Update app.py

app.py

@@ -136,10 +136,9 @@ def call_model_with_messages(client: InferenceClient, messages: List[Dict[str, s
 
     Estratégia (ordem):
       1) client.completions.create(messages=...)
-      2) client.completions(messages=...) ou client.completions.create(prompt=...)
-      3) client.chat.create / client.chat_completion / client.chat_complete (variações)
-      4) client.text_generation(prompt=...)
-      5) tentar chamar diretamente funções encontradas que contenham 'create'/'generate'/'complet'
+      2) client.chat.create / client.chat(...) / client.chat_completion.create / client.chat_completion(...)
+      3) client.text_generation(prompt=...)
+      4) tentar chamar diretamente funções encontradas que contenham 'create'/'generate'/'complet'
 
     Retorna o objeto cru retornado pela biblioteca ou lança RuntimeError com info de debug.
     """
@@ -156,22 +155,43 @@ def call_model_with_messages(client: InferenceClient, messages: List[Dict[str, s
     except Exception as e:
         logger.debug("completions.create falhou: %s", e)
 
-    # 2) variações de chat_*
-    for attr in ("chat", "chat_completion", "chat_complete", "chat_completions"):
+    # 2) tentar chat / chat_completion namespaces (há no runtime mostrado)
+    # suporte: client.chat.create, client.chat(...), client.chat_completion.create, client.chat_completion(...)
+    for chat_ns in ("chat", "chat_completion", "chat_completions"):
         try:
-            chat_obj = getattr(client, attr, None)
-            if chat_obj is not None:
-                # se chat_obj tem .create
-                if hasattr(chat_obj, "create"):
-                    logger.info(f"Chamando {attr}.create(messages=...)")
-                    return chat_obj.create(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
-                if callable(chat_obj):
-                    logger.info(f"Chamando {attr}(messages=...)")
-                    return chat_obj(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
+            ns = getattr(client, chat_ns, None)
+            if ns is None:
+                continue
+            # ns pode ser um objeto com .create ou chamável diretamente
+            if hasattr(ns, "create"):
+                logger.info(f"Chamando {chat_ns}.create(messages=...)")
+                return ns.create(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
+            if callable(ns):
+                logger.info(f"Chamando {chat_ns}(messages=...)")
+                return ns(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
         except Exception as e:
-            logger.debug("%s falhou: %s", attr, e)
+            logger.debug("%s falhou: %s", chat_ns, e)
 
-    # 3) tentar gerar prompt concatenado e usar text_generation
+    # 3) tentar diretamente client.chat (que pelo debug pode existir como atributo com métodos internos)
+    try:
+        if hasattr(client, "chat"):
+            chat_obj = getattr(client, "chat")
+            # se chat_obj tem create
+            if hasattr(chat_obj, "create"):
+                logger.info("Chamando client.chat.create(messages=...)")
+                return chat_obj.create(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
+            # se chat_obj tem chat_completion
+            if hasattr(chat_obj, "chat_completion") and hasattr(chat_obj.chat_completion, "create"):
+                logger.info("Chamando client.chat.chat_completion.create(messages=...)")
+                return chat_obj.chat_completion.create(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
+            # se chat_obj é chamável
+            if callable(chat_obj):
+                logger.info("Chamando client.chat(messages=...)")
+                return chat_obj(messages=messages, max_new_tokens=max_new_tokens, temperature=temperature)
+    except Exception as e:
+        logger.debug("client.chat path falhou: %s", e)
+
+    # 4) gerar prompt concatenado e usar text_generation
     prompt = _messages_to_prompt(messages)
     try:
         if hasattr(client, "text_generation"):
@@ -184,7 +204,7 @@ def call_model_with_messages(client: InferenceClient, messages: List[Dict[str, s
     except Exception as e:
         logger.debug("text_generation/generate falhou: %s", e)
 
-    # 4) última tentativa: procurar métodos nomeados úteis
+    # 5) última tentativa: procurar métodos nomeados úteis
     candidate_methods = [m for m in dir(client) if any(k in m for k in ("create", "generate", "complete", "run"))]
     for name in candidate_methods:
         try: