Upload 10 files

Dockerfile

@@ -6,7 +6,7 @@ WORKDIR /app
 RUN apt-get update && apt-get install -y curl
 
 # # Download the model
-RUN curl -Lo qwen2.5-coder-3b-instruct-q4_k_m.gguf https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF/resolve/main/qwen2.5-coder-0.5b-instruct-q4_k_m.gguf?download=true
+RUN curl -Lo qwen2.5-coder-3b-instruct-q4_k_m.gguf https://huggingface.co/DebopamC/Text-to-SQL__Qwen2.5-Coder-3B-FineTuned/resolve/main/Text-to-SQL-Qwen2.5-Coder-3B-FineTuned.gguf?download=true
 
 # Install build tools required for llama-cpp-python
 RUN apt-get update && apt-get install -y build-essential

utils/llm_logic.py

@@ -4,6 +4,7 @@ from langchain_core.messages import HumanMessage, SystemMessage, AIMessage
 import streamlit as st
 import multiprocessing
 from langchain_community.chat_models import ChatLlamaCpp
+from langchain_google_genai import ChatGoogleGenerativeAI
 
 local_model = "qwen2.5-coder-3b-instruct-q4_k_m.gguf"
 
@@ -29,7 +30,7 @@ stop = [
 ]
 
 
-def get_llm():
+def get_local_llm():
     cache_llm = ChatLlamaCpp(
         temperature=0.0,
         model_path=local_model,
@@ -45,7 +46,20 @@ def get_llm():
     return cache_llm
 
 
-llm = get_llm()
+local_llm = get_local_llm()
+
+
+def get_gemini_llm():
+    gemini = ChatGoogleGenerativeAI(
+        model="gemini-1.5-flash",
+        temperature=0,
+        max_tokens=None,
+        timeout=None,
+        max_retries=2,
+    )
+    return gemini
+
+gemini_llm = get_gemini_llm()
 
 
 db_schema = """### **customers**
@@ -151,7 +165,7 @@ QUESTION: {question}
 """
 
 
-def classify_question(question: str, use_default_schema: bool = True):
+def classify_question(question: str, llm , use_default_schema: bool = True):
     classification_system_prompt_local = classification_system_prompt  # Initialize here
     if use_default_schema:
         classification_system_prompt_local = classification_system_prompt_local.format(
@@ -170,8 +184,14 @@ def classify_question(question: str, use_default_schema: bool = True):
     return response.content.strip().upper()
 
 
-def generate_llm_response(prompt: str, use_default_schema: bool = True):
-    question_type = classify_question(prompt, use_default_schema)
+def generate_llm_response(prompt: str, llm: str,  use_default_schema: bool = True):
+
+    if llm == "gemini":
+        llm = gemini_llm
+    else:
+        llm = local_llm
+
+    question_type = classify_question(prompt, llm, use_default_schema)
     chosen_schema = None
     if use_default_schema:
         chosen_schema = db_schema

🤖SQL_Agent.py

@@ -18,29 +18,12 @@ st.set_page_config(
     initial_sidebar_state="expanded",
 )
 
-default_db_questions = {
-    "easy": [
-        "Retrieve all customer IDs and their corresponding cities from the `customers` table.",
-        "List all products along with their category names from the `products` table.",
-        "Fetch the order IDs and their purchase timestamps from the `orders` table.",
-        "Display the distinct payment types available in the `payments` table.",
-        "Find the total number of rows in the `customers` table.",
-    ],
-    "medium": [
-        "Retrieve the total payment value for each order from the `payments` table, grouped by `order_id`.",
-        "Find all orders where the total shipping charges (sum of `shipping_charges`) exceed 100.",
-        "List the names of cities and the number of customers in each city, sorted in descending order of the number of customers.",
-    ],
-    "hard": [
-        "Write a query to find the total revenue (sum of `price` + `shipping_charges`) generated for each product category in the `order_items` table, joined with the `products` table.",
-        "Identify the top 5 products with the highest total sales value (sum of `price`) across all orders.",
-    ],
-}
-
-
+default_db_questions = {}
 default_dfs = load_data()
 selected_df = default_dfs
 use_default_schema = True
+llm_option = "gemini"
+
 
 st.markdown(
     """
@@ -139,14 +122,33 @@ st.caption(
 )
 
 
-col1, col2 = st.columns([2, 1], vertical_alignment="bottom")
+col1, col2, col3 = st.columns([1, 1, 1], vertical_alignment="top")
 with col1:
     # Button to refresh the conversation
     if st.button("Start New Conversation", type="primary"):
         st.session_state.chat_history = []
         st.session_state.conversation_turns = 0
         st.rerun()
+    if selected_df == default_dfs:
+        with st.popover("Default Database Queries 📚 - Trial"):
+            default_db_questions = load_defaultdb_queries()
+            st.markdown(default_db_questions)
 with col2:
+    llm_option_radio = st.radio(
+        "Choose LLM Model",
+        ["Gemini 1.5-Flash", "FineTuned Qwen2.5-Coder-3B for SQL"],
+        captions=[
+            "Used via API",
+            "Run Locally on this Server. Extremely Slow because of Free vCPUs",
+        ],
+        label_visibility="collapsed",
+    )
+    if llm_option_radio == "Gemini 1.5-Flash":
+        llm_option = "gemini"
+    else:
+        llm_option = "qwen"
+
+with col3:
     disabled_selection = True
     if (
         "uploaded_dataframes" in st.session_state
@@ -170,10 +172,6 @@ with col2:
             selected_df = default_dfs
             # print(selected_df)
             use_default_schema = True
-if selected_df == default_dfs:
-    with st.popover("Default Database Queries 📚 - Trial"):
-        default_db_questions = load_defaultdb_queries()
-        st.markdown(default_db_questions)
 
 # Initialize chat history in session state
 if "chat_history" not in st.session_state:
@@ -234,11 +232,18 @@ if st.session_state.conversation_turns < MAX_TURNS:
         with st.chat_message("assistant"):
             message_placeholder = st.empty()
             full_response = ""
+            spinner_text = ""
+            if llm_option == "gemini":
+                spinner_text = (
+                    "Using Gemini-1.5-Flash to run your query. Please wait...😊"
+                )
+            else:
+                spinner_text = "I know it is taking a lot of time. To run the model I'm using `Free` small vCPUs provided by `HuggingFace Spaces` for deployment. Thank you so much for your patience😊"
             with st.spinner(
-                "I know it is taking a lot of time. To run the model I'm using `Free` small vCPUs provided by `HuggingFace Spaces` for deployment. Thank you so much for your patience😊"
+                spinner_text,
             ):
                 for response_so_far in generate_llm_response(
-                    prompt, use_default_schema
+                    prompt, llm_option, use_default_schema
                 ):
                     # Remove <sql> and </sql> tags for streaming display
                     streaming_response = response_so_far.replace("<sql>", "").replace(