Upload app.py with huggingface_hub

app.py

@@ -1,6 +1,5 @@
 import os
 import logging
-import threading
 import traceback
 import spaces
 import gradio as gr
@@ -13,38 +12,14 @@ MODEL_REPO = "sakamakismile/gemma-4-12B-coder-fable5-composer2.5-GGUF"
 MODEL_FILE = "gemma-4-12B-coder-fable5-composer2.5-Q4_K_M.gguf"
 MODEL_PATH = os.path.join(os.path.dirname(os.path.abspath(__file__)), MODEL_FILE)
 
-def ensure_model():
-    if not os.path.exists(MODEL_PATH):
-        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__)),
-        )
-        log.info("Download complete")
-
-threading.Thread(target=ensure_model, daemon=True).start()
-
-_llm = None
-_llm_lock = threading.Lock()
-
-def load_model():
-    global _llm
-    if _llm is not None:
-        return _llm
-    with _llm_lock:
-        if _llm is not None:
-            return _llm
-        log.info("Loading model into GPU...")
-        from llama_cpp import Llama as _Llama
-        _llm = _Llama(
-            model_path=MODEL_PATH,
-            n_gpu_layers=-1,
-            n_ctx=8192,
-            verbose=False
-        )
-        log.info("Model loaded")
-        return _llm
+if not os.path.exists(MODEL_PATH):
+    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__)),
+    )
+    log.info("Download complete")
 
 @spaces.GPU
 def cuda_test():
@@ -57,28 +32,40 @@ def cuda_test():
 
 @spaces.GPU
 def generate(messages, max_tokens=1024, temperature=0.7, top_p=0.95):
-    llm = load_model()
-    log.info(f"Generating (max_tokens={max_tokens}, temp={temperature})")
-    output = llm.create_chat_completion(
-        messages=messages,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        top_p=top_p,
-    )
-    return output["choices"][0]["message"]["content"].strip()
-
-def predict(message, history):
     try:
-        if not os.path.exists(MODEL_PATH):
-            return "Model is still downloading... Please wait ~5 minutes and try again."
-        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)
+        log.info("Importing llama_cpp...")
+        from llama_cpp import Llama as _Llama
+        log.info("Import OK, loading model...")
+        llm = _Llama(
+            model_path=MODEL_PATH,
+            n_gpu_layers=-1,
+            n_ctx=8192,
+            verbose=False
+        )
+        log.info("Model loaded, generating...")
+        output = llm.create_chat_completion(
+            messages=messages,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+        )
+        result = output["choices"][0]["message"]["content"].strip()
+        log.info("Generation complete")
+        return result
     except Exception as e:
-        return f"Error: {traceback.format_exc()}"
+        err = f"GPU Error: {type(e).__name__}: {e}\n{traceback.format_exc()}"
+        log.error(err)
+        return err
+
+def predict(message, history):
+    if not os.path.exists(MODEL_PATH):
+        return "Model is still downloading... Please wait ~5 minutes and try again."
+    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)
 
 def check_status():
     exists = os.path.exists(MODEL_PATH)