-        chunks = []
-        for page in reader.pages:
-            txt = page.extract_text() or ""
-            chunks.append(txt)
-            if sum(len(x) for x in chunks) >= max_chars:
-                break
-        return "\n\n".join(chunks)[:max_chars]
-    except Exception as e:
-        return f"[Gagal membaca PDF: {e}]"
-
-def extract_docx(path, max_chars=40000):
-    try:
-        from docx import Document
-        doc = Document(path)
-        parts = [p.text for p in doc.paragraphs if p.text.strip()]
-        return "\n".join(parts)[:max_chars]
-    except Exception as e:
-        return f"[Gagal membaca DOCX: {e}]"
-
-def extract_xlsx(path, max_chars=40000):
-    try:
-        from openpyxl import load_workbook
-        wb = load_workbook(path, read_only=True, data_only=True)
-        parts = []
-        for ws in wb.worksheets:
-            parts.append(f"## Sheet: {ws.title}")
-            for row in ws.iter_rows(values_only=True):
-                vals = ["" if v is None else str(v) for v in row]
-                parts.append(" | ".join(vals))
-                if sum(len(x) for x in parts) >= max_chars:
-                    break
-        return "\n".join(parts)[:max_chars]
+        ext = Path(path).suffix.lower()
+        if ext in TEXT_EXTENSIONS:
+            with open(path, "r", encoding="utf-8", errors="ignore") as f:
+                return f.read(max_chars)
+        elif ext == ".pdf":
+            from pypdf import PdfReader
+            reader = PdfReader(path)
+            return "\n\n".join([page.extract_text() or "" for page in reader.pages])[:max_chars]
+        elif ext == ".docx":
+            from docx import Document
+            doc = Document(path)
+            return "\n".join([p.text for p in doc.paragraphs if p.text.strip()])[:max_chars]
+        elif ext in {".xlsx", ".xlsm"}:
+            from openpyxl import load_workbook
+            wb = load_workbook(path, read_only=True, data_only=True)
+            rows = []
+            for ws in wb.worksheets:
+                for r in ws.iter_rows(values_only=True):
+                    rows.append(" | ".join(["" if v is None else str(v) for v in r]))
+            return "\n".join(rows)[:max_chars]
     except Exception as e:
-        return f"[Gagal membaca XLSX: {e}]"
+        return f"[Gagal membaca lampiran: {e}]"
+    return ""
 
-def inspect_uploaded_files(files):
+def inspect_files(files):
     vision_blocks = []
-    text_parts = []
-    summary = []
-
+    text_blocks = []
+    filenames = []
     for raw in files or []:
         path = normalize_path(raw)
         if not path or not os.path.exists(path):
             continue
-
         name = os.path.basename(path)
         ext = Path(path).suffix.lower()
+        filenames.append(name)
 
         if ext in IMAGE_EXTENSIONS:
-            vision_blocks.append({"type": "image", "image": file_uri(path)})
-            summary.append(name)
+            vision_blocks.append({"type": "image", "image": Path(os.path.abspath(path)).as_uri()})
         elif ext in VIDEO_EXTENSIONS:
-            vision_blocks.append({"type": "video", "video": file_uri(path), "fps": 1.0})
-            summary.append(name)
-        elif ext in TEXT_EXTENSIONS:
-            text_parts.append(f"### [FILE: {name}]\n```text\n{extract_text_file(path)}\n```")
-            summary.append(name)
-        elif ext == ".pdf":
-            text_parts.append(f"### [PDF: {name}]\n{extract_pdf(path)}")
-            summary.append(name)
-        elif ext == ".docx":
-            text_parts.append(f"### [DOCX: {name}]\n{extract_docx(path)}")
-            summary.append(name)
-        elif ext in {".xlsx", ".xlsm"}:
-            text_parts.append(f"### [SPREADSHEET: {name}]\n{extract_xlsx(path)}")
-            summary.append(name)
+            vision_blocks.append({"type": "video", "video": Path(os.path.abspath(path)).as_uri(), "fps": 1.0})
+        else:
+            t = extract_text(path)
+            if t:
+                text_blocks.append(f"### [FILE: {name}]\n```\n{t}\n```")
+    return vision_blocks, "\n\n".join(text_blocks), filenames
 
