Update app.py

app.py

@@ -10,10 +10,14 @@ from sentence_transformers import SentenceTransformer
 
 app = FastAPI()
 
+import os
+os.environ["GPT4ALL_GPU"] = "0"   # 🚫 disables CUDA usage
+
+
 # =====================
 # MODEL SETUP
 # =====================
-model = GPT4All("gpt4all-falcon-newbpe-q4_0.gguf")
+model = GPT4All("gguf-gpt4all-j-v1.3-groovy")
 model = GPT4All(MODEL_NAME)  # Automatically downloads .gguf model (no Hugging Face)
 embedder = SentenceTransformer("all-MiniLM-L6-v2")