revert changes

app.py

@@ -4,61 +4,35 @@ import spaces
 import torch
 import os
 import re
-import threading
-import queue
-import time
 
 zero = torch.Tensor([0]).cuda()
 print(zero.device)  # <-- 'cpu' 🤔
 
-def stream_output(process, q):
-    for line in iter(process.stdout.readline, b''):
-        q.put(line.decode('utf-8').strip())
-    process.stdout.close()
-
 @spaces.GPU
 def run_evaluation(model_name):
     print(zero.device)  # <-- 'cuda:0' 🤗
 
     results = []
-    manifest_logs = []
 
+    # Use the secret HF token from the Hugging Face space
     if "HF_TOKEN" not in os.environ:
-        return "Error: HF_TOKEN not found in environment variables.", "Error: Cannot start manifest server without HF_TOKEN."
+        return "Error: HF_TOKEN not found in environment variables."
 
     manifest_process = None
-    log_queue = queue.Queue()
     try:
+        # Start manifest server in background with explicit CUDA_VISIBLE_DEVICES
         manifest_cmd = f"""
         cd duckdb-nsql/ && 
-        CUDA_VISIBLE_DEVICES=0 HF_TOKEN={os.environ['HF_TOKEN']} python -m manifest.api.app \
+        CUDA_VISIBLE_DEVICES=0 HF_TOKEN={os.environ['HF_TOKEN']}  python -m manifest.api.app \
         --model_type huggingface \
         --model_generation_type text-generation \
         --model_name_or_path {model_name} \
         --fp16 \
         --device 0
         """
-        manifest_process = subprocess.Popen(manifest_cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True, bufsize=1, universal_newlines=True)
-
-        threading.Thread(target=stream_output, args=(manifest_process, log_queue), daemon=True).start()
-
+        manifest_process = subprocess.Popen(manifest_cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
         results.append("Started manifest server in background.")
 
-        # Wait for the server to initialize (adjust time as needed)
-        server_ready = False
-        for _ in range(60):  # Increased timeout to 60 seconds
-            try:
-                line = log_queue.get(timeout=1)
-                manifest_logs.append(line)
-                if "Running on" in line:  # Server is ready
-                    server_ready = True
-                    break
-            except queue.Empty:
-                pass
-
-        if not server_ready:
-            raise Exception("Manifest server failed to start within the expected time.")
-
         # Run inference
         inference_cmd = f"""
         cd duckdb-nsql/ && 
@@ -100,7 +74,7 @@ def run_evaluation(model_name):
         if metrics:
             results.append(f"Evaluation completed:\n{metrics}")
         else:
-            results.append("Evaluation completed, but couldn't get metrics.")
+            results.append("Evaluation completed, but get metrics.")
 
     except subprocess.CalledProcessError as e:
         results.append(f"Error occurred: {str(e)}")
@@ -113,24 +87,15 @@ def run_evaluation(model_name):
             manifest_process.terminate()
             results.append("Terminated manifest server.")
 
-        # Collect any remaining logs
-        while True:
-            try:
-                line = log_queue.get_nowait()
-                manifest_logs.append(line)
-            except queue.Empty:
-                break
-
-    return "\n\n".join(results), "\n".join(manifest_logs)
+    return "\n\n".join(results)
 
 with gr.Blocks() as demo:
     gr.Markdown("# DuckDB SQL Evaluation App")
 
     model_name = gr.Textbox(label="Model Name (e.g., Qwen/Qwen2.5-7B-Instruct)")
     start_btn = gr.Button("Start Evaluation")
-    output = gr.Textbox(label="Evaluation Output", lines=20)
-    manifest_output = gr.Textbox(label="Manifest Server Logs", lines=20)
+    output = gr.Textbox(label="Output", lines=20)
 
-    start_btn.click(fn=run_evaluation, inputs=[model_name], outputs=[output, manifest_output])
+    start_btn.click(fn=run_evaluation, inputs=[model_name], outputs=output)
 
 demo.launch()