Update src/pipeline.py

src/pipeline.py

@@ -9,7 +9,6 @@ from transformers import T5EncoderModel, T5TokenizerFast, CLIPTokenizer, CLIPTex
 import diffusers
 import gc
 from diffusers import FluxTransformer2DModel, DiffusionPipeline
-from optimum.quanto import freeze, qfloat8, quantize
 
 torch.set_float32_matmul_precision("high")
 
@@ -32,13 +31,11 @@ def fart():
 
 def load_pipeline() -> Pipeline:
     fart()
-    transformer = FluxTransformer2DModel.from_pretrained(ckpt_id, subfolder="transformer", torch_dtype=torch.bfloat16)
-    quantize(transformer, weights=qfloat8)
-    freeze(transformer)
+
     pipeline = DiffusionPipeline.from_pretrained(ckpt_id, torch_dtype=torch.bfloat16)
-    pipeline.transformer = transformer
     pipeline.vae = torch.compile(pipeline.vae, mode="max-autotune", fullgraph=False)
     pipeline.enable_sequential_cpu_offload()
+    
     for _ in range(2):
         pipeline(prompt="insensible, timbale, pothery, electrovital, actinogram, taxis, intracerebellar, centrodesmus", width=1024, height=1024, guidance_scale=0.0, num_inference_steps=4, max_sequence_length=256)
     fart()