Capability to modify the llm through UI

app.py

@@ -4,7 +4,8 @@ import threading
 import time
 from generator.compute_metrics import get_attributes_text
 from generator.generate_metrics import generate_metrics, retrieve_and_generate_response
-from config import AppConfig, ConfigConstants 
+from config import AppConfig, ConfigConstants
+from generator.initialize_llm import initialize_generation_llm, initialize_validation_llm 
 
 def launch_gradio(config : AppConfig):
     """
@@ -80,17 +81,50 @@ def launch_gradio(config : AppConfig):
             logging.error(f"Error computing metrics: {e}")
             return f"An error occurred: {e}", ""
 
+    def reinitialize_gen_llm(gen_llm_name):
+        """Reinitialize the generation LLM and return updated model info."""
+        if gen_llm_name.strip():  # Only update if input is not empty
+            config.gen_llm = initialize_generation_llm(gen_llm_name)
+        
+        # Return updated model information
+        updated_model_info = (
+            f"Embedding Model: {ConfigConstants.EMBEDDING_MODEL_NAME}\n"
+            f"Generation LLM: {config.gen_llm.name if hasattr(config.gen_llm, 'name') else 'Unknown'}\n"
+            f"Validation LLM: {config.val_llm.name if hasattr(config.val_llm, 'name') else 'Unknown'}\n"
+        )
+        return updated_model_info
+
+    def reinitialize_val_llm(val_llm_name):
+        """Reinitialize the generation LLM and return updated model info."""
+        if val_llm_name.strip():  # Only update if input is not empty
+            config.val_llm = initialize_validation_llm(val_llm_name)
+        
+        # Return updated model information
+        updated_model_info = (
+            f"Embedding Model: {ConfigConstants.EMBEDDING_MODEL_NAME}\n"
+            f"Generation LLM: {config.gen_llm.name if hasattr(config.gen_llm, 'name') else 'Unknown'}\n"
+            f"Validation LLM: {config.val_llm.name if hasattr(config.val_llm, 'name') else 'Unknown'}\n"
+        )
+        return updated_model_info
+    
     # Define Gradio Blocks layout
     with gr.Blocks() as interface:
         interface.title = "Real Time RAG Pipeline Q&A"
         gr.Markdown("### Real Time RAG Pipeline Q&A")  # Heading
         
+        # Textbox for new generation LLM name
+        with gr.Row():
+            new_gen_llm_input = gr.Textbox(label="New Generation LLM Name", placeholder="Enter LLM name to update")
+            update_gen_llm_button = gr.Button("Update Generation LLM")
+            new_val_llm_input = gr.Textbox(label="New Validation LLM Name", placeholder="Enter LLM name to update")
+            update_val_llm_button = gr.Button("Update Validation LLM")
+
         # Section to display LLM names
         with gr.Row():
             model_info = f"Embedding Model: {ConfigConstants.EMBEDDING_MODEL_NAME}\n"
             model_info += f"Generation LLM: {config.gen_llm.name if hasattr(config.gen_llm, 'name') else 'Unknown'}\n"
             model_info += f"Validation LLM: {config.val_llm.name if hasattr(config.val_llm, 'name') else 'Unknown'}\n"
-            gr.Textbox(value=model_info, label="Model Information", interactive=False)  # Read-only textbox
+            model_info_display = gr.Textbox(value=model_info, label="Model Information", interactive=False)  # Read-only textbox
 
         # State to store response and source documents
         state = gr.State(value={"query": "","response": "", "source_docs": {}})
@@ -122,7 +156,19 @@ def launch_gradio(config : AppConfig):
             inputs=[state],
             outputs=[attr_output, metrics_output]
         )
+        
+        update_gen_llm_button.click(
+            fn=reinitialize_gen_llm,
+            inputs=[new_gen_llm_input],
+            outputs=[model_info_display]  # Update the displayed model info
+        )
 
+        update_val_llm_button.click(
+            fn=reinitialize_val_llm,
+            inputs=[new_val_llm_input],
+            outputs=[model_info_display]  # Update the displayed model info
+        )
+        
         # Section to display logs
         with gr.Row():
             start_log_button = gr.Button("Start Log Update", elem_id="start_btn")  # Button to start log updates

config.py

@@ -1,7 +1,7 @@
 
 class ConfigConstants:
     # Constants related to datasets and models
-    DATA_SET_NAMES = ['covidqa', 'cuad', 'delucionqa', 'emanual', 'expertqa', 'finqa', 'hagrid', 'hotpotqa', 'msmarco', 'pubmedqa', 'tatqa', 'techqa']
+    DATA_SET_NAMES = ['covidqa', 'cuad']#, 'delucionqa', 'emanual', 'expertqa', 'finqa', 'hagrid', 'hotpotqa', 'msmarco', 'pubmedqa', 'tatqa', 'techqa']
     EMBEDDING_MODEL_NAME = "sentence-transformers/paraphrase-MiniLM-L3-v2"
     RE_RANKER_MODEL_NAME = 'cross-encoder/ms-marco-electra-base'
     GENERATION_MODEL_NAME = 'mixtral-8x7b-32768'

generator/initialize_llm.py

@@ -2,18 +2,22 @@ import logging
 import os
 from langchain_groq import ChatGroq
 
-from config import ConfigConstants
-
-def initialize_generation_llm():
+def initialize_generation_llm(input_model_name):
     os.environ["GROQ_API_KEY"] = ""
-    model_name = ConfigConstants.GENERATION_MODEL_NAME
+    
+    model_name = input_model_name    
     llm = ChatGroq(model=model_name, temperature=0.7)
+    llm.name = model_name
     logging.info(f'Generation LLM {model_name} initialized')
+    
     return llm
 
-def initialize_validation_llm():
+def initialize_validation_llm(input_model_name):
     os.environ["GROQ_API_KEY"] = ""
-    model_name = ConfigConstants.VALIDATION_MODEL_NAME
+    
+    model_name = input_model_name      
     llm = ChatGroq(model=model_name, temperature=0.7)
+    llm.name = model_name
     logging.info(f'Validation LLM {model_name} initialized')
+    
     return llm

main.py

@@ -44,10 +44,10 @@ def main():
     logging.info("Documents embedded")
     
      # Initialize the Generation LLM
-    gen_llm = initialize_generation_llm()
+    gen_llm = initialize_generation_llm(ConfigConstants.GENERATION_MODEL_NAME)
 
     # Initialize the Validation LLM
-    val_llm = initialize_validation_llm()
+    val_llm = initialize_validation_llm(ConfigConstants.VALIDATION_MODEL_NAME)
 
     #Compute RMSE and AUC-ROC for entire dataset
     #Enable below code for calculation
@@ -55,7 +55,7 @@ def main():
     #compute_rmse_auc_roc_metrics(gen_llm, val_llm, datasets[data_set_name], vector_store, 10)
     
     # Launch the Gradio app
-    config = AppConfig(vector_store= vector_store, gen_llm= gen_llm, val_llm= val_llm)
+    config = AppConfig(vector_store= vector_store, gen_llm = gen_llm, val_llm = val_llm)
     launch_gradio(config)
 
     logging.info("Finished!!!")

retriever/retrieve_documents.py

@@ -8,7 +8,7 @@ def retrieve_top_k_documents(vector_store, query, top_k=5):
     documents = vector_store.similarity_search(query, k=top_k)
     logging.info(f"Top {top_k} documents reterived for query")
 
-    documents = rerank_documents(query, documents)
+    #documents = rerank_documents(query, documents)
     
     return documents