Add debug logging to trace startup

app.py

@@ -1,10 +1,21 @@
+import sys
+print("Step 1: import start", flush=True)
+
 import os
 import torch
+print("Step 2: torch imported", flush=True)
+
 import spaces
+print("Step 3: spaces imported", flush=True)
+
 import gradio as gr
+print("Step 4: gradio imported", flush=True)
+
 from transformers import AutoModelForCausalLM, AutoTokenizer
+print("Step 5: transformers imported", flush=True)
 
 MODEL_ID = os.getenv("MODEL_ID", "GnLOLot/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking")
+print(f"Step 6: MODEL_ID = {MODEL_ID}", flush=True)
 
 model = None
 tokenizer = None
@@ -13,6 +24,7 @@ tokenizer = None
 def chat_fn(message, history):
     global model, tokenizer
     if model is None:
+        print("Loading model...", flush=True)
         tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
         model = AutoModelForCausalLM.from_pretrained(
             MODEL_ID,
@@ -22,6 +34,7 @@ def chat_fn(message, history):
         )
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
+        print("Model loaded", flush=True)
 
     messages = []
     for h in history:
@@ -44,6 +57,8 @@ def chat_fn(message, history):
 
     return tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True).strip()
 
+print("Step 7: function defined, creating Blocks...", flush=True)
+
 with gr.Blocks(title="MiniCPM5-1B Chat") as demo:
     gr.Markdown(f"# MiniCPM5-1B Chat\n**Model:** `{MODEL_ID}`\n\nPowered by ZeroGPU free GPU")
     gr.ChatInterface(
@@ -51,3 +66,5 @@ with gr.Blocks(title="MiniCPM5-1B Chat") as demo:
         title=None,
         description="First request loads the model (~30s), subsequent calls are faster."
     )
+
+print("Step 8: demo created successfully", flush=True)