subiendo la demo a hf

app.py

@@ -5,6 +5,7 @@ from multiprocessing.dummy import Pool
 import base64
 from PIL import Image, ImageOps
 import torch
+import numpy as np
 from torchvision import transforms
 from streamlit_drawable_canvas import st_canvas
 from src.model_LN_prompt import Model
@@ -83,8 +84,9 @@ def compute_sketch(_sketch, model):
 def image_search(_query, corpus, model, embeddings, n_results=N_RESULTS):
     query_embedding = compute_sketch(_query, model)
     corpus_id = 0 if corpus == "Unsplash" else 1
-    image_features = torch.tensor(
-        list([item[0] for item in embeddings[corpus_id]])).to(device)
+    image_features = torch.from_numpy(
+        np.array([item[0] for item in embeddings[corpus_id]])
+    ).to(device)
 
     dot_product = (image_features @ query_embedding.T)[:, 0]
     _, max_indices = torch.topk(

src/model_LN_prompt.py

@@ -32,14 +32,11 @@ class Model(pl.LightningModule):
 
 
     def configure_optimizers(self):
-        if self.opts.model_type == 'one_encoder':
-            model_params = list(self.dino.parameters())
-        else:
-            model_params = list(self.dino.parameters()) + list(self.clip_sk.parameters())
+        model_params = list(self.dino.parameters())
 
         optimizer = torch.optim.Adam([
-            {'params': model_params, 'lr': self.opts.clip_LN_lr},
-            {'params': [self.sk_prompt] + [self.img_prompt], 'lr': self.opts.prompt_lr}])
+            {'params': model_params, 'lr': self.opts.clip_LN_lr}]
+        )
         return optimizer
 
     def forward(self, data, dtype='image'):