initial commit

app.py

@@ -0,0 +1,52 @@
+import streamlit as st
+from transformers import PegasusForConditionalGeneration, PegasusTokenizer
+
+st.title("Paraphrase sentences")
+
+model_name = "tuner007/pegasus_paraphrase"
+torch_device = "cpu"
+tokenizer = PegasusTokenizer.from_pretrained(model_name)
+
+
+@st.cache(allow_output_mutation=True)
+def load_model():
+    model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)
+    return model
+
+
+def get_response(
+    input_text, num_return_sequences, num_beams, max_length=60, temperature=1.5
+):
+    model = load_model()
+    batch = tokenizer(
+        [input_text],
+        truncation=True,
+        padding="longest",
+        max_length=max_length,
+        return_tensors="pt",
+    ).to(torch_device)
+    translated = model.generate(
+        **batch,
+        max_length=max_length,
+        num_beams=num_beams,
+        num_return_sequences=num_return_sequences,
+        temperature=temperature
+    )
+    tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
+    return tgt_text
+
+
+num_beams = 10
+num_return_sequences = st.slider("Number of paraphrases", 1, 10, 5, 1)
+context = st.text_area(label="Enter a sentence to paraphrase", max_chars=384)
+
+with st.expander("Advanced"):
+    temperature = st.slider("Temperature", 0.1, 5.0, 1.5, 0.1)
+    max_length = st.slider("Max length", 10, 100, 60, 10)
+if context:
+    response = get_response(
+        context, num_return_sequences, num_beams, max_length, temperature
+    )
+
+    for paraphrase in response:
+        st.write(paraphrase)

requirements.txt

@@ -0,0 +1 @@
+transformers==4.12.5