loads n closet via api endpoint

app.py

@@ -79,30 +79,26 @@ def main(
     seed=None
     ):
 
-    if seed == None:
-        seed = np.random.randint(2147483647)
-    # if device contains cuda
-    if device.type == 'cuda':
-        generator = torch.Generator(device=device).manual_seed(int(seed))
-    else:
-        generator = torch.Generator().manual_seed(int(seed)) # use cpu as does not work on mps
-
     embeddings = base64_to_embedding(embeddings)
-    embeddings = torch.tensor(embeddings, dtype=torch_size).to(device)
-
-    images_list = pipe(
-        # inp.tile(n_samples, 1, 1, 1),
-        # [embeddings * n_samples],
-        embeddings,
-        guidance_scale=scale,
-        num_inference_steps=steps,
-        generator=generator,
-        )
-
+    # convert to python array
+    embeddings = embeddings.tolist()
+    results = clip_retrieval_client.query(embedding_input=embeddings)
     images = []
-    for i, image in enumerate(images_list["images"]):
-        images.append(image)
-    # images.append(embedding_image)
+    for result in results:
+        if len(images) >= n_samples:
+            break
+        # dowload image
+        import requests
+        from io import BytesIO
+        response = requests.get(result["url"])
+        if not response.ok:
+            continue
+        try:
+            bytes = BytesIO(response.content)
+            image = Image.open(bytes)
+            images.append(image)
+        except Exception as e:
+            print(e)
     return images
 
 def on_image_load_update_embeddings(image_data):
@@ -146,10 +142,10 @@ def update_average_embeddings(embedding_base64s_state, embedding_powers):
         return gr.Text.update('')
 
     # TODO toggle this to support average or sum
-    final_embedding = final_embedding / num_embeddings
+    # final_embedding = final_embedding / num_embeddings
     
     # normalize embeddings in numpy
-    final_embedding /= np.linalg.norm(final_embedding)
+    # final_embedding /= np.linalg.norm(final_embedding)
     
     embeddings_b64 = embedding_to_base64(final_embedding)
     return embeddings_b64
@@ -368,12 +364,12 @@ Try uploading a few images and/or add some text prompts and click generate image
         with gr.Accordion(f"Avergage embeddings in base 64", open=False):
             average_embedding_base64 = gr.Textbox(show_label=False)
     with gr.Row():
-        submit = gr.Button("Generate images")
+        submit = gr.Button("Search embedding space")
     with gr.Row():
         with gr.Column(scale=1, min_width=200):
             scale = gr.Slider(0, 25, value=3, step=1, label="Guidance scale")
         with gr.Column(scale=1, min_width=200):
-            n_samples = gr.Slider(1, 4, value=1, step=1, label="Number images")
+            n_samples = gr.Slider(1, 16, value=4, step=1, label="Number images")
         with gr.Column(scale=1, min_width=200):
             steps = gr.Slider(5, 50, value=25, step=5, label="Steps")
         with gr.Column(scale=1, min_width=200):