Upload src/pipeline.py with huggingface_hub

src/pipeline.py

@@ -30,10 +30,9 @@ REVISION = "5ef0012f11a863e5111ec56540302a023bc8587b"
 TinyVAE = "madebyollin/taef1"
 TinyVAE_REV = "2d552378e58c9c94201075708d7de4e1163b2689"
 
-# @torch.inference_mode()
+
 def load_pipeline() -> Pipeline:
-    # path = os.path.join(HF_HUB_CACHE, "models--jokerbit--flux.1-schnell-Robert-int8wo/snapshots/5ef0012f11a863e5111ec56540302a023bc8587b/transformer")
-    path = os.path.join(HF_HUB_CACHE, "models--RobertML--FLUX.1-schnell-int8wo/snapshots/307e0777d92df966a3c0f99f31a6ee8957a9857a")
+    path = os.path.join(HF_HUB_CACHE, "models--jokerbit--flux.1-schnell-Robert-int8wo/snapshots/5ef0012f11a863e5111ec56540302a023bc8587b/transformer") 
     transformer = FluxTransformer2DModel.from_pretrained(
             path, 
             use_safetensors=False,
@@ -49,19 +48,19 @@ def load_pipeline() -> Pipeline:
     ).to("cuda")
 
     pipeline.transformer.to(memory_format=torch.channels_last)
-    pipeline.transformer = torch.compile(pipeline.transformer, mode="reduce-overhead")
+    pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=True)
     # pipeline.vae.to(memory_format=torch.channels_last)
     # quantize_(pipeline.vae, int8_weight_only())
     # pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")
     # pipeline.to("cuda")
     PROMPT = 'semiconformity, peregrination, quip, twineless, emotionless, tawa, depickle'
-  
-    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)
+    with torch.inference_mode():  
+        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.inference_mode()
 def infer(request: TextToImageRequest, pipeline: Pipeline, generator: torch.Generator) -> Image:
 
     return pipeline(