-    return vision_blocks, "\n\n".join(text_parts), summary
-
-# ==============================================================================
-# 3. FAST WEB SEARCH (DUCKDUCKGO)
-# ==============================================================================
 def web_search(query, max_results=4):
     query = (query or "").strip()
     if not query:
@@ -213,41 +182,16 @@ def web_search(query, max_results=4):
         with DDGS() as ddgs:
             for r in ddgs.text(query, max_results=max_results, region="wt-wt", safesearch="moderate"):
                 results.append({
-                    "title": r.get("title") or "Dokumentasi",
+                    "title": r.get("title") or "Rujukan",
                     "url": r.get("href") or "",
                     "snippet": r.get("body") or "",
                 })
         return results
-    except Exception as e:
-        print(f"[Web Search Warn] {e}")
+    except Exception:
         return []
 
-def format_search_context(results):
-    if not results:
-        return ""
-    chunks = ["## HASIL DOKUMENTASI/WEB TERBARU:"]
-    for i, r in enumerate(results, 1):
-        chunks.append(f"[{i}] {r['title']} ({r['url']})\nSnippet: {r['snippet']}")
-    return "\n\n".join(chunks)
-
-def sources_html(results):
-    if not results:
-        return '<div class="sources-empty">Pencarian web nonaktif atau belum ada kueri.</div>'
-    chips = []
-    for i, r in enumerate(results, 1):
-        title = gr.utils.sanitize_html(r["title"][:75])
-        url = r["url"]
-        domain = re.sub(r"^https?://(www\.)?", "", url).split("/")[0]
-        chips.append(
-            f'<a class="source-chip" href="{url}" target="_blank" title="{title}">'
-            f'<span class="source-domain">{domain}</span>'
-            f'<span class="source-title">{title}</span>'
-            f'</a>'
-        )
-    return f'<div class="sources-grid">{"".join(chips)}</div>'
-
 # ==============================================================================
-# 4. CHAT GENERATION ENGINE (ZEROGPU A100)
+# 3. CHAT GENERATION (ZEROGPU)
 # ==============================================================================
 @spaces.GPU(duration=120)
 def generate_chat(
@@ -265,13 +209,20 @@ def generate_chat(
     user_text = (message or {}).get("text", "").strip()
     files = (message or {}).get("files", [])
 
-    vision_blocks, file_context, file_names = inspect_uploaded_files(files)
+    vision_blocks, file_context, filenames = inspect_files(files)
 
     search_results = []
     if enable_web_search and user_text:
         search_results = web_search(user_text, max_results=4)
-    search_context = format_search_context(search_results)
 
+    search_context = ""
+    if search_results:
+        chunks = ["## HASIL PENCARIAN TERBARU:"]
+        for i, r in enumerate(search_results, 1):
+            chunks.append(f"[{i}] {r['title']} ({r['url']})\nSnippet: {r['snippet']}")
+        search_context = "\n\n".join(chunks)
+
+    # Siapkan pesan
     sys_content = system_prompt.strip() if system_prompt else SYSTEM_PROMPT
     messages = [{"role": "system", "content": sys_content}]
 
@@ -281,24 +232,24 @@ def generate_chat(
         if role in {"user", "assistant"} and isinstance(content, str):
             messages.append({"role": role, "content": content})
 
-    current_turn_content = []
-    current_turn_content.extend(vision_blocks)
+    current_turn = []
+    current_turn.extend(vision_blocks)
 
-    context_sections = []
+    contexts = []
     if file_context:
-        context_sections.append(file_context)
+        contexts.append(file_context)
     if search_context:
-        context_sections.append(search_context)
+        contexts.append(search_context)
 
-    if context_sections:
-        full_text = "\n\n".join(context_sections) + (f"\n\n## INSTRUKSI USER:\n{user_text}" if user_text else "")
-        current_turn_content.append({"type": "text", "text": full_text})
+    if contexts:
+        combined = "\n\n".join(contexts) + (f"\n\n## INSTRUKSI:\n{user_text}" if user_text else "")
+        current_turn.append({"type": "text", "text": combined})
     elif user_text:
-        current_turn_content.append({"type": "text", "text": user_text})
+        current_turn.append({"type": "text", "text": user_text})
     elif not vision_blocks:
-        current_turn_content.append({"type": "text", "text": "Halo!"})
+        current_turn.append({"type": "text", "text": "Halo!"})
 
-    messages.append({"role": "user", "content": current_turn_content})
+    messages.append({"role": "user", "content": current_turn})
 
     try:
         model_inputs = processor.apply_chat_template(
@@ -350,312 +301,97 @@ def generate_chat(
     thread.start()
 
     badges = []
-    if file_names:
-        badges.append(f"Attachment: {', '.join(file_names[:2])}{'...' if len(file_names) > 2 else ''}")
+    if filenames:
+        badges.append(f"📁 {len(filenames)} file")
     if search_results:
-        badges.append(f"Web Search: {len(search_results)} rujukan")
+        badges.append(f"🌐 {len(search_results)} sumber")
+    prefix = f"`{' · '.join(badges)}`\n\n" if badges else ""
 
