Update app.py

app.py

@@ -9,6 +9,7 @@ import numpy as np
 import faiss
 import csv
 import datetime
+import joblib
 
 from huggingface_hub import hf_hub_download
 encoder_text_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename=os.environ['ENCODER_TEXT'],
@@ -30,7 +31,31 @@ sys.path.append(os.environ['PRIVATE_DIR'])
 from models import *
 
 preprocess_model, model = get_models()
-    
+
+index_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename=os.environ['INDEX'],
+                            use_auth_token=os.environ['TOKEN'])
+indexnames_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename=os.environ['INDEX_NAMES'],
+                            use_auth_token=os.environ['TOKEN'])
+catalog_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename=os.environ['CATALOG'],
+                            use_auth_token=os.environ['TOKEN'])
+
+#url
+url_dict={}
+with open(catalog_path) as csv_file:
+    csv_reader = csv.reader(csv_file, delimiter=';')
+    for row in csv_reader:
+      f = row[2].split('/')[-1]
+      url_dict[f.split('/')[-1][:-4]] = row[2]
+
+# names index
+audio_names = joblib.load(open(indexnames_path, 'rb'))
+
+# load embed audio catalog
+index = faiss.read_index(index_path)
+
+encoder_text = tf.keras.models.load_model(encoder_text_path)
+
+
 def process(prompt, lang):
 
     now = datetime.datetime.now()
@@ -46,30 +71,13 @@ def process(prompt, lang):
     print("   text representation computed.")
 
     # Embed text
-    #from models import *
-    encoder_text = tf.keras.models.load_model(encoder_text_path)
     embed_query  = encoder_text.predict(embed_prompt["pooled_output"])
     faiss.normalize_L2(embed_query)
     print("   text embed computed.")
 
-    # load embed audio catalog
-    index = faiss.read_index("BMG_221022.index")
-
     # distance computing
     D, I = index.search(embed_query, TOP)
 
-    # names index
-    import joblib
-    audio_names = joblib.load(open('BMG_221022_names.index', 'rb'))
-
-    #url
-    url_dict={}
-    with open("bmg_clean.csv") as csv_file:
-        csv_reader = csv.reader(csv_file, delimiter=';')
-        for row in csv_reader:
-          f = row[2].split('/')[-1]
-          url_dict[f.split('/')[-1][:-4]] = row[2]
-
     # output : top N audio file names
     print(I)
     print(D)
@@ -78,7 +86,6 @@ def process(prompt, lang):
         print(audio_names[I[0][i]], " with distance ", D[0][i])
         print("    url : ", url_dict[audio_names[I[0][i]]])
 
-
     return [url_dict[audio_names[I[0][0]]], url_dict[audio_names[I[0][1]]], url_dict[audio_names[I[0][2]]], url_dict[audio_names[I[0][3]]], url_dict[audio_names[I[0][4]]]]
 
 inputs = [gr.Textbox(label="Input", value="type your description", max_lines=2),