Update modules/inference.py

modules/inference.py

@@ -1,3 +1,5 @@
+import spaces
+
 import os
 import gradio as gr
 import copy
@@ -10,6 +12,7 @@ import time
 from modules.load_presets import load_presets_value
 from modules.load_model import *
 
+@spaces.GPU
 def generate_text(message, history, system_prompt, preset, temperature, max_tokens, top_p, top_k, repeat_penalty, model, n_ctx, n_gpu_layers, n_threads, verbose, f16_kv, logits_all, vocab_only, use_mmap, use_mlock, n_batch, last_n_tokens_size, low_vram, rope_freq_base, rope_freq_scale):
     dir = os.getcwd()
     global llm