Add lookup tables to NLU model

.gitattributes

@@ -41,3 +41,4 @@ woz_nlu_agent/models/nlu/checkpoint filter=lfs diff=lfs merge=lfs -text
 woz_nlu_agent/models/nlu/component_2_LexicalSyntacticFeaturizer.feature_to_idx_dict.pkl filter=lfs diff=lfs merge=lfs -text
 woz_nlu_agent/models/nlu/component_3_CountVectorsFeaturizer.pkl filter=lfs diff=lfs merge=lfs -text
 woz_nlu_agent/models/nlu/** filter=lfs diff=lfs merge=lfs -text
+woz_nlu_agent/models/nlu-lookup-1/** filter=lfs diff=lfs merge=lfs -text

app.py

@@ -10,12 +10,13 @@ from transformers import AutoTokenizer, AutoModelForSequenceClassification
 
 
 # Rasa intent + entity extractor
-RASA_MODEL_PATH = "woz_nlu_agent/models/nlu"
-interpreter = None
+RASA_MODEL_PATH = "woz_nlu_agent/models/nlu-lookup-1"
+interpreter = Interpreter.load(RASA_MODEL_PATH)
 
 # OOS classifier
-tokenizer = AutoTokenizer.from_pretrained("msamogh/autonlp-cai-out-of-scope-649919116")
-model = AutoModelForSequenceClassification.from_pretrained("msamogh/autonlp-cai-out-of-scope-649919116")
+OOS_MODEL= "msamogh/autonlp-cai-out-of-scope-649919116"
+tokenizer = AutoTokenizer.from_pretrained(OOS_MODEL)
+model = AutoModelForSequenceClassification.from_pretrained(OOS_MODEL)
 
 
 MODEL_TYPES = {
@@ -30,7 +31,7 @@ def predict(model_type, input):
     elif MODEL_TYPES[model_type] == "oos":
         return oos_predict(input)
     elif MODEL_TYPES[model_type] == "intent_transformer":
-        return "TODO:: intent_transformer"
+        return "WIP: intent_transformer"
 
     
 def oos_predict(input):
@@ -67,15 +68,6 @@ def rasa_predict(input):
 
 
 def main():
-    global interpreter
-    print("Loading model...")
-    print(os.listdir("woz_nlu_agent/models/nlu"))
-    print(open("woz_nlu_agent/models/nlu/metadata.json", "r").read())
-    import json
-    print(json.load(open("woz_nlu_agent/models/nlu/metadata.json", "r")))
-    
-    interpreter = Interpreter.load(RASA_MODEL_PATH)
-    print("Model loaded.")
     iface = gr.Interface(fn=predict, inputs=[gr.inputs.Dropdown(list(MODEL_TYPES.keys())), "text"], outputs="text")
     iface.launch()