Moved spaces GPU calls back to main functions as otherwise it doesn't seem to work correctly

app.py

@@ -171,12 +171,12 @@ with app:
 
         with gr.Accordion("Prompt settings", open = True):
             number_of_prompts = gr.Number(value=1, label="Number of prompts to send to LLM in sequence", minimum=1, maximum=3)
-            system_prompt_textbox = gr.Textbox(label="System prompt", lines = 4, value = system_prompt)
-            initial_table_prompt_textbox = gr.Textbox(label = "Prompt 1", lines = 8, value = initial_table_prompt)
+            system_prompt_textbox = gr.Textbox(label="Initial system prompt", lines = 4, value = system_prompt)
+            initial_table_prompt_textbox = gr.Textbox(label = "Initial topics prompt", lines = 8, value = initial_table_prompt)
             prompt_2_textbox = gr.Textbox(label = "Prompt 2", lines = 8, value = prompt2, visible=False)
             prompt_3_textbox = gr.Textbox(label = "Prompt 3", lines = 8, value = prompt3, visible=False)
-            add_to_existing_topics_system_prompt_textbox = gr.Textbox(label="Summary system prompt", lines = 4, value = add_existing_topics_system_prompt)
-            add_to_existing_topics_prompt_textbox = gr.Textbox(label = "Summary prompt", lines = 8, value = add_existing_topics_prompt)
+            add_to_existing_topics_system_prompt_textbox = gr.Textbox(label="Additional topics system prompt", lines = 4, value = add_existing_topics_system_prompt)
+            add_to_existing_topics_prompt_textbox = gr.Textbox(label = "Additional topics prompt", lines = 8, value = add_existing_topics_prompt)
             
         log_files_output = gr.File(label="Log file output", interactive=False)
         conversation_metadata_textbox = gr.Textbox(label="Query metadata - usage counts and other parameters", interactive=False, lines=8)

tools/chatfuncs.py

@@ -2,7 +2,6 @@ from typing import TypeVar
 import torch.cuda
 import os
 import time
-import spaces
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
 from tools.helper_functions import RUN_LOCAL_MODEL
@@ -126,7 +125,6 @@ def get_model_path():
         print(f"Checking default Hugging Face folder. Downloading model from Hugging Face Hub if not found")
         return hf_hub_download(repo_id=repo_id, filename=filename)
 
-@spaces.GPU
 def load_model(local_model_type:str=local_model_type, gpu_layers:int=gpu_layers, max_context_length:int=context_length, gpu_config:llama_cpp_init_config_gpu=gpu_config, cpu_config:llama_cpp_init_config_cpu=cpu_config, torch_device:str=torch_device):
     '''
     Load in a model from Hugging Face hub via the transformers package, or using llama_cpp_python by downloading a GGUF file from Huggingface Hub. 
@@ -173,7 +171,7 @@ def load_model(local_model_type:str=local_model_type, gpu_layers:int=gpu_layers,
     print(load_confirmation)
     return model, tokenizer
 
-@spaces.GPU
+
 def call_llama_cpp_model(formatted_string:str, gen_config:str, model=model):
     """
     Calls your generation model with parameters from the LlamaCPPGenerationConfig object.

tools/llm_api_call.py

@@ -10,6 +10,7 @@ import json
 import math
 import string
 import re
+import spaces
 from rapidfuzz import process, fuzz
 from tqdm import tqdm
 from gradio import Progress
@@ -878,7 +879,7 @@ def write_llm_output_and_logs(responses: List[ResponseObject],
 
     return topic_table_out_path, reference_table_out_path, unique_topics_df_out_path, topic_with_response_df, markdown_table, out_reference_df, out_unique_topics_df, batch_file_path_details, is_error
 
-
+@spaces.GPU
 def extract_topics(in_data_file,
               file_data:pd.DataFrame,
               existing_topics_table:pd.DataFrame,
@@ -1614,7 +1615,7 @@ def summarise_output_topics_query(model_choice:str, in_api_key:str, temperature:
 
     return latest_response_text, conversation_history, whole_conversation_metadata
 
-
+@spaces.GPU
 def summarise_output_topics(summarised_references:pd.DataFrame,
                             unique_table_df:pd.DataFrame,
                             reference_table_df:pd.DataFrame,