Dynamic Model Loading

src/conversation_terminator/remote/model.py

@@ -14,6 +14,10 @@ def __new__(cls, context):
         return cls.instance
 
     async def inference(self, request: ModelRequest):
+        if request.model != 'NA':
+            model_name = str(request.model)
+            self.tokenizer = BertTokenizer.from_pretrained(model_name)
+            self.model = TFBertForSequenceClassification.from_pretrained(model_name)
         inputs = self.tokenizer(request.text,return_tensors="np", padding=True)
         outputs = self.model(inputs.input_ids, inputs.attention_mask)
         probabilities = tf.nn.sigmoid(outputs.logits)

src/conversation_terminator/remote/request.py

@@ -2,8 +2,9 @@
 
 
 class ModelRequest():
-    def __init__(self, text):
+    def __init__(self, text, model='NA'):
         self.text = text
+        self.model = model
 
     def to_json(self):
         return json.dumps(self, default=lambda o: o.__dict__,

src/text_classification/grievance_recognition/local/model.py

@@ -16,6 +16,11 @@ def __new__(cls, context):
 
 
     async def inference(self,  request: ModelRequest):
+        if request.model != 'NA':
+            model_name = request.model
+            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+            self.model = AutoModelForSequenceClassification.from_pretrained(model_name)
+
         inputs = self.tokenizer(request.text, return_tensors="pt")
         inputs = {key: value.to(self.device) for key, value in inputs.items()}
         with torch.no_grad():

src/text_classification/grievance_recognition/local/request.py

@@ -3,8 +3,9 @@
 
 
 class ModelRequest():
-    def __init__(self, text):
+    def __init__(self, text, model='NA'):
         self.text = text
+        self.model = model
 
     def to_json(self):
         return json.dumps(self, default=lambda o: o.__dict__,