|
import json |
|
import os |
|
import shutil |
|
import requests |
|
|
|
import gradio as gr |
|
from huggingface_hub import Repository, InferenceClient |
|
|
|
HF_TOKEN = os.environ.get("HF_TOKEN", None) |
|
API_URL = "https://api-inference.huggingface.co/models/tiiuae/falcon-180B-chat" |
|
BOT_NAME = "Falcon" |
|
|
|
STOP_SEQUENCES = ["\nUser:", "<|endoftext|>", " User:", "###"] |
|
|
|
EXAMPLES = [["climate change"], ["2308.15699"], ["hallucination"], ["2308.00205"], ["large language model"], ["2308.05204"], ["2308.10873"], ["2308.06355"],["2308.01684"],["2308.00352"],["2308.07773"]] |
|
|
|
client = InferenceClient( |
|
API_URL, |
|
headers={"Authorization": f"Bearer {HF_TOKEN}"}, |
|
) |
|
|
|
id_dict = {} |
|
for i in range(0,4): |
|
fname = "arxiv_2023_" + str(i) |
|
with open(fname, "r") as f: |
|
for line in f: |
|
D = json.loads(line) |
|
id_dict[D['id']] = D |
|
|
|
|
|
def format_prompt_summarize(message, history, system_prompt, keyword): |
|
|
|
prompt = "" |
|
prompt += "System: You are scholarly RESEARCH ASSISTANT who can read the ARXIV scholarly article.\n" |
|
prompt += "User: READ ALL THE TITLEs and ABSTRACTs of various article below\n" |
|
prompt += "Generate a SUMMARY of all the articles below relevant to the research for the field of \"" + keyword + "\"\n" |
|
prompt += "SUGGEST FIVE IMPORTANT FINDINGS or ORIGINAL CONTRIBUTIONS of OBSERVATIONs for the field of \"" + keyword + "\" that summarizes the work.\n" |
|
prompt += "Each BULLET POINT must be be less than 15 WORDS. \n" |
|
prompt += "Output the FIVE KEY FINDINGS as BULLET POINTS with UNDERLINE OR BOLDEN KEY PHRASES.\n" |
|
prompt += "Propose ONE CREATIVE ACTIONABLE IDEA for FUTURE extension of the RESEARCH\n. You MUST output the CREATIVE IDEA with a BULB OR IDEA OR THINKING emoji.\n" |
|
prompt += "Output ONE CREATIVE IDEA for FUTURE extension with a RANDOM emoji\n" |
|
prompt += "Choose an UNRELATED or ORTHOGONAL field where the FINDINGS of the article can be applied.\n" |
|
prompt += "In a new line, OUTPUT ONE CRAZY IDEA in 20 WORDS how the KEY FINDINGS of RESEARCH article can be applied in an ORTHOGONAL or UNRELATED FIELD with a CRAZY IDEA emoji \n" |
|
prompt += message + "\n" |
|
|
|
mock_prompt = "" |
|
if system_prompt == "": |
|
mock_prompt += f"System: {system_prompt}\n" |
|
for user_prompt, bot_response in history: |
|
mock_prompt += f"User: {user_prompt}\n" |
|
mock_prompt += f"Falcon: {bot_response}\n" |
|
mock_prompt += f"""User: {message} |
|
Falcon:""" |
|
return prompt |
|
|
|
|
|
|
|
def format_prompt(message, history, system_prompt): |
|
|
|
prompt = "" |
|
prompt += "System: You are scholarly RESEARCH ASSISTANT who can read the ARXIV scholarly article.\n" |
|
prompt += "READ THE TITLE and ABSTRACT of the article below\n" |
|
prompt += "After understanding the ABSTRACT, SUGGEST 4 IMPORTANT FINDINGS or ORIGINAL CONTRIBUTIONS of OBSERVATIONs that summarizes the work.\n" |
|
prompt += "Each BULLET POINT must be be less than 15 WORDS. \n" |
|
prompt += "Output the FOUR KEY FINDINGS as BULLET POINTS with UNDERLINE OR BOLDEN KEY PHRASES.\n" |
|
prompt += "Propose ONE CREATIVE ACTIONABLE IDEA for FUTURE extension of the RESEARCH\n. You MUST output the CREATIVE IDEA with a BULB OR IDEA OR THINKING emoji.\n" |
|
prompt += "Output ONE CREATIVE IDEA for FUTURE extension with a RANDOM emoji\n" |
|
prompt += "Choose an UNRELATED or ORTHOGONAL field where the FINDINGS of the article can be applied.\n" |
|
prompt += "In a new line, OUTPUT ONE CRAZY IDEA in 20 WORDS how the KEY FINDINGS of RESEARCH article can be applied in an ORTHOGONAL or UNRELATED FIELD with a CRAZY IDEA emoji \n" |
|
prompt += "User:" + message + "\n" |
|
mock_prompt = "" |
|
if system_prompt == "": |
|
mock_prompt += f"System: {system_prompt}\n" |
|
for user_prompt, bot_response in history: |
|
mock_prompt += f"User: {user_prompt}\n" |
|
mock_prompt += f"Falcon: {bot_response}\n" |
|
mock_prompt += f"""User: {message} |
|
Falcon:""" |
|
return prompt |
|
|
|
seed = 42 |
|
|
|
def generate( |
|