-    meta_tag = f'<div class="context-tag">{" · ".join(badges)}</div>\n\n' if badges else ""
     output = ""
-
     for chunk in streamer:
         output += chunk
-        yield meta_tag + output, search_results
+        yield prefix + output
 
     thread.join()
-
     if errors:
-        yield output + f"\n\n> ⚠️ **Runtime Issue:** `{errors[0]}`", search_results
+        yield output + f"\n\n> ⚠️ **Error:** `{errors[0]}`"
 
 # ==============================================================================
-# 5. MODERN UI SYSTEM (LINEAR / CURSOR AESTHETICS)
+# 4. TAMPILAN BERSIH GRADIO 6 (TANPA CSS GLITCH)
 # ==============================================================================
-WELCOME_MD = """### Siap membantu audit kode dan arsitektur sistem.
+WELCOME_MD = """### 🦅 GarudaCoder-27B
+Asisten coding berbahasa Indonesia berbasis **Qwen3.8-27B** di atas akselerator **NVIDIA A100**.
 
-Pilih topik di bawah atau lampirkan file dokumen/gambar langsung:
+Mendukung analisis kode, diagram arsitektur, ekstraksi file dokumen/spreadsheet, dan pencarian web.
 """
 
-CSS = r"""
-@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');
-
-:root {
-    --bg-base: #0a0c10;
-    --bg-surface: #11141b;
-    --bg-subtle: #181d26;
-    --border-subtle: rgba(255, 255, 255, 0.07);
-    --border-active: rgba(255, 255, 255, 0.15);
-    --text-primary: #f0f3f6;
-    --text-secondary: #8c96a5;
-    --text-muted: #566171;
-    --accent: #10b981;
-    --font-sans: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
-    --font-mono: 'JetBrains Mono', monospace;
-}
-
-body, .gradio-container {
-    background-color: var(--bg-base) !important;
-    color: var(--text-primary) !important;
-    font-family: var(--font-sans) !important;
-    max-width: 1380px !important;
-    margin: 0 auto !important;
-}
-
-/* Header Navbar */
-.app-nav {
-    display: flex;
-    justify-content: space-between;
-    align-items: center;
-    padding: 18px 0;
-    border-bottom: 1px solid var(--border-subtle);
-    margin-bottom: 16px;
-}
-.brand-group {
-    display: flex;
-    align-items: center;
-    gap: 12px;
-}
-.brand-badge {
-    background: var(--bg-subtle);
-    border: 1px solid var(--border-subtle);
-    border-radius: 8px;
-    padding: 6px 9px;
-    font-size: 14px;
-    font-weight: 700;
-}
-.brand-name {
-    font-size: 16px;
-    font-weight: 600;
-    letter-spacing: -0.02em;
-    color: var(--text-primary);
-}
-.brand-desc {
-    font-size: 12px;
-    color: var(--text-secondary);
-}
-.status-pill {
-    display: inline-flex;
-    align-items: center;
-    gap: 6px;
-    font-size: 11px;
-    font-family: var(--font-mono);
-    color: var(--text-secondary);
-    background: var(--bg-surface);
-    border: 1px solid var(--border-subtle);
-    padding: 4px 10px;
-    border-radius: 9999px;
-}
-.status-dot {
-    width: 6px;
-    height: 6px;
-    border-radius: 50%;
-    background-color: var(--accent);
-}
-
-/* Chatbox Area */
-.chat-viewport {
-    background: var(--bg-surface) !important;
-    border: 1px solid var(--border-subtle) !important;
-    border-radius: 12px !important;
-}
-
-/* Source Reference Chips (Perplexity Style) */
-.sources-grid {
-    display: flex;
-    flex-wrap: wrap;
-    gap: 8px;
-    padding: 4px 0;
-}
-.source-chip {
-    display: inline-flex;
-    flex-direction: column;
-    background: var(--bg-subtle);
-    border: 1px solid var(--border-subtle);
-    border-radius: 8px;
-    padding: 6px 10px;
-    text-decoration: none !important;
-    transition: border-color 0.15s ease, background 0.15s ease;
-    max-width: 100%;
-}
-.source-chip:hover {
-    border-color: var(--border-active);
-    background: #1e2430;
-}
-.source-domain {
-    font-size: 10px;
-    font-family: var(--font-mono);
-    color: var(--accent);
-    text-transform: uppercase;
-    letter-spacing: 0.04em;
-}
-.source-title {
-    font-size: 11px;
-    font-weight: 500;
-    color: var(--text-primary) !important;
-    white-space: nowrap;
-    overflow: hidden;
-    text-overflow: ellipsis;
-    max-width: 240px;
-}
-.sources-empty {
-    font-size: 12px;
-    color: var(--text-muted);
-    font-style: italic;
-}
-
-/* Context Badge in Response */
-.context-tag {
-    display: inline-block;
-    font-family: var(--font-mono);
-    font-size: 10px;
-    color: var(--text-secondary);
-    background: var(--bg-subtle);
