Inital commit of feedback demo

.gitignore

@@ -0,0 +1,2 @@
+.venv/*
+*.pyc

app.py

@@ -0,0 +1,36 @@
+import streamlit as st
+import streamlit.components.v1 as components
+
+
+from src.utils import summarize_en
+
+
+st.write("## 👥 Getting User Feedback with Tally forms")
+
+
+st.write("It's one thing to build performant models, it's another thing to prototype and test their performance in front of real users. By mixing Streamlit & Tally, this has never been easier.")
+
+st.write("The Streamlit app below uses google/pegasus-xsum to summarize an article. Run a summary, and share your feedback. 💬")
+
+placeholder_text = """Contemporary climate change includes both global warming and its impacts on Earth's weather patterns. There have been previous periods of climate change, but the current changes are distinctly more rapid and not due to natural causes. Instead, they are caused by the emission of greenhouse gases, mostly carbon dioxide (CO2) and methane. Burning fossil fuels for energy use creates most of these emissions. Certain agricultural practices, industrial processes, and forest loss are additional sources. Greenhouse gases are transparent to sunlight, allowing it through to heat the Earth's surface. When the Earth emits that heat as infrared radiation the gases absorb it, trapping the heat near the Earth's surface. As the planet heats up it causes changes like the loss of sunlight-reflecting snow cover, amplifying global warming. From: https://en.wikipedia.org/wiki/Climate_change"""
+
+feedback_url = """https://tally.so/embed/w2EX7A?alignLeft=1&hideTitle=1&transparentBackground=1"""
+ 
+with st.form("my_form"):
+    st.write("#### Your long text goes here 🇬🇧↘️")
+    txt_input = st.text_area("", placeholder_text,height=300)
+
+    # Every form must have a submit button.
+    submitted = st.form_submit_button("🤖 Summarize ")
+    if submitted:
+        with st.spinner(text="Summarizing text, this could take 10 seconds"):
+            
+            
+            summary = summarize_en(txt_input)
+        st.markdown("### Summary by Pegasus")
+        st.markdown(">{}".format(summary))
+        # embed streamlit docs in a streamlit app
+        components.iframe(feedback_url, height=800, scrolling=True)
+
+# st.write("Outside the form")
+

makefile

@@ -0,0 +1,7 @@
+setup:
+	( \
+	   python -m venv .venv; \
+       . .venv/bin/activate; \
+	   pip install --upgrade pip; \
+       pip install -r requirements.txt; \
+    )

requirements.txt

@@ -0,0 +1,3 @@
+transformers[torch]
+streamlit
+sentencepiece

src/utils.py

@@ -0,0 +1,98 @@
+"""
+Shared utility methods for this module.
+"""
+
+from ctypes import Array
+import datetime
+import re
+
+from transformers import AutoModelForSequenceClassification, AutoTokenizer, MBartForConditionalGeneration, MBartTokenizer, pipeline
+from transformers import PegasusTokenizer, PegasusForConditionalGeneration
+
+
+
+def lowercase_string(string: str) -> str:
+    """Returns a lowercased string
+    Args:
+        string: String to lowercase
+    Returns:
+        String in lowercase
+    """
+    if isinstance(string, str):
+        return string.lower()
+    return None
+
+
+from functools import lru_cache
+
+@lru_cache
+def get_sentiment_pipeline():
+    model_name = "nlptown/bert-base-multilingual-uncased-sentiment"
+    model = AutoModelForSequenceClassification.from_pretrained(model_name)
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    sentiment_pipeline = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)
+    return sentiment_pipeline
+
+def score_sentiment(input: str):
+    """Score sentiment of an input string with a pretrained Transformers Pipeline
+
+    Args:
+        input (str): Text to be scored
+
+    Returns:
+        tuple: (label, score)
+    """
+    sentiment_pipeline = get_sentiment_pipeline()
+    result = sentiment_pipeline(input.lower())[0]
+    # print("label:{0} input:{1}".format(result['label'], input))
+    return result['label'], result['score']
+
+
+@lru_cache
+def get_summarization_pipeline_nl():
+
+    undisputed_best_model = MBartForConditionalGeneration.from_pretrained(
+        "ml6team/mbart-large-cc25-cnn-dailymail-nl"
+    )
+    tokenizer = MBartTokenizer.from_pretrained("facebook/mbart-large-cc25")
+    summarization_pipeline = pipeline(
+        task="summarization",
+        model=undisputed_best_model,
+        tokenizer=tokenizer,
+    )
+    summarization_pipeline.model.config.decoder_start_token_id = tokenizer.lang_code_to_id[
+        "nl_XX"
+    ]
+    return summarization_pipeline
+
+def summarize_nl(input: str) -> str:
+    summarization_pipeline = get_summarization_pipeline_nl()
+    summary = summarization_pipeline(
+        input,
+        do_sample=True,
+        top_p=0.75,
+        top_k=50,
+        # num_beams=4,
+        min_length=50,
+        early_stopping=True,
+        truncation=True,
+    )[0]["summary_text"]
+    return summary
+
+
+
+@lru_cache
+def get_pegasus():
+    model = PegasusForConditionalGeneration.from_pretrained("google/pegasus-xsum")
+    tokenizer = PegasusTokenizer.from_pretrained("google/pegasus-xsum")
+    return model, tokenizer
+
+def summarize_en(input: str) -> str:
+
+    model, tokenizer = get_pegasus()
+    inputs = tokenizer(input, max_length=1024, return_tensors="pt")
+
+    # Generate Summary
+    summary_ids = model.generate(inputs["input_ids"])
+    result = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] 
+    return result