Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,81 @@
+import os
+import logging
+import threading
+import spaces
+import gradio as gr
+from huggingface_hub import hf_hub_download
+from llama_cpp import Llama
+
+logging.basicConfig(level=logging.INFO)
+log = logging.getLogger(__name__)
+
+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)
+
+_llm = None
+_downloaded = threading.Event()
+
+def download_model():
+    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()
+
+threading.Thread(target=download_model, daemon=True).start()
+
+@spaces.GPU
+def generate(messages, max_tokens=1024, temperature=0.7, top_p=0.95):
+    global _llm
+    if _llm is None:
+        _downloaded.wait()
+        log.info("Loading model into GPU...")
+        _llm = Llama(
+            model_path=MODEL_PATH,
+            n_gpu_layers=-1,
+            n_ctx=8192,
+            verbose=False
+        )
+        log.info("Model loaded")
+    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):
+    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)
+
+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!"
+    )
+
+    gr.Markdown("---\nFirst request is slow (~5 min) while the 7.38 GB model downloads. Subsequent requests are fast.")
+
+demo.queue(default_concurrency_limit=1)
+
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)