Update raw.py

raw.py

@@ -17,15 +17,20 @@ import gradio as gr
 huggingface_token = os.getenv("HUGGINFACE_TOKEN")
 MAX_SEED = 1000000
 
-quant_config = TransformersBitsAndBytesConfig(load_in_8bit=True,)
-text_encoder_2_8bit = T5EncoderModel.from_pretrained(
+# quant_config = TransformersBitsAndBytesConfig(load_in_8bit=True,)
+# text_encoder_2_8bit = T5EncoderModel.from_pretrained(
+#     "LPX55/FLUX.1-merged_uncensored",
+#     subfolder="text_encoder_2",
+#     quantization_config=quant_config,
+#     torch_dtype=torch.bfloat16,
+#     token=huggingface_token
+# )
+text_encoder_2_unquant = T5EncoderModel.from_pretrained(
     "LPX55/FLUX.1-merged_uncensored",
     subfolder="text_encoder_2",
-    quantization_config=quant_config,
     torch_dtype=torch.bfloat16,
     token=huggingface_token
 )
-
 # good_vae = AutoencoderKL.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="vae", torch_dtype=torch.bfloat16, token=huggingface_token).to("cuda")
 
 # Load pipeline
@@ -36,7 +41,7 @@ text_encoder_2_8bit = T5EncoderModel.from_pretrained(
 pipe = FluxControlNetPipeline.from_pretrained(
     "LPX55/FLUX.1M-8step_upscaler-cnet",
     torch_dtype=torch.bfloat16,
-    text_encoder_2=text_encoder_2_8bit,
+    text_encoder_2=text_encoder_2_unquant,
     token=huggingface_token
 )
 # adapter_id = "alimama-creative/FLUX.1-Turbo-Alpha"
@@ -45,14 +50,14 @@ pipe = FluxControlNetPipeline.from_pretrained(
 
 pipe.to("cuda")
 
-try:
-    pipe.vae.enable_slicing()
-except:
-    print("debug-2")
-try:
-    pipe.vae.enable_tiling()
-except:
-    print("debug-3")
+# try:
+#     pipe.vae.enable_slicing()
+# except:
+#     print("debug-2")
+# try:
+#     pipe.vae.enable_tiling()
+# except:
+#     print("debug-3")
 
 # pipe.load_lora_weights(adapter_id, adapter_name="turbo")
 # pipe.load_lora_weights(adapter_id2, adapter_name="real")