Update app.py

app.py

@@ -7,34 +7,30 @@ from groq import Groq
 from mistralai import Mistral
 import google.generativeai as genai
 
-# --- CONFIGURAÇÕES GLOBAIS ---
+# --- CONFIGURAÇÕES ---
 
-# 1. LOCAL (H200 - ZeroGPU)
+# 1. LOCAL (H200)
 LOCAL_MODEL_ID = "Qwen/Qwen2.5-Coder-32B-Instruct"
 local_model = None
 local_tokenizer = None
 
-# 2. GROQ CLIENT
+# 2. CLIENTES DE NUVEM
 groq_key = os.environ.get("GROQ_API_KEY")
 groq_client = Groq(api_key=groq_key) if groq_key else None
 
-# 3. MISTRAL CLIENT
 mistral_key = os.environ.get("MISTRAL_API_KEY")
 mistral_client = Mistral(api_key=mistral_key) if mistral_key else None
 
-# 4. GEMINI CLIENT (NOVO!)
 gemini_key = os.environ.get("GEMINI_API_KEY")
 if gemini_key:
     genai.configure(api_key=gemini_key)
 
-# --- FUNÇÕES DE EXECUÇÃO ---
-
-# A. Função Local H200 (Cota Limitada)
+# --- FUNÇÃO 1: LOCAL H200 (ZeroGPU) ---
 @spaces.GPU(duration=60)
 def run_local_h200(messages):
     global local_model, local_tokenizer
     if local_model is None:
-        print(f"🐢 Cold Start: Carregando {LOCAL_MODEL_ID}...")
+        print(f"🐢 Carregando {LOCAL_MODEL_ID}...")
         local_tokenizer = AutoTokenizer.from_pretrained(LOCAL_MODEL_ID)
         local_model = AutoModelForCausalLM.from_pretrained(
             LOCAL_MODEL_ID, torch_dtype=torch.bfloat16, device_map="cuda"
@@ -45,90 +41,88 @@ def run_local_h200(messages):
     outputs = local_model.generate(**inputs, max_new_tokens=2048, temperature=0.6, do_sample=True)
     return local_tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
 
-# B. Função Groq
+# --- FUNÇÃO 2: GROQ ---
 def run_groq(messages, model_id):
-    if not groq_client: return "❌ Erro: Chave GROQ_API_KEY ausente."
+    if not groq_client: return "❌ Erro: Configure a GROQ_API_KEY."
     try:
         completion = groq_client.chat.completions.create(
             model=model_id, messages=messages, temperature=0.7, max_tokens=4096
         )
         return completion.choices[0].message.content
-    except Exception as e: return f"❌ Erro Groq: {e}"
+    except Exception as e: return f"❌ Groq Error: {e}"
 
-# C. Função Mistral
+# --- FUNÇÃO 3: MISTRAL (IDs Novos da sua lista!) ---
 def run_mistral(messages, model_id):
-    if not mistral_client: return "❌ Erro: Chave MISTRAL_API_KEY ausente."
+    if not mistral_client: return "❌ Erro: Configure a MISTRAL_API_KEY."
     try:
+        print(f"🇫🇷 Mistral Target: {model_id}")
         res = mistral_client.chat.complete(model=model_id, messages=messages)
         return res.choices[0].message.content
-    except Exception as e: return f"❌ Erro Mistral: {e}"
+    except Exception as e: return f"❌ Mistral Error: {e}"
 
-# D. Função Gemini (NOVO!)
+# --- FUNÇÃO 4: GEMINI (Modo "Hacker" - Aceita o que vc mandar) ---
 def run_gemini(messages, model_id):
-    if not gemini_key: return "❌ Erro: Chave GEMINI_API_KEY ausente."
+    if not gemini_key: return "❌ Erro: Configure a GEMINI_API_KEY."
     try:
         model = genai.GenerativeModel(model_id)
-        
-        # O Gemini funciona melhor com string direta ou histórico formatado
-        # Vamos converter o histórico do chat num prompt único para garantir contexto
-        full_prompt = ""
-        for m in messages:
-            role = "User" if m['role'] == 'user' else "Model"
-            full_prompt += f"{role}: {m['content']}\n"
-        
+        full_prompt = "\n".join([f"{m['role'].title()}: {m['content']}" for m in messages])
         response = model.generate_content(full_prompt)
         return response.text
-    except Exception as e: return f"❌ Erro Gemini: {e}"
+    except Exception as e: return f"❌ Gemini Error ({model_id}): {e}"
 
