Update app.py

app.py

@@ -2,7 +2,7 @@ from transformers import AutoTokenizer
 from exllamav2 import (
     ExLlamaV2,
     ExLlamaV2Config,
-    ExLlamaV2Cache,
+    ExLlamaV2Cache_CPU,
     ExLlamaV2Tokenizer
 )
 from exllamav2.generator import (
@@ -10,6 +10,11 @@ from exllamav2.generator import (
     ExLlamaV2Sampler
 )
 import torch
+import os
+
+# disable CUDA
+os.environ["CUDA_VISIBLE_DEVICES"] = "-1"  # Disables GPU detection
+os.environ["EXLLAMA_NO_CUDA"] = "1"  # Forces CPU mode in ExLlamaV2
 
 # Configure model
 model_dir = "TheBloke_Wizard-Vicuna-13B-GPTQ"  # Path to downloaded model
@@ -21,7 +26,7 @@ config.prepare()
 
 # Load model
 model = ExLlamaV2(config)
-cache = ExLlamaV2Cache(model)
+cache = ExLlamaV2Cache_CPU(model)
 model.load_autosplit(cache)
 
 # Load tokenizer (HF-compatible)