show manifest output

app.py

@@ -4,38 +4,60 @@ import spaces
 import torch
 import os
 import re
+import threading
+import queue
 
 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."
+        return "Error: HF_TOKEN not found in environment variables.", "Error: Cannot start manifest server without HF_TOKEN."
 
     manifest_process = None
+    log_queue = queue.Queue()
     try:
         # Start manifest server in background with explicit CUDA_VISIBLE_DEVICES
         manifest_cmd = f"""
-        CUDA_VISIBLE_DEVICES=0 HF_TOKEN={os.environ['HF_TOKEN']} cd duckdb-nsql/ && \
-        python -m manifest.api.app \
+        cd duckdb-nsql/ && 
+        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.PIPE)
+        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()
+
         results.append("Started manifest server in background.")
 
+        # Wait for the server to initialize (adjust time as needed)
+        for _ in range(30):
+            try:
+                line = log_queue.get(timeout=1)
+                manifest_logs.append(line)
+                if "Running on" in line:  # Server is ready
+                    break
+            except queue.Empty:
+                pass
+
         # Run inference
         inference_cmd = f"""
-        cd duckdb-nsql/ && \
+        cd duckdb-nsql/ && 
         python eval/predict.py \
         predict \
         eval/data/dev.json \
@@ -59,7 +81,7 @@ def run_evaluation(model_name):
 
         # Run evaluation
         eval_cmd = f"""
-        cd duckdb-nsql/ && \
+        cd duckdb-nsql/ && 
         python eval/evaluate.py evaluate \
         --gold eval/data/dev.json \
         --db eval/data/databases/ \
@@ -74,7 +96,7 @@ def run_evaluation(model_name):
         if metrics:
             results.append(f"Evaluation completed:\n{metrics}")
         else:
-            results.append("Evaluation completed, but get metrics.")
+            results.append("Evaluation completed, but couldn't get metrics.")
 
     except subprocess.CalledProcessError as e:
         results.append(f"Error occurred: {str(e)}")
@@ -87,15 +109,24 @@ def run_evaluation(model_name):
             manifest_process.terminate()
             results.append("Terminated manifest server.")
 
-    return "\n\n".join(results)
+        # 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)
 
 with gr.Blocks() as demo:
-    gr.Markdown("# DuckDB-NSQL Evaluation App")
+    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="Output", lines=20)
+    output = gr.Textbox(label="Evaluation Output", lines=20)
+    manifest_output = gr.Textbox(label="Manifest Server Logs", lines=20)
 
-    start_btn.click(fn=run_evaluation, inputs=[model_name], outputs=output)
+    start_btn.click(fn=run_evaluation, inputs=[model_name], outputs=[output, manifest_output])
 
 demo.launch()