-    border: 1px solid var(--border-subtle);
-    border-radius: 4px;
-    padding: 3px 6px;
-    margin-bottom: 6px;
-}
-
-/* Prompt Suggestion Chips */
-.quick-prompt-row {
-    display: flex;
-    gap: 8px;
-    margin-bottom: 12px;
-    flex-wrap: wrap;
-}
-.quick-prompt-btn {
-    background: var(--bg-surface) !important;
-    border: 1px solid var(--border-subtle) !important;
-    color: var(--text-secondary) !important;
-    font-size: 11px !important;
-    border-radius: 20px !important;
-    padding: 4px 12px !important;
-    cursor: pointer !important;
-    font-weight: 400 !important;
-    transition: all 0.15s ease !important;
-}
-.quick-prompt-btn:hover {
-    color: var(--text-primary) !important;
-    border-color: var(--border-active) !important;
-    background: var(--bg-subtle) !important;
-}
-
-/* Sidebar Box Minimalist */
-.sidebar-panel {
-    background: var(--bg-surface) !important;
-    border: 1px solid var(--border-subtle) !important;
-    border-radius: 12px !important;
-    padding: 16px !important;
-}
-"""
-
-# ==============================================================================
-# 6. GRADIO 6.x COMPATIBLE BLOCKS & LAUNCH
-# ==============================================================================
 with gr.Blocks(title="GarudaCoder-27B") as demo:
-    # ------------------ TOP NAV BAR ------------------
-    gr.HTML("""
-        <div class="app-nav">
-            <div class="brand-group">
-                <div class="brand-badge">🦅 GC-27B</div>
-                <div>
-                    <div class="brand-name">GarudaCoder Workspace</div>
-                    <div class="brand-desc">Indonesian Specialized Coding Intelligence · Qwen3.8 Architecture</div>
-                </div>
-            </div>
-            <div class="status-pill">
-                <span class="status-dot"></span>
-                <span>ZeroGPU A100 · 4-Bit Native</span>
-            </div>
-        </div>
-    """)
-
     with gr.Row():
-        # ------------------ SETTINGS / SIDEBAR ------------------
-        with gr.Column(scale=1, min_width=270):
-            with gr.Group(elem_classes="sidebar-panel"):
-                gr.Markdown("#### Runtime Controls")
-                enable_thinking = gr.Checkbox(value=False, label="Deep Reasoning (CoT)")
-                enable_web_search = gr.Checkbox(value=False, label="Live Web Context (DDG)")
-
-                with gr.Accordion("Sampling & Temperature", open=False):
-                    temperature = gr.Slider(0.0, 1.5, value=0.6, step=0.05, label="Temperature")
-                    top_p = gr.Slider(0.1, 1.0, value=0.85, step=0.05, label="Top-P")
-                    max_new_tokens = gr.Slider(128, 4096, value=1024, step=64, label="Max Tokens")
-
-                with gr.Accordion("System Instructions", open=False):
-                    system_prompt = gr.Textbox(value=SYSTEM_PROMPT, label="Prompt", lines=6, show_label=False)
-
-            with gr.Group(elem_classes="sidebar-panel", visible=True):
-                gr.Markdown("#### Web References")
-                sources_box = gr.HTML('<div class="sources-empty">Pencarian web nonaktif.</div>')
-
-        # ------------------ CHAT WORKSPACE ------------------
+        with gr.Column(scale=1, min_width=260):
+            gr.Markdown("### ⚙️ Pengaturan")
+            enable_thinking = gr.Checkbox(value=False, label="🧠 Thinking Mode (CoT)")
+            enable_web_search = gr.Checkbox(value=False, label="🌐 Live Web Search")
+            
+            with gr.Accordion("Parameter Generasi", open=False):
+                temperature = gr.Slider(0.0, 1.5, value=0.6, step=0.05, label="Temperature")
+                top_p = gr.Slider(0.1, 1.0, value=0.85, step=0.05, label="Top-P")
+                max_new_tokens = gr.Slider(128, 4096, value=1024, step=64, label="Max Tokens")
+
+            with gr.Accordion("Instruksi Sistem", open=False):
+                system_prompt = gr.Textbox(value=SYSTEM_PROMPT, lines=6, label="System Prompt")
+
         with gr.Column(scale=3):
-            # Pada Gradio 6, type="messages" dihapus karena sudah jadi format standar bawaan
             chatbot = gr.Chatbot(
                 value=[{"role": "assistant", "content": WELCOME_MD}],
-                height=620,
+                height=600,
                 show_label=False,
-                elem_classes="chat-viewport"
             )
 
