Update src/pipeline.py

src/pipeline.py

@@ -13,7 +13,7 @@ from diffusers import FluxTransformer2DModel, DiffusionPipeline
 from torchao.quantization import quantize_, int8_weight_only, fpx_weight_only
 
 os.environ['PYTORCH_CUDA_ALLOC_CONF']="expandable_segments:True"
-os.environ["TOKENIZERS_PARALLELISM"] = "True"
+os.environ["TOKENIZERS_PARALLELISM"] = "False"
 torch._dynamo.config.suppress_errors = True
 
 Pipeline = None
@@ -21,12 +21,11 @@ ids = "black-forest-labs/FLUX.1-schnell"
 Revision = "741f7c3ce8b383c54771c7003378a50191e9efe9"
 
 def load_pipeline() -> Pipeline:
-    vae = AutoencoderKL.from_pretrained(ids,revision=Revision, subfolder="vae", local_files_only=True, torch_dtype=torch.bfloat16,)
-    quantize_(vae, int8_weight_only())
+    vae = AutoencoderTiny.from_pretrained("slobers/tt1",revision="ec746bf42d91e3335760895281f070df54f2196a", torch_dtype=torch.bfloat16,)
     text_encoder_2 = T5EncoderModel.from_pretrained("city96/t5-v1_1-xxl-encoder-bf16", revision = "1b9c856aadb864af93c1dcdc226c2774fa67bc86", torch_dtype=torch.bfloat16).to(memory_format=torch.channels_last)
     path = os.path.join(HF_HUB_CACHE, "models--RobertML--FLUX.1-schnell-int8wo/snapshots/307e0777d92df966a3c0f99f31a6ee8957a9857a")
     transformer = FluxTransformer2DModel.from_pretrained(path, torch_dtype=torch.bfloat16, use_safetensors=False).to(memory_format=torch.channels_last)
-    pipeline = DiffusionPipeline.from_pretrained(ids, revision=Revision, transformer=transformer, text_encoder_2=text_encoder_2, torch_dtype=torch.bfloat16,)
+    pipeline = DiffusionPipeline.from_pretrained(ids, revision=Revision, vae=vae, transformer=transformer, text_encoder_2=text_encoder_2, torch_dtype=torch.bfloat16,)
     pipeline.to("cuda")
 
     for _ in range(3):