updates on app for evaluations

app.py

@@ -1,152 +1,113 @@
 import os
 import gradio as gr
 from huggingface_hub import InferenceClient
-from sqlalchemy import create_engine
-
-# Clients for each model provider
-llama_client = InferenceClient(provider="sambanova", api_key=os.environ["HF_TOKEN"])
-minimax_client = InferenceClient(provider="novita", api_key=os.environ["HF_TOKEN"])
-mistral_client = InferenceClient(provider="together", api_key=os.environ["HF_TOKEN"])
-
-# Global objects
-db_connection = None  
-
-def get_sqlalchemy_connection():
-    server = os.getenv("SQL_SERVER")
-    database = os.getenv("SQL_DATABASE")
-    username = os.getenv("SQL_USERNAME")
-    password = os.getenv("SQL_PASSWORD")
-
-    connection_url = f"mssql+pymssql://{username}:{password}@{server}/{database}"
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+
+HF_TOKEN = os.environ.get("HF_TOKEN")
+
+# ------------------------
+# API Clients
+# ------------------------
+llama_client = InferenceClient(
+    provider="fireworks-ai",
+    api_key=HF_TOKEN,
+)
+
+qwen_client = InferenceClient(
+    provider="featherless-ai",
+    api_key=HF_TOKEN,
+)
+
+# ------------------------
+# Mixtral Local Setup
+# ------------------------
+mixtral_model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+mixtral_tokenizer = AutoTokenizer.from_pretrained(mixtral_model_id)
+mixtral_model = AutoModelForCausalLM.from_pretrained(
+    mixtral_model_id, torch_dtype=torch.float16
+).to("cuda")
+
+# ------------------------
+# Unified Inference Function with Chat History
+# ------------------------
+def run_model_with_history(model_name, user_input, chat_history):
+    messages = chat_history + [{"role": "user", "content": user_input}]
 
     try:
-        engine = create_engine(connection_url)
-        conn = engine.connect()
-        print("✅ SQLAlchemy + pymssql connection successful")
-        return conn
-    except Exception as e:
-        print(f"❌ SQLAlchemy connection failed: {e}")
-        return None
-
-def get_sql_connection():
-    global db_connection
-
-    if db_connection is not None:
-        try:
-            db_connection.cursor()  # test if still open
-            return db_connection
-        except Exception as e:
-            print(f"❌ SQL connection failed: {e}")
-            db_connection = None  # reset if broken
-
-    # Reconnect if needed
-    db_connection = get_sqlalchemy_connection()
-    return db_connection
-    
-# Format chat history for Markdown display
-def format_chat_history(chat_history):
-    formatted = ""
-    for msg in chat_history:
-        role = msg["role"]
-        content = msg["content"]
-        if isinstance(content, list):  # For LLaMA image+text input
-            for item in content:
-                if "text" in item:
-                    formatted += f"**{role.capitalize()}:** {item['text']}\n\n"
-                elif "image_url" in item:
-                    formatted += f"**{role.capitalize()}:** 🖼️ Image: {item['image_url']['url']}\n\n"
-        else:
-            formatted += f"**{role.capitalize()}:** {content}\n\n"
-    return formatted.strip()
-
-# Main chat handler
-def chat_with_model(model_choice, prompt, image_url, chat_history):
-    if not prompt:
-        return "❌ Please enter a text prompt.", chat_history, "", ""
-
-    if chat_history is None:
-        chat_history = []
-        
-    conn = get_sql_connection()
-    if conn is None:
-        return "❌ Failed to connect to database.", chat_history, "", ""
-
-    try:
-        # === LLaMA 4 ===
-        if model_choice == "LLaMA 4 (SambaNova)":
-            user_msg = [{"type": "text", "text": prompt}]
-            if image_url:
-                user_msg.append({"type": "image_url", "image_url": {"url": image_url}})
-            chat_history.append({"role": "user", "content": user_msg})
-
-            response = llama_client.chat.completions.create(
+        if model_name == "LLaMA 4":
+            result = llama_client.chat.completions.create(
                 model="meta-llama/Llama-4-Maverick-17B-128E-Instruct",
-                messages=chat_history
+                messages=messages
             )
-            bot_msg = response.choices[0].message.content
-            chat_history.append({"role": "assistant", "content": bot_msg})
-
-        # === MiniMax ===
-        elif model_choice == "MiniMax M1 (Novita)":
-            chat_history.append({"role": "user", "content": prompt})
-            response = minimax_client.chat.completions.create(
-                model="MiniMaxAI/MiniMax-M1-80k",
-                messages=chat_history
+            reply = result.choices[0].message.content
+
+        elif model_name == "Qwen3 14B":
+            result = qwen_client.chat.completions.create(
+                model="Qwen/Qwen3-14B",
+                messages=messages
             )
-            bot_msg = response.choices[0].message.content
-            chat_history.append({"role": "assistant", "content": bot_msg})
-
-        # === Mistral ===
-        elif model_choice == "Mistral Mixtral-8x7B (Together)":
-            chat_history.append({"role": "user", "content": prompt})
-            response = mistral_client.chat.completions.create(
-                model="mistralai/Mixtral-8x7B-Instruct-v0.1",
-                messages=chat_history
+            reply = result.choices[0].message.content
+
+        elif model_name == "Mixtral 8x7B":
+            full_prompt = ""
+            for msg in messages:
+                prefix = "User: " if msg["role"] == "user" else "Assistant: "
+                full_prompt += f"{prefix}{msg['content']}\n"
+            inputs = mixtral_tokenizer(full_prompt, return_tensors="pt").to("cuda")
+            outputs = mixtral_model.generate(
+                **inputs,
+                max_new_tokens=512,
+                do_sample=True,
+                temperature=0.7,
+                top_k=50,
+                top_p=0.95
             )
-            bot_msg = response.choices[0].message.content
-            chat_history.append({"role": "assistant", "content": bot_msg})
+            reply = mixtral_tokenizer.decode(outputs[0], skip_special_tokens=True)
 