-# --- ROTEADOR CENTRAL ---
+# --- ROTEADOR ---
 def router(message, history, model_selector):
-    # Formata histórico
     messages = []
     for user_msg, bot_msg in history:
         if user_msg: messages.append({"role": "user", "content": user_msg})
         if bot_msg: messages.append({"role": "assistant", "content": bot_msg})
     messages.append({"role": "user", "content": message})
 
-    print(f"🔀 Roteando para: {model_selector}")
-
-    # --- SELEÇÃO ---
-    if "H200" in model_selector:
-        return run_local_h200(messages)
+    # --- MAPEAMENTO DE MODELOS ---
     
-    elif "Gemini 1.5 Pro" in model_selector:
-        return run_gemini(messages, "gemini-2.5-pro")
-    elif "Gemini 1.5 Flash" in model_selector:
-        return run_gemini(messages, "gemini-2.5-flash")
+    # Rota Google (Seus modelos "Future")
+    if "Gemini 3" in model_selector:
+        return run_gemini(messages, "gemini-3.0-pro-exp") # Tentativa de ID
+    elif "Gemini 2.5 Flash" in model_selector:
+        return run_gemini(messages, "gemini-2.0-flash-exp") # O Experimental atual
         
+    # Rota Groq
     elif "Llama 3.3" in model_selector:
         return run_groq(messages, "llama-3.3-70b-versatile")
-    elif "Llama 3.1 8B" in model_selector:
-        return run_groq(messages, "llama-3.1-8b-instant")
         
-    elif "Mistral Large" in model_selector:
-        return run_mistral(messages, "mistral-medium-2508")
-    elif "Codestral" in model_selector:
-        return run_mistral(messages, "codestral-latest")
+    # Rota Mistral (Sua Lista VIP)
+    elif "Mistral Large 2512" in model_selector:
+        return run_mistral(messages, "mistral-large-2512")
+    elif "Pixtral Large" in model_selector:
+        return run_mistral(messages, "pixtral-large-latest") # Multimodal!
+    elif "Magistral Medium" in model_selector:
+        return run_mistral(messages, "magistral-medium-latest") # Exclusivo
+    elif "Codestral 2508" in model_selector:
+        return run_mistral(messages, "codestral-2508")
         
+    # Rota Local
+    elif "H200" in model_selector:
+        return run_local_h200(messages)
     else:
-        return "⚠️ Modelo não reconhecido."
+        return "⚠️ Modelo não configurado no roteador."
 
-# --- INTERFACE ---
+# --- UI ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🔀 APIDOST Router V4: Quarteto Fantástico")
+    gr.Markdown("# 🔀 APIDOST Router V5: Unleashed")
     
     with gr.Row():
         model_dropdown = gr.Dropdown(
             choices=[
-                "✨ Google: Gemini 1.5 Pro (Raciocínio Avançado)",
-                "✨ Google: Gemini 1.5 Flash (Rápido)",
-                "☁️ Groq: Llama 3.3 70B (Versátil)", 
-                "☁️ Groq: Llama 3.1 8B (Instantâneo)",
-                "🇫🇷 Mistral: Large 3 (SOTA Europeu)",
-                "🇫🇷 Mistral: Codestral (Código)",
-                "🔥 Local H200: Qwen 2.5 Coder 32B (Cota ZeroGPU)"
+                "✨ Google: Gemini 3 (Experimental)",
+                "✨ Google: Gemini 2.5 Flash",
+                "☁️ Groq: Llama 3.3 70B",
+                "🇫🇷 Mistral: Large 2512 (Dez/25)",
+                "🇫🇷 Mistral: Pixtral Large (Vision)",
+                "🇫🇷 Mistral: Magistral Medium (VIP)",
+                "🇫🇷 Mistral: Codestral 2508 (Code)",
+                "🔥 Local H200: Qwen 2.5 Coder"
             ],
-            value="✨ Google: Gemini 1.5 Flash (Rápido)", 
+            value="🇫🇷 Mistral: Large 2512 (Dez/25)", 
             label="Escolha o Cérebro",
             interactive=True
         )