fundemental rewrite to make it WAY MORE AWESOME

app.py

@@ -1,6 +1,5 @@
-from simplet5 import SimpleT5
 import streamlit as st
-import pytorch_lightning as pl
+import csv 
 
 
 st.set_page_config(page_title="Reassuring Parables")
@@ -11,8 +10,10 @@ st.caption("Find me on Linkedin: https://www.linkedin.com/in/allen-roush-2772101
 st.image("https://imgs.xkcd.com/comics/reassuring.png")
 st.caption("From https://xkcd.com/1263/")
 
+
 # instantiate
-model = SimpleT5()
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
 
 # load (supports t5, mt5, byT5 models)
 #model.from_pretrained("t5","t5-base")
@@ -79,16 +80,49 @@ def train_model():
 
 # load trained T5 model
 
+
+
 with st.spinner("Please wait while the model loads:"):
-    model = SimpleT5()
-    model.load_model("t5","Hellisotherpeople/T5_Reassuring_Parables")
+    tokenizer = AutoTokenizer.from_pretrained("Hellisotherpeople/T5_Reassuring_Parables")
+    model = AutoModelForSeq2SeqLM.from_pretrained("Hellisotherpeople/T5_Reassuring_Parables")
+
+form = st.sidebar.form("choose_settings")
 
+form.header("Main Settings")
+
+number_of_parables = form.number_input("Select how many reassuring parables you want to generate", value = 100, max_value = 1000)
+max_length_of_parable = form.number_input("What's the max length of the parable?", value = 20, max_value = 128)
+min_length_of_parable = form.number_input("What's the min length of the parable?", value = 0, max_value = max_length_of_parable)
+top_k = form.number_input("What value of K should we use for Top-K sampling? Set to zero to disable", value = 50)
+form.caption("In Top-K sampling, the K most likely next words are filtered and the probability mass is redistributed among only those K next words. ")
+top_p = form.number_input("What value of P should we use for Top-p sampling? Set to zero to disable", value = 0.95, max_value = 1.0, min_value = 0.0)
+form.caption("Top-p sampling chooses from the smallest possible set of words whose cumulative probability exceeds the probability p. The probability mass is then redistributed among this set of words.")
+temperature = form.number_input("How spicy/interesting do we want our models output to be", value = 1.05, min_value = 0.0)
+form.caption("Setting this higher decreases the likelihood of high probability words and increases the likelihood of low probability (and presumably more interesting) words")
+form.caption("For more details on what these settings mean, see here: https://huggingface.co/blog/how-to-generate")
+form.form_submit_button("Generate some Reassuring Parables!")
 
-number_of_parables = st.sidebar.number_input("Select how many reassuring parables you want to generate", value = 100)
 #seed_value = st.sidebar.number_input("Select a seed value - change this to get different output", 42) ## Doesn't work :(
 
+input_ids = tokenizer.encode("Computers will never", return_tensors='pt')
+
+sample_outputs = model.generate(
+    input_ids,
+    do_sample=True, 
+    max_length=max_length_of_parable, 
+    min_length=min_length_of_parable,
+    top_k=top_k, 
+    top_p=top_p, 
+    num_return_sequences=number_of_parables,
+    temperature=temperature
+)
+
+
 
 with st.spinner("Generating Reassuring Parables"):
     #pl.seed_everything(seed_value)
-    st.write(model.predict("Computers will never", num_return_sequences = number_of_parables, num_beams = number_of_parables))
+    list_of_parables = []
+    for i, sample_output in enumerate(sample_outputs):
+        list_of_parables.append(tokenizer.decode(sample_output, skip_special_tokens=True))
+    st.write(list_of_parables)
 

requirements.txt

@@ -1,2 +1,2 @@
 streamlit==1.2.0
-simplet5
+transformers