-            # Quick Prompt Suggestions (Interactive Pills)
-            with gr.Row(elem_classes="quick-prompt-row"):
-                p1 = gr.Button("Memory leak pada Python generator?", elem_classes="quick-prompt-btn")
-                p2 = gr.Button("Arsitektur Microservices vs Modular Monolith", elem_classes="quick-prompt-btn")
-                p3 = gr.Button("Optimasi SQL Query (N+1 Problem)", elem_classes="quick-prompt-btn")
+            with gr.Row():
+                p1 = gr.Button("Memory leak pada Python?", size="sm")
+                p2 = gr.Button("Microservices vs Modular Monolith", size="sm")
+                p3 = gr.Button("Optimasi N+1 Query SQL", size="sm")
 
             message = gr.MultimodalTextbox(
-                placeholder="Tanyakan analisis kode, upload dokumen teknis (PDF/Docx/Sheet), atau screenshot error...",
+                placeholder="Ketik pertanyaan atau lampirkan gambar, video, dan file dokumen...",
                 file_types=["image", "video", "file"],
                 file_count="multiple",
-                submit_btn="Send",
-                stop_btn="Stop",
+                show_label=False,
+                submit_btn="Kirim",
+                stop_btn="Batal",
             )
 
-    # ------------------ INTERAKSI & EVENTS ------------------
-    def set_prompt_p1():
-        return {"text": "Bagaimana cara mendeteksi dan mencegah memory leak pada penggunaan generator & closure di Python?", "files": []}
-    def set_prompt_p2():
-        return {"text": "Kapan sebaiknya beralih dari Modular Monolith ke Microservices? Berikan trade-off engineeringnya.", "files": []}
-    def set_prompt_p3():
-        return {"text": "Jelaskan apa itu masalah N+1 Query pada ORM dan 3 strategi konkret untuk mengatasinya.", "files": []}
-
-    p1.click(set_prompt_p1, outputs=[message])
-    p2.click(set_prompt_p2, outputs=[message])
-    p3.click(set_prompt_p3, outputs=[message])
+    p1.click(lambda: {"text": "Bagaimana cara mendeteksi dan mencegah memory leak pada Python?", "files": []}, outputs=[message])
+    p2.click(lambda: {"text": "Kapan sebaiknya beralih dari Modular Monolith ke Microservices?", "files": []}, outputs=[message])
+    p3.click(lambda: {"text": "Jelaskan apa itu N+1 Query Problem dan cara mengatasinya.", "files": []}, outputs=[message])
 
     def stream_handler(msg, hist, sp, temp, tp, mnt, eth, ews):
         user_text = (msg or {}).get("text", "")
         files = (msg or {}).get("files", [])
 
-        user_content_display = []
+        user_content = []
         for f in files:
             path = normalize_path(f)
             if path:
-                user_content_display.append({"path": path})
+                user_content.append({"path": path})
         if user_text:
-            user_content_display.append(user_text)
+            user_content.append(user_text)
 
         new_history = list(hist or [])
-        new_history.append({"role": "user", "content": user_content_display if user_content_display else user_text})
+        new_history.append({"role": "user", "content": user_content if user_content else user_text})
         new_history.append({"role": "assistant", "content": ""})
 
-        last_search = []
-        for partial_text, search_res in generate_chat(msg, hist, sp, temp, tp, mnt, eth, ews):
+        for partial_text in generate_chat(msg, hist, sp, temp, tp, mnt, eth, ews):
             new_history[-1]["content"] = partial_text
-            last_search = search_res
-            yield new_history, sources_html(last_search)
+            yield new_history
 
     message.submit(
         stream_handler,
         inputs=[message, chatbot, system_prompt, temperature, top_p, max_new_tokens, enable_thinking, enable_web_search],
-        outputs=[chatbot, sources_box]
+        outputs=[chatbot]
     )
 
-# launch() di Gradio 6 menerima parameter theme dan css langsung di sini
 if __name__ == "__main__":
     demo.queue(max_size=16, default_concurrency_limit=1).launch(
-        theme=gr.themes.Base(),
-        css=CSS
+        theme=gr.themes.Soft(primary_hue="emerald", neutral_hue="slate")
     )