|
|
import openai |
|
|
import streamlit as st |
|
|
from transformers import AutoTokenizer |
|
|
import pandas as pd |
|
|
import numpy as np |
|
|
|
|
|
@st.cache(allow_output_mutation=True) |
|
|
def get_model(): |
|
|
tokenizer = AutoTokenizer.from_pretrained("gpt2") |
|
|
return tokenizer |
|
|
|
|
|
tokenizer = get_model() |
|
|
|
|
|
number_of_probs = st.slider("Length", 5, 99) |
|
|
temp = st.slider('Temperature', 0.7, 5.0, 1.0) |
|
|
tokens = st.slider('Length', 3, 25, 20) |
|
|
|
|
|
def run_generate(bad_words): |
|
|
bad_words = bad_words.split() |
|
|
bad_word_ids = [] |
|
|
for bad_word in bad_words: |
|
|
bad_word = " " + bad_word |
|
|
ids = tokenizer(bad_word).input_ids |
|
|
ids = str(ids) |
|
|
ids = ids.replace("]", ": -30").replace("[", "").replace(", ", ":-30, ") |
|
|
bad_word_ids.append(ids) |
|
|
bad_word_ids = str(bad_word_ids) |
|
|
bad_word_ids = bad_word_ids.replace("['", "{").replace("']", "}").replace("'", "") |
|
|
bad_word_ids = bad_word_ids + "," |
|
|
print(bad_word_ids) |
|
|
return bad_word_ids |
|
|
|
|
|
with st.form(key='my_form'): |
|
|
prompt = st.text_area(label='Enter sentence', value=" ") |
|
|
bad_words = st.text_area("Words You Do Not Want Generated", " core lemon height time ") |
|
|
key = st.text_area("Key", " core lemon height time ") |
|
|
openai.api_key = key |
|
|
bad_words = run_generate(bad_words) |
|
|
yes =eval(bad_words) |
|
|
bad_wordss = (yes[0]) |
|
|
submit_button = st.form_submit_button(label='Submit') |
|
|
submit_button2 = st.form_submit_button(label='Submit Log Probs') |
|
|
if submit_button: |
|
|
response = openai.Completion.create(engine="code-davinci-002", prompt=prompt, temperature=temp, max_tokens=tokens, top_p=0.8, n=50, |
|
|
logit_bias = bad_wordss,) |
|
|
st.write(response) |
|
|
if submit_button2: |
|
|
response = openai.Completion.create(engine="code-davinci-002", prompt=prompt + " ", top_p=0.1, logprobs=number_of_probs, max_tokens=1, logit_bias = bad_wordss,) |
|
|
scores = pd.DataFrame([response["choices"][0]["logprobs"]["top_logprobs"][0]]).T |
|
|
scores.columns = ["logprob"] |
|
|
scores["%"] = scores["logprob"].apply(lambda x: 100*np.e**x) |
|
|
g = scores.sort_values(by="%", ascending=False) |
|
|
first_column = g.iloc[:, 0] |
|
|
okay = first_column.to_string() |
|
|
h = [] |
|
|
for i in range(len(okay)): |
|
|
try: |
|
|
if okay[i] == "-": |
|
|
okay = okay[0:i] + okay[i+9:] |
|
|
h.append(okay) |
|
|
except: |
|
|
pass |
|
|
words2 = (h[-1]) |
|
|
words = words2.split() |
|
|
st.write(words) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
