Update bt_generator.py

bt_generator.py

@@ -4,13 +4,9 @@ from llama_cpp import Llama
 import textwrap
 import re
 import spaces
+import functools
 
 # Download only the behavior-tree model shard
-model_path = hf_hub_download(
-    repo_id="Inventors-Hub/SwarmChat-models",
-    repo_type="model",
-    filename="Falcon3-10B-Instruct-BehaviorTree-3epochs.Q4_K_M.gguf",
-)
 
 
 # llm = Llama(model_path=model_path, n_ctx=1024*4)
@@ -22,10 +18,17 @@ model_path = hf_hub_download(
 #     use_mmap=True,         # mmap file
 #     use_mlock=False,
 # )
-llm = None
+
 
 @spaces.GPU
+@functools.lru_cache(maxsize=1)
 def gpu_llm():
+    model_path = hf_hub_download(
+        repo_id="Inventors-Hub/SwarmChat-models",
+        repo_type="model",
+        filename="Falcon3-10B-Instruct-BehaviorTree-3epochs.Q4_K_M.gguf",
+    )
+
     llm = Llama(
         model_path=model_path,
         n_ctx=1024*4,            # down from 4096
@@ -111,6 +114,7 @@ def generate_behavior_tree(task_prompt: str) -> str:
     prompt = construct_prompt(task_prompt)
 
     print("\n\n",prompt,"\n\n")
+    llm = gpu_llm()
 
     output = llm(
         prompt,