Atualizar app/app.py

app/app.py

@@ -87,6 +87,17 @@ def rate_limited(f):
         return f(*args, **kwargs)
     return wrapper
 
+def call_ollama_with_fallback(payload):
+    for model in FALLBACK_MODELS:
+        payload["model"] = model
+        try:
+            r = requests.post(f"{app.config['OLLAMA_BASE_URL']}/api/generate", json=payload)
+            r.raise_for_status()
+            return r.json(), model
+        except Exception:
+            continue
+    raise Exception("All models failed")
+
 
 def admin_required(f):
     @wraps(f)
@@ -989,23 +1000,23 @@ def generate():
     if not model_id:
         return jsonify({"error": f"Unsupported model '{model_name}'."}), 400
 
-    ollama_url = f"{app.config['OLLAMA_BASE_URL']}/api/generate"
-    payload = {"model": model_id, "prompt": prompt}
+    # ollama_url = f"{app.config['OLLAMA_BASE_URL']}/api/generate"
+    # payload = {"model": model_id, "prompt": prompt}
+    # try:
+    #     response = requests.post(ollama_url, json=payload, timeout=60)
+    #     response.raise_for_status()
+    # except requests.RequestException as e:
+    #     return jsonify({"error": "Failed to call Ollama API.", "details": str(e)}), 502
+    # try:
+    #     ollama_data = response.json()
+    # except json.JSONDecodeError:
+    #     return jsonify({"error": "Invalid JSON response from Ollama."}), 502
+    #output_text = ollama_data.get("response") or ollama_data.get("output") or ""
+    #return jsonify({"model": model_name, "output": output_text})
 
-    try:
-        response = requests.post(ollama_url, json=payload, timeout=60)
-        response.raise_for_status()
-    except requests.RequestException as e:
-        return jsonify({"error": "Failed to call Ollama API.", "details": str(e)}), 502
+    result, used_model = call_ollama_with_fallback({"prompt": prompt})
+    return jsonify({"model_used": used_model, "output": result.get("response")})
 
-    try:
-        ollama_data = response.json()
-    except json.JSONDecodeError:
-        return jsonify({"error": "Invalid JSON response from Ollama."}), 502
-
-    output_text = ollama_data.get("response") or ollama_data.get("output") or ""
-
-    return jsonify({"model": model_name, "output": output_text})
 
 
 if __name__ == "__main__":