added deployer config

.github/workflows/deploy.yml

@@ -0,0 +1,22 @@
+# Only modify variables that have a comment above them
+# Contact IWG if you wish to alter the template otherwise
+
+name: Deploy
+on:
+  push:
+    branches: main
+  pull_request:
+      branches: ['*']
+
+jobs:
+  deploy:
+    name: Environments
+    uses: arg-tech/deployment-templates/.github/workflows/default-deploy-template.yml@main
+    secrets: inherit
+    with: 
+      # Specify the target production server
+      target_production_server_nickname: argand
+      # Define a URL for your app, without the http:// or www prefixes
+      full_app_url: vanilla-am-caasr.amfws.arg.tech
+      # The port that is exposed on localhost (must be the same as in docker-compose.yml)
+      app_port: 5015

Dockerfile

@@ -10,13 +10,14 @@ RUN pip3 install torch
 RUN pip3 install numpy
 RUN pip3 install transformers
 RUN pip3 install Cython
-RUN pip3 install xaif_eval
+RUN pip3 install xaif_eval==0.0.9
 RUN pip3  install scikit-learn 
-RUN pip3 install amf-fast-inference
+#RUN pip3 install amf-fast-inference
+RUN pip3 install amf-fast-inference==0.0.3
 
 
 COPY . /app
 WORKDIR /app
 RUN pip install -r requirements.txt
-EXPOSE 5007
+EXPOSE 5015
 CMD python ./main.py

docker-compose.yml

@@ -4,5 +4,5 @@ services:
   dam:
     build: .
     ports:
-      - "5007:5007"
+      - "5015:5015"
 

main.py

@@ -1,7 +1,7 @@
 
 
 from src.caasr import CAASRArgumentStructure
-from transformers import GPT2Tokenizer,pipeline
+from transformers import GPT2Tokenizer,pipeline, AutoModelForSequenceClassification
 
 from amf_fast_inference import model
 
@@ -12,12 +12,12 @@
 
 logging.basicConfig(datefmt='%H:%M:%S', level=logging.DEBUG)
 
-
 app = Flask(__name__)
 metrics = PrometheusMetrics(app)
 model_name = "debela-arg/dialogtp-am-medium"
 loader = model.ModelLoader(model_name)
-model = loader.load_model()        
+model = loader.load_model()   
+#model =     AutoModelForSequenceClassification.from_pretrained(model_name)
 tokenizer = GPT2Tokenizer.from_pretrained(model_name)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model.to(device)
@@ -47,5 +47,5 @@ def caasra():
 		return info	
 
 if __name__ == "__main__":
-    app.run(host="0.0.0.0", port=int("5007"), debug=False)	  
+    app.run(host="0.0.0.0", port=int("5015"), debug=False)	  
 

src/caasr.py

@@ -3,7 +3,7 @@
 import pandas as pd
 from itertools import combinations
 import torch.nn.functional as F
-from xaif_eval import xaif
+from xaif_eval import xaif 
 import json
 from src.custom_model import MicroAndMacroContextPT2ForTokenClassification
 from src.data_utils import prepare_inputs
@@ -12,7 +12,7 @@
 logging.basicConfig(level=logging.INFO)
 
 from transformers import pipeline
-from amf_fast_inference import model
+#from amf_fast_inference import model
 
 
 
@@ -119,26 +119,31 @@ def get_argument_structure(self,):
         predicted_relations = []
         propositions = []
         for (p1, p2), relation in zip(combined_texts,predictions):
-            predicted_relations.append((p1,p2,relation))
-            if p1 not in propositions:
-                propositions.append(p1)
-            if p2 not in propositions:
-                propositions.append(p2)
+            if relation in ["CA", "RA",'MA']:
+                predicted_relations.append((p1,p2,relation))
+                if p1 not in propositions:
+                    propositions.append(p1)
+                if p2 not in propositions:
+                    propositions.append(p2)
 
         generator = ArgumentStructureGenerator()
         refined_structure = generator.generate_argument_structure_from_relations(propositions, predicted_relations)
         logging.info(refined_structure)
         relation_encoder = {0: "None", 1: "RA", 2: "CA", 3: "MA"}
         for conclussion_id, premise_relation_list in refined_structure.items():
             #print(node_id_prpos[conclussion_id], node_id_prpos[premise_relation_list[0]], relation_encoder[premise_relation_list[1]])
-            premise_id,AR_type  = premise_relation_list[0], premise_relation_list[1]
-            logging.info(AR_type)
-            if AR_type in ['CA', 'RA']:
+            #premise_id,AR_type  = premise_relation_list[0], premise_relation_list[1]
+            premises, relations = premise_relation_list[:len(premise_relation_list)//2], premise_relation_list[len(premise_relation_list)//2:]
+            for premise_id,AR_type in zip (premises, relations):
                 logging.info(AR_type)
-
-                aif.add_component("argument_relation", AR_type, conclussion_id, premise_id)
-
-        return(aif.aif)
+                if AR_type=="MA":
+                    AR_type = "RA"
+                if AR_type in ['CA', 'RA',"MA"]:
+                    logging.info(AR_type)
+
+                    aif.add_component("argument_relation", AR_type, conclussion_id, premise_id)
+
+        return(aif.xaif)
 
 '''
 if __name__ == "__main__":

src/data_utils.py

@@ -45,7 +45,7 @@ def tokenize_and_concat(propo1, propso2, argument, tokenizer):
 
 
 def prepare_inputs(data, tokenizer):
-    # Implementation of prepare_inputs
+    # Implementation of prepare_inputs 
     tokenized_results = []
     position_embeddings = []
     micro_labels =  []
@@ -54,9 +54,11 @@ def prepare_inputs(data, tokenizer):
     prop_1_texts = data['prop_1']
     prop_2_texts = data['prop_2']
     for prop_1, prop_2, argument in zip(prop_1_texts, prop_2_texts, arguments):
-        tokenized_input,propos = tokenize_and_concat(prop_1, prop_2, argument,tokenizer)
-        tokenized_results.append(tokenized_input) 
-        input_data.append(propos)
+
+        if  (prop_1 not in ["Good evening both", "Rishi Sunak","Thank you, Rishi Sunak", "THANK YOU", "Rishi Sunak"]) and (prop_2 not in ["Good evening both", "Rishi Sunak","Thank you, Rishi Sunak", "THANK YOU", "Rishi Sunak"]):
+            tokenized_input,propos = tokenize_and_concat(prop_1, prop_2, argument,tokenizer)
+            tokenized_results.append(tokenized_input) 
+            input_data.append(propos)
     return tokenized_results, input_data