Update app.py

app.py

@@ -11,7 +11,7 @@ import csv
 import datetime
 
 from huggingface_hub import hf_hub_download
-encoder_text_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename="encoder_text_retrievaltext_bmg_221022_54.h5",
+encoder_text_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename=os.environ['ENCODER_TXT'],
                             use_auth_token=os.environ['TOKEN'])
 print("DEBUG ", encoder_text_path)                                
 # NO GPU
@@ -19,58 +19,23 @@ os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
 
 
-
-
 python_path = hf_hub_download(repo_id="PierreHanna/TextRetrieval", repo_type="space", filename="models.py",
-                            use_auth_token=os.environ['TOKEN'], cache_dir="./ph")
+                            use_auth_token=os.environ['TOKEN'])
+
 print(python_path)
-os.system('ls -la')
+#os.system('ls -la')
 #from models import *
 
-sys.path.append("ph/spaces--PierreHanna--TextRetrieval/snapshots/ee43bbe093de2cd1b2fbda7c04d00ed4d360d730/")
-
-#import site
-#site.addsitedir('./ph.spaces--PierreHanna--TextRetrieval.snapshots.ee43bbe093de2cd1b2fbda7c04d00ed4d360d730')
-#from ph.spaces--PierreHanna--TextRetrieval.snapshots.ee43bbe093de2cd1b2fbda7c04d00ed4d360d730.models import *
+sys.path.append(os.environ['PRIVATE_DIR'])
 from models import *
 
-def make_preprocess_model(sentence_features, tfhub_handle_preprocess, seq_length=128):
-  """Returns Model mapping string features to BERT inputs.
-  """
-
-  input_segments = [
-      tf.keras.layers.Input(shape=(), dtype=tf.string, name=ft)
-      for ft in sentence_features]
-
-  bert_preprocess = hub.load(tfhub_handle_preprocess)
-  tokenizer = hub.KerasLayer(bert_preprocess.tokenize, name='tokenizer')
-  segments = [tokenizer(s) for s in input_segments]
-
-  truncated_segments = segments
-  packer = hub.KerasLayer(bert_preprocess.bert_pack_inputs,
-                          arguments=dict(seq_length=seq_length),
-                          name='packer')
-  model_inputs = packer(truncated_segments)
-  return tf.keras.Model(input_segments, model_inputs)
-
 def process(prompt, lang):
 
     # Getting prompt user
     #prompt = input("Audio Search - enter text : ")
     #print(prompt)
-
-    # prompt embedding
-    bert_model_name = 'small_bert/bert_en_uncased_L-4_H-512_A-8' 
-    tfhub_handle_encoder = 'https://tfhub.dev/tensorflow/small_bert/bert_en_uncased_L-4_H-512_A-8/1'
-    tfhub_handle_preprocess = 'https://tfhub.dev/tensorflow/bert_en_uncased_preprocess/3'
-
-    MAX_LENGTH = 130 # MAX de 512 !!! TENSORFLOW !!!
-    TOP = 10
-
-
-    bert_preprocess_model = make_preprocess_model(['my_input'], tfhub_handle_preprocess, seq_length = MAX_LENGTH)
-    bert_model = hub.KerasLayer(tfhub_handle_encoder)
-
+    preprocess_model, model = get_models()
+    
     now = datetime.datetime.now()
     print()
     print('*************')
@@ -79,8 +44,8 @@ def process(prompt, lang):
     print('*************')
     print()
     prompt=[prompt]
-    text_preprocessed = bert_preprocess_model([np.array(prompt)])
-    embed_prompt = bert_model(text_preprocessed)
+    text_preprocessed = preprocess_model([np.array(prompt)])
+    embed_prompt = model(text_preprocessed)
     print("   text representation computed.")
 
     # Embed text