remove offload-to-cpu (disk swapping), keep preload + other opts, ~42 min timing

app.py

@@ -62,13 +62,13 @@ def generate(prompt, negative_prompt, resolution, seed):
         "--cfg-scale", str(CFG),
         "--sampling-method", "euler_a",
         "-o", output_path,
-        "--offload-to-cpu",
         "--diffusion-fa",
         "--diffusion-conv-direct",
         "--vae-tiling",
         "--vae-conv-direct",
         "--tensor-type-rules", "^vae=f32",
-        "--mmap",
+        # "--offload-to-cpu",  # causes disk swapping with --mmap, slower
+        # "--mmap",            # use with preload only if --offload-to-cpu is off
         "-v",
     ]
     if seed >= 0:
@@ -128,7 +128,7 @@ with gr.Blocks(title="Z-Anime (CPU)") as demo:
     gr.Markdown(
         "**[Z-Anime 6B](https://huggingface.co/SeeSee21/Z-Anime)** S3-DiT Q5_0 GGUF "
         "(distill 4-step) via [sd.cpp](https://github.com/leejet/stable-diffusion.cpp) | "
-        "~25 min at 512x512 on free CPU"
+        "~42 min at 512x512 on free CPU"
     )
     with gr.Row():
         with gr.Column():