refactor: Simplify model configuration by replacing dynamic GPU mapping with a static dictionary, and enhance bot response function to include a seed value for reproducibility in responses.

app.py

@@ -5,63 +5,26 @@
 
 import os
 import json
+import random
 import datetime
 import gradio as gr
 import pandas as pd
-import subprocess
-import time
 from pathlib import Path
 from huggingface_hub import CommitScheduler
 from huggingface_hub import InferenceClient
 
 HF_TOKEN = os.getenv("HF_TOKEN")
 
-MODELS = dict()
-
-# Launch models via vLLM
-model_gpu_mapping = [
-    # (0, 1000), 
-    # (0, 1500),
-    # (1, 2000), 
-    # (1, 2500),
-    # (2, 3000), 
-    # (2, 3500),
-    (2, 4000), 
-    # (3, 4500),
-    (2, 5000), 
-    # (4, 5500),
-    (3, 6000), 
-    # (5, 6500),
-    (3, 7000), 
-    # (6, 7500),
-]
-
-for index, (gpu_id, iter_num) in enumerate(model_gpu_mapping):
-    formatted_iter_num = f"{iter_num:07d}"
-    model_name = f"Elfsong/VLM_stage_2_iter_{formatted_iter_num}"
-    arena_key = f"Local-Model-{iter_num:05d}"
-
-    port = 9000 + index
-    print(f"🚀 Launching {model_name} on port {port} (GPU {gpu_id}) ...")
-    log_file = open(f"./logs/vllm_{formatted_iter_num}.log", "w")
-    
-    subprocess.Popen(
-        [
-            "python", "-m", "vllm.entrypoints.openai.api_server",
-            "--model", model_name,
-            "--port", str(port),
-            "--quantization", "bitsandbytes",
-            "--gpu-memory-utilization", "0.3",
-            "--trust-remote-code",
-        ],
-        env={**os.environ, "CUDA_VISIBLE_DEVICES": str(gpu_id)},
-        stdout=log_file,
-        stderr=log_file,
-    )
-
-    time.sleep(5) # Wait for initialization
-    MODELS[arena_key] = f"http://localhost:{port}/v1"
-print(f"✅ Launched {len(MODELS)} models. Check logs in ./logs/ directory.")
+# Model configuration - these should match the models launched by launch_models.py
+MODELS = {
+    "Local-Model-00500": "http://localhost:9000/v1",
+    "Local-Model-01000": "http://localhost:9001/v1",
+    "Local-Model-01500": "http://localhost:9002/v1",
+    "Local-Model-02000": "http://localhost:9003/v1",
+    "Local-Model-02500": "http://localhost:9004/v1",
+    "Local-Model-03000": "http://localhost:9005/v1",
+    "Local-Model-03500": "http://localhost:9006/v1",
+}
 
 DATA_DIR = Path("logs")
 DATA_DIR.mkdir(exist_ok=True)
@@ -88,7 +51,7 @@ def save_feedback(model_name, history, feedback_data: gr.LikeData):
     
     print(f"Feedback logged for {model_name}")
 
-def bot_response(user_message, history, model_name, system_message, thinking_mode, max_tokens, temperature, top_p):
+def bot_response(user_message, history, model_name, system_message, thinking_mode, max_tokens, temperature, top_p, seed_val):
     if not user_message or user_message.strip() == "":
         yield history, ""
         return
@@ -114,6 +77,7 @@ def bot_response(user_message, history, model_name, system_message, thinking_mod
             temperature=temperature,
             top_p=top_p,
             model=model_name,
+            seed=seed_val,
         )
         
         response_text = ""
@@ -145,6 +109,7 @@ with gr.Blocks() as demo:
         max_t = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max Tokens")
         temp = gr.Slider(minimum=0.0, maximum=2.0, value=0.0, step=0.05, label="Temperature")
         top_p_val = gr.Slider(minimum=0.0, maximum=1.0, value=1.0, step=0.05, label="Top-p")
+        seed_val = gr.Slider(minimum=-1, maximum=4294967295, value=random.randint(0, 4294967295), step=1, label="Seed")
 
     gr.Markdown("# ⚔️ Chatbot Arena")
 
@@ -164,12 +129,12 @@ with gr.Blocks() as demo:
             btn_b = gr.Button("Send to Model B")
 
     # --- Bind Events ---
-    a_inputs = [msg_a, chatbot_a, model_a_name, system_msg, thinking_mode, max_t, temp, top_p_val]
+    a_inputs = [msg_a, chatbot_a, model_a_name, system_msg, thinking_mode, max_t, temp, top_p_val, seed_val]
     msg_a.submit(bot_response, a_inputs, [chatbot_a, msg_a])
     btn_a.click(bot_response, a_inputs, [chatbot_a, msg_a])
     chatbot_a.like(save_feedback, [model_a_name, chatbot_a], None)
 
-    b_inputs = [msg_b, chatbot_b, model_b_name, system_msg, thinking_mode, max_t, temp, top_p_val]
+    b_inputs = [msg_b, chatbot_b, model_b_name, system_msg, thinking_mode, max_t, temp, top_p_val, seed_val]
     msg_b.submit(bot_response, b_inputs, [chatbot_b, msg_b])
     btn_b.click(bot_response, b_inputs, [chatbot_b, msg_b])
     chatbot_b.like(save_feedback, [model_b_name, chatbot_b], None)
@@ -185,4 +150,4 @@ with gr.Blocks() as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", share=False)
+    demo.launch(share=True)