maybe fixed :)

app.py

@@ -6,8 +6,7 @@ from sklearn.svm import LinearSVC
 from sklearn import preprocessing
 import pandas as pd
 
-from transformers import CLIPVisionModelWithProjection
-from diffusers import LCMScheduler, AutoencoderTiny, EulerDiscreteScheduler, UNet2DConditionModel, StableDiffusionXLPipeline
+from diffusers import LCMScheduler, AutoencoderTiny, EulerDiscreteScheduler, UNet2DConditionModel
 from diffusers.models import ImageProjection
 from patch_sdxl import SDEmb
 import torch
@@ -40,16 +39,10 @@ unet.load_state_dict(load_file(hf_hub_download(sdxl_lightening, ckpt), device="c
 pipe = SDEmb.from_pretrained(model_id, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
 pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taesdxl", torch_dtype=torch.float16)
 pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
-
-image_preproc = CLIPImageProcessor.from_pretrained('runwayml/stable-diffusion-v1-5', subfolder='feature_extractor')
-image_encoder = CLIPVisionModelWithProjection.from_pretrained('h94/IP-Adapter', subfolder='sdxl_models/image_encoder', torch_dtype=torch.float16)
-pipe.unet._load_ip_adapter_weights(torch.load(hf_hub_download('h94/IP-Adapter', 'sdxl_models/ip-adapter_sdxl.bin'), map_location="cpu"))
-pipe.register_modules(image_encoder = image_encoder)
-pipe.register_modules(feature_extractor = image_preproc)
-
 pipe.to(device='cuda')
+pipe.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter_sdxl.bin")
 
-output_hidden_state = True
+output_hidden_state = False
 #######################
 
 @spaces.GPU
@@ -64,7 +57,7 @@ def predict(
             im_emb = torch.zeros(1, 1280, dtype=torch.float16, device='cuda')
         image = pipe(
             prompt=prompt,
-            ip_adapter_emb=[im_emb.to('cuda')],
+            ip_adapter_emb=im_emb.to('cuda'),
             height=1024,
             width=1024,                                                                                                             
             num_inference_steps=2,
@@ -124,7 +117,7 @@ def next_image(embs, ys, calibrate_prompts):
 
             w = 1# if len(embs) % 2 == 0 else 0
             im_emb = w * lin_class.coef_.to(device=DEVICE, dtype=torch.float16)
-            prompt= '' if glob_idx % 2 == 0 else rng_prompt
+            prompt= 'an image' if glob_idx % 2 == 0 else rng_prompt
             print(prompt)
             image, im_emb = predict(prompt, im_emb)
             embs.append(im_emb)
@@ -153,9 +146,9 @@ def start(_, embs, ys, calibrate_prompts):
 
 
 def choose(choice, embs, ys, calibrate_prompts):
-    if choice == 'Like':
+    if choice == 'Like (L)':
         choice = 1
-    elif choice == 'Neither':
+    elif choice == 'Neither (Space)':
         _ = embs.pop(-1)
         img, embs, ys, calibrate_prompts = next_image(embs, ys, calibrate_prompts)
         return img, embs, ys, calibrate_prompts