Add Image-Caption matching app

app.py

@@ -1,4 +1,39 @@
+import os
+import sys
+
 import streamlit as st
+from huggingface_hub import snapshot_download
+from transformers import AutoTokenizer
+
+
+LOCAL_PATH = snapshot_download("flax-community/clip-spanish")
+sys.path.append(LOCAL_PATH)
+
+from modeling_hybrid_clip import FlaxHybridCLIP
+from test_on_image import run_inference
+
+
+def save_file_to_disk(uplaoded_file):
+    temp_file = os.path.join("/tmp", uplaoded_file.name)
+    with open(temp_file,"wb") as f:
+        f.write(uploaded_file.getbuffer())
+    return temp_file
+
+# load the saved model
+tokenizer = AutoTokenizer.from_pretrained("dccuchile/bert-base-spanish-wwm-cased")
+model = FlaxHybridCLIP.from_pretrained(LOCAL_PATH)
+
+
+st.title("Image-Caption Matching")
+uploaded_file = st.file_uploader("Choose an image...", type="jpg")
+text_input = st.text_input("Type a caption")
 
-x = st.slider('Select a value')
-st.write(x, 'squared is', x * x)
+if uploaded_file is not None and text_input:
+    local_image_path = None
+    try:
+        local_image_path = save_file_to_disk(uploaded_file)
+        score = run_inference(local_image_path, text_input, model)[0][0].tolist()
+        st.write('%s (%.2f%%)' % score)
+    finally:
+        if local_image_path:
+            os.remove(local_image_path)