upd: load types

api_server.py

@@ -13,13 +13,16 @@ os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
 from tensorflow import keras
 from flask import Flask, jsonify, request
 
-load_type = 'remote_hub_from_pretrained'
+load_type = ''
 """
-local,
-remote_hub_download - /cache error even using TRANSFORMERS_CACHE to root folder
-remote_hub_from_pretrained
+local;
+remote_hub_download; - /cache error even using TRANSFORMERS_CACHE & cache_dir to local folder
+remote_hub_from_pretrained; - /cache error even using TRANSFORMERS_CACHE & cache_dir to local folder
+remote_hub_pipeline; - needs config.json and this is not easy to grasp how to do it with custom models
+https://discuss.huggingface.co/t/how-to-create-a-config-json-after-saving-a-model/10459/4
 """
 
+
 REPO_ID = "1vash/mnist_demo_model"
 
 # Load the saved model into memory
@@ -29,9 +32,14 @@ elif load_type == 'remote_hub_download':
     from huggingface_hub import hf_hub_download
     model = keras.models.load_model(hf_hub_download(repo_id=REPO_ID, filename="saved_model.pb"))
 elif load_type == 'remote_hub_from_pretrained':
+    # https://huggingface.co/docs/hub/keras
     from huggingface_hub import from_pretrained_keras
     model = from_pretrained_keras(REPO_ID, cache_dir='./artifacts/')
-
+elif load_type == 'remote_hub_pipeline':
+    from transformers import pipeline
+    classifier = pipeline("image-classification", model=REPO_ID)
+else:
+    pass
 
 # Initialize the Flask application
 app = Flask(__name__)