Upload src/pipeline.py with huggingface_hub

src/pipeline.py

@@ -55,17 +55,17 @@ def load_pipeline() -> Pipeline:
 
     pipeline.to(memory_format=torch.channels_last)
     pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=False)
-    quantize_(pipeline.vae, int8_weight_only())
-    pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")
+    # quantize_(pipeline.vae, int8_weight_only())
+    # pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")
     
     PROMPT = 'semiconformity, peregrination, quip, twineless, emotionless, tawa, depickle'
-    with torch.inference_mode():  
+    with torch.no_grad():  
         for _ in range(4):
             pipeline(prompt="onomancy, aftergo, spirantic, Platyhelmia, modificator, drupaceous, jobbernowl, hereness", width=1024, height=1024, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256)
     torch.cuda.empty_cache()
     return pipeline
 
-@torch.inference_mode()
+@torch.no_grad()
 def infer(request: TextToImageRequest, pipeline: Pipeline, generator: torch.Generator) -> Image:
 
     return pipeline(