Update app.py

app.py

@@ -7,6 +7,7 @@ import torch
 import gradio as gr
 from threading import Lock
 from contextlib import contextmanager
+from huggingface_hub import snapshot_download
 
 # --- LOGGING FOR UI ---
 LOG_BUFFER = []
@@ -20,10 +21,10 @@ def log(message):
             LOG_BUFFER.pop(0)
     return "\n".join(LOG_BUFFER)
 
-# 🚀 Initialization log
+# 🚀 Initialization
 _initial_logs = log("🚀 Initializing Ultimate Z-Image Turbo CPU Edition...")
 
-# CPU THREAD OPTIMIZATION
+# --- ENVIRONMENT SETUP ---
 CPU_THREADS = min(8, os.cpu_count() or 1)
 os.environ["OMP_NUM_THREADS"] = str(CPU_THREADS)
 os.environ["MKL_NUM_THREADS"] = str(CPU_THREADS)
@@ -35,6 +36,9 @@ os.environ["HF_HUB_DISABLE_TELEMETRY"] = "1"
 os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "0"
 os.environ["TRANSFORMERS_CACHE"] = "./hf_cache"
 os.environ["HF_DATASETS_CACHE"] = "./hf_cache"
+os.environ["HF_HUB_OFFLINE"] = "1"
+os.environ["TRANSFORMERS_OFFLINE"] = "1"
+os.environ["HF_DATASETS_OFFLINE"] = "1"
 
 torch.set_num_threads(CPU_THREADS)
 torch.set_grad_enabled(False)
@@ -61,6 +65,23 @@ pipe = None
 _pipe_lock = Lock()
 _generation_lock = Lock()
 
+# --- Pre-download full snapshot once ---
+MODEL_ID = "Tongyi-MAI/Z-Image-Turbo"
+MODEL_LOCAL = os.path.join(CACHE_DIR, "Z-Image-Turbo-snapshot")
+os.makedirs(MODEL_LOCAL, exist_ok=True)
+
+if not os.listdir(MODEL_LOCAL):
+    log("📥 Downloading full model snapshot, please wait...")
+    snapshot_download(
+        repo_id=MODEL_ID,
+        cache_dir=MODEL_LOCAL,
+        local_dir=MODEL_LOCAL,
+        local_dir_use_symlinks=False
+    )
+    log(f"📦 Model snapshot cached at: {MODEL_LOCAL}")
+else:
+    log(f"📦 Model snapshot already exists at: {MODEL_LOCAL}")
+
 @contextmanager
 def managed_memory():
     try:
@@ -81,9 +102,9 @@ def load_pipeline():
         start_load = time.time()
 
         pipe = ZImagePipeline.from_pretrained(
-            "Tongyi-MAI/Z-Image-Turbo",
+            MODEL_LOCAL,
             torch_dtype=DTYPE,
-            cache_dir=CACHE_DIR,
+            local_files_only=True,
             low_cpu_mem_usage=True
         )
 
@@ -163,6 +184,7 @@ def generate(prompt, quality_mode, seed, progress=gr.Progress()):
 
             return image, seed
 
+# --- GRADIO UI ---
 with gr.Blocks(title="🚀 Z-Image Turbo Pro Max + Live Logs") as demo:
     gr.Markdown("## GPU‑FREE CPU Turbo — Live Logs Below")