Upload app.py with huggingface_hub

app.py

@@ -1,6 +1,7 @@
 import os
 import logging
 import threading
+import traceback
 import spaces
 import gradio as gr
 from huggingface_hub import hf_hub_download
@@ -14,30 +15,50 @@ MODEL_PATH = os.path.join(os.path.dirname(os.path.abspath(__file__)), MODEL_FILE
 
 _llm = None
 _downloaded = threading.Event()
+_download_error = None
 
 def download_model():
+    global _download_error
     if os.path.exists(MODEL_PATH):
         log.info("Model already downloaded")
         _downloaded.set()
         return
     log.info("Downloading model (7.38 GB)...")
-    hf_hub_download(
-        repo_id=MODEL_REPO,
-        filename=MODEL_FILE,
-        local_dir=os.path.dirname(os.path.abspath(__file__)),
-        resume=True
-    )
-    log.info("Download complete")
-    _downloaded.set()
+    try:
+        hf_hub_download(
+            repo_id=MODEL_REPO,
+            filename=MODEL_FILE,
+            local_dir=os.path.dirname(os.path.abspath(__file__)),
+            resume=True
+        )
+        log.info("Download complete")
+    except Exception as e:
+        _download_error = str(e)
+        log.error(f"Download failed: {e}")
+    finally:
+        _downloaded.set()
 
 threading.Thread(target=download_model, daemon=True).start()
 
+@spaces.GPU
+def cuda_test():
+    import torch
+    return {
+        "cuda_available": torch.cuda.is_available(),
+        "device_count": torch.cuda.device_count(),
+        "device_name": torch.cuda.get_device_name(0) if torch.cuda.is_available() else "N/A",
+    }
+
 @spaces.GPU
 def generate(messages, max_tokens=1024, temperature=0.7, top_p=0.95):
     global _llm
+    _downloaded.wait()
+
+    if _download_error:
+        raise RuntimeError(f"Model download failed: {_download_error}")
+
     if _llm is None:
         from llama_cpp import Llama as _Llama
-        _downloaded.wait()
         log.info("Loading model into GPU...")
         _llm = _Llama(
             model_path=MODEL_PATH,
@@ -46,6 +67,7 @@ def generate(messages, max_tokens=1024, temperature=0.7, top_p=0.95):
             verbose=False
         )
         log.info("Model loaded")
+
     log.info(f"Generating (max_tokens={max_tokens}, temp={temperature})")
     output = _llm.create_chat_completion(
         messages=messages,
@@ -56,22 +78,36 @@ def generate(messages, max_tokens=1024, temperature=0.7, top_p=0.95):
     return output["choices"][0]["message"]["content"].strip()
 
 def predict(message, history):
-    messages = []
-    for user_msg, assistant_msg in history:
-        messages.append({"role": "user", "content": user_msg})
-        messages.append({"role": "assistant", "content": assistant_msg})
-    messages.append({"role": "user", "content": message})
-    return generate(messages)
+    try:
+        messages = []
+        for user_msg, assistant_msg in history:
+            messages.append({"role": "user", "content": user_msg})
+            messages.append({"role": "assistant", "content": assistant_msg})
+        messages.append({"role": "user", "content": message})
+        return generate(messages)
+    except Exception as e:
+        return f"Error: {e}\n\n{traceback.format_exc()}"
+
+def check_status():
+    dl_status = "downloading" if not _downloaded.is_set() else ("done" if not _download_error else f"failed: {_download_error}")
+    cuda = cuda_test()
+    return f"Model: {dl_status}\nGPU: {cuda}"
 
 with gr.Blocks(title="Gemma Coder Zero", theme=gr.themes.Soft()) as demo:
     gr.Markdown("# Gemma 4 12B Coder Zero")
     gr.Markdown("Powered by llama.cpp on ZeroGPU (RTX Pro 6000 Blackwell)")
 
-    gr.ChatInterface(
-        fn=predict,
-        title="Gemma Coder",
-        description="Ask any coding question!"
-    )
+    with gr.Tabs():
+        with gr.TabItem("Chat"):
+            gr.ChatInterface(
+                fn=predict,
+                title="Gemma Coder",
+                description="Ask any coding question!"
+            )
+        with gr.TabItem("Status"):
+            status_btn = gr.Button("Check GPU & Model Status")
+            status_out = gr.Textbox(label="Status")
+            status_btn.click(fn=check_status, outputs=status_out)
 
     gr.Markdown("---\nFirst request is slow (~5 min) while the 7.38 GB model downloads. Subsequent requests are fast.")