Update app.py

app.py

@@ -24,11 +24,14 @@ except NVMLError as error:
 
 
 os.environ["RWKV_JIT_ON"] = '1'
-os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
 
 # Model strat to use
 MODEL_STRAT="cpu bf16"
+os.environ["RWKV_CUDA_ON"] = '0' # if '1' then use CUDA kernel for seq mode (much faster)
+
+# Switch to GPU mode
 if HAS_GPU == True :
+    os.environ["RWKV_CUDA_ON"] = '1' 
     MODEL_STRAT = "cuda bf16"
 
 # Load the model accordingly