change threading options for onnx inference

app.py

@@ -87,6 +87,10 @@ hide_streamlit_style = """
             """
 st.markdown(hide_streamlit_style, unsafe_allow_html=True)
 
+options = ort.SessionOptions()
+options.intra_op_num_threads=1
+options.inter_op_num_threads=1
+
 
 @st.cache(allow_output_mutation=True, suppress_st_warning=True, max_entries=None, ttl=None)
 def create_model_dir(chkpt, model_dir):
@@ -180,6 +184,9 @@ if select_task=='README':
 if select_task == 'Detect Sentiment':
     t1=time.time()
     tokenizer_sentiment,sentiment_session = sentiment_task_selected(task=select_task)
+    ##below 2 steps are slower as caching is not enabled
+    # tokenizer_sentiment = AutoTokenizer.from_pretrained(sent_mdl_dir)
+    # sentiment_session = ort.InferenceSession(f"{sent_onnx_mdl_dir}/{sent_onnx_mdl_name}")
     t2 = time.time()
     st.write(f"Total time to load Model is {(t2-t1)*1000:.1f} ms")
 
@@ -210,7 +217,9 @@ if select_task == 'Detect Sentiment':
 
 if select_task=='Zero Shot Classification':
     t1=time.time()
-    tokenizer_zs,zs_session = zs_task_selected(task=select_task)
+    tokenizer_zs,session_zs = zs_task_selected(task=select_task)
+    # tokenizer_zs= AutoTokenizer.from_pretrained(zs_mdl_dir)
+    # session_zs = ort.InferenceSession(f"{zs_onnx_mdl_dir}/{zs_onnx_mdl_name}")
     t2 = time.time()
     st.write(f"Total time to load Model is {(t2-t1)*1000:.1f} ms")
 
@@ -225,7 +234,7 @@ if select_task=='Zero Shot Classification':
 
     if response1:
         start = time.time()
-        df_output = zero_shot_classification_onnx(premise=input_texts, labels=input_lables, _session=zs_session,
+        df_output = zero_shot_classification_onnx(premise=input_texts, labels=input_lables, _session=session_zs,
                                                _tokenizer=tokenizer_zs)
         end = time.time()
         st.write("")