Update app.py

app.py

@@ -45,6 +45,8 @@ MAX_FRAMES_MODEL = 160
 MIN_DURATION = round(MIN_FRAMES_MODEL / FIXED_FPS, 1)
 MAX_DURATION = round(MAX_FRAMES_MODEL / FIXED_FPS, 1)
 
+CACHE_DIR = os.path.expanduser("~/.cache/huggingface/")
+
 SCHEDULER_MAP = {
     "FlowMatchEulerDiscrete": FlowMatchEulerDiscreteScheduler,
     "SASolver": SASolverScheduler,
@@ -62,6 +64,12 @@ pipe = WanImageToVideoPipeline.from_pretrained(
 original_scheduler = copy.deepcopy(pipe.scheduler)
 print(original_scheduler)
 
+if os.path.exists(CACHE_DIR):
+    shutil.rmtree(CACHE_DIR)
+    print("Deleted Hugging Face cache.")
+else:
+    print("No hub cache found.")
+
 quantize_(pipe.text_encoder, Int8WeightOnlyConfig())
 quantize_(pipe.transformer, Float8DynamicActivationFloat8WeightConfig())
 quantize_(pipe.transformer_2, Float8DynamicActivationFloat8WeightConfig())
@@ -326,7 +334,8 @@ def interpolate_video(generated_video, multiplier, request: gr.Request):
 def generate_interpolate(generated_video, multiplier, request: gr.Request):
     return interpolate_video(generated_video, multiplier, request)
 
-with gr.Blocks() as demo:
+
+with gr.Blocks(delete_cache=(3600, 10800)) as demo:
     gr.Markdown("# WAMU - Wan 2.2 I2V (14B)")
     gr.Markdown("## ℹ️ **A Note on Performance:** This version prioritizes a straightforward setup over maximum speed, so performance may vary.")
     gr.Markdown("run Wan 2.2 in just 4-8 steps, fp8 quantization & AoT compilation - compatible with 🧨 diffusers and ZeroGPU⚡️")