Update app.py

app.py

@@ -225,9 +225,16 @@ def generate_reply(
     max_new_tokens: int,
     temperature:    float,
     top_p:          float,
+    model_choice:   str = "",
 ) -> Generator[str, None, None]:
     """Main generation function."""
 
+    # Model switching (may take 1-2 min on first switch)
+    target = model_choice if model_choice in MODELS else DEFAULT_MODEL
+    if target != _loaded_model_name:
+        yield f"⏳ Loading **{target}**... (최초 전환 시 1-2분 소요)"
+        _load_model(target)
+
     use_think = "Thinking" in thinking_mode
     max_new_tokens = min(int(max_new_tokens), 8192)
 
@@ -308,27 +315,33 @@ footer { display: none !important; }
 .model-box .st { font-size: 10px; color: #78716c; margin-top: 4px; }
 """
 
-_mcfg = MODELS[DEFAULT_MODEL]
-MODEL_INFO_HTML = (
-    f'<div class="model-box">'
-    f'<b>{"⚡" if _mcfg["arch"]=="MoE" else "🏆"} {DEFAULT_MODEL}</b> '
-    f'<span style="font-size:9px;padding:2px 6px;border-radius:6px;background:rgba(109,40,217,.08);color:#6d28d9;font-weight:700">{_mcfg["arch"]}</span><br>'
-    f'<div class="st">{_mcfg["active"]} active / {_mcfg["total"]} total · 👁️ Vision · {_mcfg["ctx"]} context</div>'
-    f'<div class="st">{_mcfg["desc"]}</div>'
-    f'<div class="st" style="margin-top:6px">'
-    f'<a href="https://huggingface.co/{MODELS[DEFAULT_MODEL]["id"]}" target="_blank" style="color:#6d28d9;font-weight:700;text-decoration:none">🤗 Model Card ↗</a> · '
-    f'<a href="https://deepmind.google/models/gemma/gemma-4/" target="_blank" style="color:#059669;font-weight:700;text-decoration:none">🔬 DeepMind ↗</a>'
-    f'</div></div>'
-)
+def _model_info_html(name):
+    m = MODELS.get(name, MODELS[DEFAULT_MODEL])
+    icon = "⚡" if m["arch"] == "MoE" else "🏆"
+    return (
+        f'<div class="model-box">'
+        f'<b>{icon} {name}</b> '
+        f'<span style="font-size:9px;padding:2px 6px;border-radius:6px;background:rgba(109,40,217,.08);color:#6d28d9;font-weight:700">{m["arch"]}</span><br>'
+        f'<div class="st">{m["active"]} active / {m["total"]} total · 👁️ Vision · {m["ctx"]} context</div>'
+        f'<div class="st">{m["desc"]}</div>'
+        f'<div class="st" style="margin-top:6px">'
+        f'<a href="https://huggingface.co/{m["id"]}" target="_blank" style="color:#6d28d9;font-weight:700;text-decoration:none">🤗 Model Card ↗</a> · '
+        f'<a href="https://deepmind.google/models/gemma/gemma-4/" target="_blank" style="color:#059669;font-weight:700;text-decoration:none">🔬 DeepMind ↗</a>'
+        f'</div></div>'
+    )
 
 with gr.Blocks(title="Gemma 4 Playground") as demo:
 
     gr.Markdown("## 💎 Gemma 4 Playground\nGoogle DeepMind · Apache 2.0 · Vision · Thinking")
 
     with gr.Row():
-        # ── Sidebar (compact) ──
+        # ── Sidebar ──
         with gr.Column(scale=0, min_width=280):
-            gr.HTML(MODEL_INFO_HTML)
+            model_dd = gr.Dropdown(
+                choices=list(MODELS.keys()), value=DEFAULT_MODEL, label="Model",
+                info="⚡MoE=Fast | 🏆Dense=Best quality (전환 시 1-2분)",
+            )
+            model_info = gr.HTML(value=_model_info_html(DEFAULT_MODEL))
             image_input = gr.Image(label="👁️ Image (Vision)", type="filepath", height=140)
             thinking_radio = gr.Radio(["⚡ Fast", "🧠 Thinking"], value="⚡ Fast", label="Mode")
             with gr.Accordion("⚙️ Settings", open=False):
@@ -350,21 +363,22 @@ with gr.Blocks(title="Gemma 4 Playground") as demo:
                 send_btn = gr.Button("↑", variant="primary", scale=0, min_width=48, elem_id="send-btn")
 
     # ── Events ──
+    model_dd.change(fn=_model_info_html, inputs=[model_dd], outputs=[model_info])
     preset_dd.change(fn=lambda k: PRESETS.get(k, PRESETS["general"]), inputs=[preset_dd], outputs=[sys_prompt])
 
     def user_msg(msg, hist):
         if not msg.strip(): return "", hist
         return "", hist + [{"role": "user", "content": msg}]
 
-    def bot_reply(hist, think, img, sysp, maxt, tmp, tp):
+    def bot_reply(hist, think, img, sysp, maxt, tmp, tp, model):
         if not hist or hist[-1]["role"] != "user": return hist
         txt, past = hist[-1]["content"], hist[:-1]
         hist = hist + [{"role": "assistant", "content": ""}]
-        for chunk in generate_reply(txt, past, think, img, sysp, maxt, tmp, tp):
+        for chunk in generate_reply(txt, past, think, img, sysp, maxt, tmp, tp, model):
             hist[-1]["content"] = chunk
             yield hist
 
-    ins = [chatbot, thinking_radio, image_input, sys_prompt, max_tok, temp, topp]
+    ins = [chatbot, thinking_radio, image_input, sys_prompt, max_tok, temp, topp, model_dd]
     send_btn.click(user_msg, [chat_input, chatbot], [chat_input, chatbot], queue=False).then(bot_reply, ins, chatbot)
     chat_input.submit(user_msg, [chat_input, chatbot], [chat_input, chatbot], queue=False).then(bot_reply, ins, chatbot)
     clear_btn.click(lambda: [], None, chatbot, queue=False)