Update src/pipeline.py

src/pipeline.py

@@ -593,14 +593,13 @@ def load_pipeline() -> Pipeline:
     dtype, device = torch.bfloat16, "cuda"
     
     text_encoder_2 = T5EncoderModel.from_pretrained(
-        "silentdriver/aadb864af9", revision = "060dabc7fa271c26dfa3fd43c16e7c5bf3ac7892", torch_dtype=torch.bfloat16
+        "city96/t5-v1_1-xxl-encoder-bf16", revision = "1b9c856aadb864af93c1dcdc226c2774fa67bc86", torch_dtype=torch.bfloat16
     ).to(memory_format=torch.channels_last)
-
     
     
-    vae = AutoencoderTiny.from_pretrained("silentdriver/7815792fb4", revision="bdb7d88ebe5a1c6b02a3c0c78651dd57a403fdf5", torch_dtype=dtype)
+    vae = AutoencoderTiny.from_pretrained("RobertML/FLUX.1-schnell-vae_int8", revision="2e4fdf1337ed12ed202e23e6f25269aa7d6d3f2a", torch_dtype=dtype)
     
-    path = os.path.join(HF_HUB_CACHE, "models--silentdriver--7d92df966a/snapshots/add1b8d9a84c728c1209448c4a695759240bad3c")
+    path = os.path.join(HF_HUB_CACHE, "models--RobertML--FLUX.1-schnell-int8wo/snapshots/307e0777d92df966a3c0f99f31a6ee8957a9857a")
     generator = torch.Generator(device=device)
     model = FluxTransformer2DModel.from_pretrained(path, torch_dtype=dtype, use_safetensors=False, generator= generator).to(memory_format=torch.channels_last)
     torch.backends.cudnn.benchmark = True