Update app.py

app.py

@@ -20,8 +20,12 @@ HF_TOKEN = os.getenv("HF_TOKEN")
 # Ensure that the minimal version of diffusers is installed
 check_min_version("0.30.2")
 
-quant_config = BitsAndBytesConfig(load_in_8bit=True)
-text_encoder_8bit = T5EncoderModel.from_pretrained(
+quant_config = TransformersBitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_use_double_quant=True,
+)
+
+text_encoder_2_4bit = T5EncoderModel.from_pretrained(
     "black-forest-labs/FLUX.1-dev",
     subfolder="text_encoder_2",
     quantization_config=quant_config,
@@ -29,6 +33,14 @@ text_encoder_8bit = T5EncoderModel.from_pretrained(
     use_safetensors=True,
     token=HF_TOKEN
 )
+# text_encoder_8bit = T5EncoderModel.from_pretrained(
+#     "black-forest-labs/FLUX.1-dev",
+#     subfolder="text_encoder_2",
+#     quantization_config=quant_config,
+#     torch_dtype=torch.bfloat16,
+#     use_safetensors=True,
+#     token=HF_TOKEN
+# )
 # Build pipeline
 controlnet = FluxControlNetModel.from_pretrained(
     "LPX55/FLUX.1-Inpainting-8step_uncensored", 
@@ -47,7 +59,7 @@ pipe = FluxControlNetInpaintingPipeline.from_pretrained(
     # device_map="balanced",
     token=HF_TOKEN
 )
-pipe.text_encoder_2 = text_encoder_8bit
+pipe.text_encoder_2 = text_encoder_2_4bit
 
 # pipe.load_lora_weights(
 #     hf_hub_download("ByteDance/Hyper-SD", "Hyper-FLUX.1-dev-8steps-lora.safetensors"), adapter_name="hyper-sd"