commit

app.py

@@ -411,7 +411,6 @@ CHECKPOINTS = ESRGANUpscalerCheckpoints(
     }
 )
 
-# initialize the enhancer, on the cpu
 device = DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 DTYPE = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
 enhancer = ESRGANUpscaler(checkpoints=CHECKPOINTS, device=DEVICE, dtype=DTYPE)
@@ -505,6 +504,9 @@ function custom(){
 image_pipe = FluxPipeline.from_pretrained(base, torch_dtype=torch.bfloat16).to(device)
 #image_pipe.enable_model_cpu_offload()
 
+pegasus_tokenizer = PegasusTokenizerFast.from_pretrained("google/pegasus-xsum"),
+pegasus_model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum")
+
 # functionality
 
 @spaces.GPU(duration=180)
@@ -513,7 +515,7 @@ def upscaler(
     prompt: str = "masterpiece, best quality, highres",
     negative_prompt: str = "worst quality, low quality, normal quality",
     seed: int = int(str(random.random()).split(".")[1]),
-    upscale_factor: int = 8,
+    upscale_factor: int = 4,
     controlnet_scale: float = 0.6,
     controlnet_decay: float = 1.0,
     condition_scale: int = 6,
@@ -553,9 +555,7 @@ def upscaler(
 
 @spaces.GPU(duration=180)
 def summarize_text(
-    text, max_length=30, num_beams=16, early_stopping=True, 
-    pegasus_tokenizer = PegasusTokenizerFast.from_pretrained("google/pegasus-xsum"),
-    pegasus_model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum")
+    text, max_length=30, num_beams=16, early_stopping=True
 ):
     log(f'CALL summarize_text')
     summary = pegasus_tokenizer.decode( pegasus_model.generate(