prompt, history, system_prompt="", temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0, |
|
): |
|
temperature = float(temperature) |
|
if temperature < 1e-2: |
|
temperature = 1e-2 |
|
top_p = float(top_p) |
|
global seed |
|
generate_kwargs = dict( |
|
temperature=temperature, |
|
max_new_tokens=max_new_tokens, |
|
top_p=top_p, |
|
repetition_penalty=repetition_penalty, |
|
stop_sequences=STOP_SEQUENCES, |
|
do_sample=True, |
|
seed=seed, |
|
) |
|
seed = seed + 1 |
|
|
|
title = "INPUT ARXI ID" |
|
abstract = "" |
|
if prompt in id_dict: |
|
title = id_dict[prompt]['title'] |
|
abstract = id_dict[prompt]['abstract'] |
|
prompt = f"TITLE: {title} ABSTRACT: {abstract}\n" |
|
output = f"<b>Title: </b> {title} \n <br>" |
|
formatted_prompt = format_prompt(prompt, history, system_prompt) |
|
else: |
|
keyword = prompt |
|
counter= 0 |
|
for d in id_dict: |
|
title = id_dict[d]['title'] |
|
abstract = id_dict[d]['abstract'] |
|
if keyword in title or keyword in abstract: |
|
counter+=1 |
|
prompt += "ARTICLE " + str(counter) + "\n" |
|
prompt += f"TITLE: {title} ABSTRACT: {abstract}\n" |
|
if counter >= 4: |
|
break |
|
|
|
prompt += "Keyword: " + keyword + "\n" |
|
formatted_prompt = format_prompt_summarize(prompt, history, system_prompt, keyword) |
|
output = "Articles related to the keyword " + keyword + "\n" |
|
|
|
|
|
|
|
|
|
|
|
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False) |
|
|
|
|
|
for response in stream: |
|
output += response.token.text |
|
|
|
for stop_str in STOP_SEQUENCES: |
|
if output.endswith(stop_str): |
|
output = output[:-len(stop_str)] |
|
output = output.rstrip() |
|
yield output |
|
yield output |
|
return output |
|
|
|
|
|
additional_inputs=[ |
|
gr.Textbox("", label="Optional system prompt"), |
|
gr.Slider( |
|
label="Temperature", |
|
value=0.9, |
|
minimum=0.0, |
|
maximum=1.0, |
|
step=0.05, |
|
interactive=True, |
|
info="Higher values produce more diverse outputs", |
|
), |
|
gr.Slider( |
|
label="Max new tokens", |
|
value=256, |
|
minimum=0, |
|
maximum=8192, |
|
step=64, |
|
interactive=True, |
|
info="The maximum numbers of new tokens", |
|
), |
|
gr.Slider( |
|
label="Top-p (nucleus sampling)", |
|
value=0.90, |
|
minimum=0.0, |
|
maximum=1, |
|
step=0.05, |
|
interactive=True, |
|
info="Higher values sample more low-probability tokens", |
|
), |
|
gr.Slider( |
|
label="Repetition penalty", |
|
value=1.2, |
|
minimum=1.0, |
|
maximum=2.0, |
|
step=0.05, |
|
interactive=True, |
|
info="Penalize repeated tokens", |
|
) |
|
] |
|
|
|
|
|
with gr.Blocks() as demo: |
|
with gr.Row(): |
|
with gr.Column(scale=0.4): |
|
gr.Image("better_banner.jpeg", elem_id="banner-image", show_label=False) |
|
with gr.Column(): |
|
gr.Markdown( |
|
""" |
|
# |
|
** The idea is inspired by CREATIVE WHACK PACK https://apps.apple.com/us/app/creative-whack-pack/id307306326 |
|
|
|
** ##Researchers need INSPIRATION to come up with CREATIVE IDEAS. |
|
** ###We use Falcon 180B to |
|
<br> - generate a <b>SUMMARY</b> of the arxiv articles (only August articles are supported) |
|
<br> - generate a <b>CREATIVE IDEA </b> for future extension |
|
<br> - generate a </b>CRAZY IDEA</b> for application in an orthogonal field. |
|
|
|
This should hopefully CONNECT unrelated fields and inspire researchers to come up with CREATIVE IDEAS. |
|
## Please input ARXIV ID or a query, see examples below (limited to 15K articles from August 2023) |
|
➡️️ **Intended Use**: this demo is intended to showcase how LLMs can be used to generate creative ideas for future extension and application in orthogonal field. |
|
|
|
⚠️ **Limitations**: the model can and will produce factually incorrect information, hallucinating facts and actions. As it has not undergone any advanced tuning/alignment, it can produce problematic outputs, especially if prompted to do so. Finally, this demo is limited to a session length of about 1,000 words. |
|
""" |
|
) |
|
|
|
gr.ChatInterface( |
|
generate, |
|
examples=EXAMPLES, |
|
additional_inputs=additional_inputs, |
|
) |
|
|
|
demo.queue(concurrency_count=100, api_open=False).launch(show_api=False) |