         else:
-            return "❌ Unsupported model selected.", chat_history, "", ""
-
-        return format_chat_history(chat_history), chat_history, "", ""
+            reply = "❌ Invalid model selection."
 
     except Exception as e:
-        return f"❌ Error: {e}", chat_history, "", ""
+        reply = f"⚠️ Error: {str(e)}"
+
+    # Update chat history
+    chat_history.append({"role": "user", "content": user_input})
+    chat_history.append({"role": "assistant", "content": reply})
 
-# Gradio interface
+    # Format display
+    chat_transcript = "\n".join([
+        f"👤 User: {msg['content']}" if msg["role"] == "user" else f"🤖 Assistant: {msg['content']}"
+        for msg in chat_history
+    ])
+
+    return chat_transcript, chat_history
+
+# ------------------------
+# Gradio UI
+# ------------------------
 with gr.Blocks() as demo:
-    gr.Markdown("## 🤖 Multi-Model Context-Aware Chatbot")
-    gr.Markdown("Supports LLaMA 4 (with optional image), MiniMax, and Mistral. Memory is preserved for multi-turn dialog.")
-
-    model_dropdown = gr.Dropdown(
-        choices=[
-            "LLaMA 4 (SambaNova)",
-            "MiniMax M1 (Novita)",
-            "Mistral Mixtral-8x7B (Together)"
-        ],
-        value="LLaMA 4 (SambaNova)",
-        label="Select Model"
+    gr.Markdown("## 🧠 Generative AI Model Evaluation with Context")
+
+    model_choice = gr.Dropdown(
+        choices=["LLaMA 4", "Qwen3 14B", "Mixtral 8x7B"],
+        label="Select Model",
+        value="LLaMA 4"
     )
 
-    prompt_input = gr.Textbox(label="Text Prompt", placeholder="Ask something...", lines=2)
-    image_url_input = gr.Textbox(label="Optional Image URL (for LLaMA only)", placeholder="https://example.com/image.jpg")
+    chat_display = gr.Textbox(label="Chat History", lines=20, interactive=False)
+    prompt_input = gr.Textbox(label="Your Prompt", lines=3, placeholder="Ask your BI question...")
 
-    submit_btn = gr.Button("💬 Generate Response")
-    reset_btn = gr.Button("🔄 Reset Conversation")
-    output_box = gr.Markdown(label="Chat History", value="")
-    state = gr.State([])
+    run_button = gr.Button("Send")
 
-    submit_btn.click(
-        fn=chat_with_model,
-        inputs=[model_dropdown, prompt_input, image_url_input, state],
-        outputs=[output_box, state, prompt_input, image_url_input]
-    )
+    # Hidden chat history state
+    chat_memory = gr.State([])
 
-    reset_btn.click(
-        fn=lambda: ("🧹 Conversation reset. You can start a new one.", [], "", ""),
-        inputs=[],
-        outputs=[output_box, state, prompt_input, image_url_input]
+    run_button.click(
+        fn=run_model_with_history,
+        inputs=[model_choice, prompt_input, chat_memory],
+        outputs=[chat_display, chat_memory]
     )
 
+# Launch app
 demo.launch()