init

app.py

@@ -0,0 +1,57 @@
+import streamlit as st
+import torch
+from transformers import AutoModelForQuestionAnswering, AutoTokenizer
+
+
+@st.cache(allow_output_mutation=True)
+def get_model():
+    # Load fine-tuned MRC model by HuggingFace Model Hub
+    HUGGINGFACE_MODEL_PATH = "bespin-global/klue-bert-base-aihub-mrc"
+    tokenizer = AutoTokenizer.from_pretrained(HUGGINGFACE_MODEL_PATH )
+    model = AutoModelForQuestionAnswering.from_pretrained(HUGGINGFACE_MODEL_PATH )
+
+    return tokenizer, model
+
+tokenizer, model = get_model()
+
+## Title
+st.title('BespinGlobal - Machine Reading Comprehension', anchor='https://huggingface.co/bespin-global/klue-bert-base-aihub-mrc')
+
+## Text
+st.text('bespin-global/klue-bert-base-aihub-mrc 모델 성능 테스트 페이지 입니다.')
+
+# Text Input
+context = st.text_area("📑 Context HERE!", placeholder="Please input some context..", height=300, on_change=None)
+
+# Text Area
+question = st.text_area("💡 Question HERE!", placeholder="Please input your question..")
+if st.button("Submit", key='question'):
+    try:
+        # Progress spinner
+        with st.spinner('Wait for it...'):
+            # Encoding
+            encodings = tokenizer(context, question, 
+                              max_length=512, 
+                              truncation=True,
+                              padding="max_length", 
+                              return_token_type_ids=False
+                              )
+            encodings = {key: torch.tensor([val]) for key, val in encodings.items()}             
+            input_ids = encodings["input_ids"]
+            attention_mask = encodings["attention_mask"]
+
+            # Predict
+            pred = model(input_ids, attention_mask=attention_mask)
+
+            start_logits, end_logits = pred.start_logits, pred.end_logits
+            token_start_index, token_end_index = start_logits.argmax(dim=-1), end_logits.argmax(dim=-1)
+            pred_ids = input_ids[0][token_start_index: token_end_index + 1]
+
+            # Decoding
+            prediction = tokenizer.decode(pred_ids)
+
+            # answer
+            st.success(prediction)
+
+    except Exception as e:
+        st.error(e)

requirements.txt

@@ -0,0 +1,3 @@
+torch
+transformers
+streamlit