from __future__ import annotations
from typing import Iterable
import gradio as gr
from gradio.themes.base import Base
from gradio.themes.utils import colors, fonts, sizes
import time
import torch
from transformers import pipeline
import pandas as pd
instruct_pipeline = pipeline(model="databricks/dolly-v2-7b", torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto")
def run_pipeline(prompt):
response = instruct_pipeline(prompt)
return response
def get_user_input(input_question, history):
return "", history + [[input_question, None]]
def get_qa_user_input(input_question, history):
return "", history + [[input_question, None]]
def dolly_chat(history):
prompt = history[-1][0]
bot_message = run_pipeline(prompt)
history[-1][1] = bot_message
return history
def qa_bot(context, history):
query = history[-1][0]
prompt = f'instruction: {query} \ncontext: {context}'
bot_message = run_pipeline(prompt)
history[-1][1] = bot_message
return history
def reset_chatbot():
return gr.update(value="")
def load_customer_support_example():
df = pd.read_csv("examples.csv")
return df['doc'].iloc[0], df['question'].iloc[0]
def load_databricks_doc_example():
df = pd.read_csv("examples.csv")
return df['doc'].iloc[1], df['question'].iloc[1]
# Referred & modified from https://gradio.app/theming-guide/
class SeafoamCustom(Base):
def __init__(
self,
*,
primary_hue: colors.Color | str = colors.emerald,
secondary_hue: colors.Color | str = colors.blue,
neutral_hue: colors.Color | str = colors.blue,
spacing_size: sizes.Size | str = sizes.spacing_md,
radius_size: sizes.Size | str = sizes.radius_md,
font: fonts.Font
| str
| Iterable[fonts.Font | str] = (
fonts.GoogleFont("Quicksand"),
"ui-sans-serif",
"sans-serif",
),
font_mono: fonts.Font
| str
| Iterable[fonts.Font | str] = (
fonts.GoogleFont("IBM Plex Mono"),
"ui-monospace",
"monospace",
),
):
super().__init__(
primary_hue=primary_hue,
secondary_hue=secondary_hue,
neutral_hue=neutral_hue,
spacing_size=spacing_size,
radius_size=radius_size,
font=font,
font_mono=font_mono,
)
super().set(
button_primary_background_fill="linear-gradient(90deg, *primary_300, *secondary_400)",
button_primary_background_fill_hover="linear-gradient(90deg, *primary_200, *secondary_300)",
button_primary_text_color="white",
button_primary_background_fill_dark="linear-gradient(90deg, *primary_600, *secondary_800)",
block_shadow="*shadow_drop_lg",
button_shadow="*shadow_drop_lg",
input_background_fill="zinc",
input_border_color="*secondary_300",
input_shadow="*shadow_drop",
input_shadow_focus="*shadow_drop_lg",
)
seafoam = SeafoamCustom()
with gr.Blocks(theme=seafoam) as demo:
with gr.Row(variant='panel'):
with gr.Column():
gr.HTML(
"""
"""
)
with gr.Column():
gr.Markdown("# **Dolly 2.0: World's First Truly Open Instruction-Tuned LLM
**")
gr.Markdown("Dolly 2.0, the first open source, instruction-following LLM, fine-tuned on a human-generated instruction dataset licensed for research and commercial use. It's a 12B parameter language model based on the EleutherAI pythia model family and fine-tuned exclusively on a new, high-quality human generated instruction following dataset, crowdsourced among Databricks employees.")
qa_bot_state = gr.State(value=[])
with gr.Tabs():
with gr.TabItem("Dolly Chat"):
with gr.Row():
with gr.Column():
chatbot = gr.Chatbot(label="Chat History")
input_question = gr.Text(
label="Instruction",
placeholder="Type prompt and hit enter.",
)
clear = gr.Button("Clear", variant="primary")
with gr.Row():
with gr.Accordion("Show example inputs I can load:", open=False):
gr.Examples(
[
["Explain to me the difference between nuclear fission and fusion."],
["Give me a list of 5 science fiction books I should read next."],
["I'm selling my Nikon D-750, write a short blurb for my ad."],
["Write a song about sour donuts"],
["Write a tweet about a new book launch by J.K. Rowling."],
],
[input_question],
[],
None,
cache_examples=False,
)
with gr.TabItem("Q&A with Context"):
with gr.Row():
with gr.Column():
input_context = gr.Text(label="Add context here", lines=10)
with gr.Column():
qa_chatbot = gr.Chatbot(label="Q&A History")
qa_input_question = gr.Text(
label="Input Question",
placeholder="Type question here and hit enter.",
)
qa_clear = gr.Button("Clear", variant="primary")
with gr.Row():
with gr.Accordion("Show example inputs I can load:", open=False):
example_1 = gr.Button("Load Customer support example")
example_2 = gr.Button("Load Databricks documentation example")
input_question.submit(
get_user_input,
[input_question, chatbot],
[input_question, chatbot],
).then(dolly_chat, [chatbot], chatbot)
clear.click(lambda: None, None, chatbot)
qa_input_question.submit(
get_qa_user_input,
[qa_input_question, qa_chatbot],
[qa_input_question, qa_chatbot],
).then(qa_bot, [input_context, qa_chatbot], qa_chatbot)
qa_clear.click(lambda: None, None, qa_chatbot)
# reset the chatbot Q&A history when input context changes
input_context.change(fn=reset_chatbot, inputs=[], outputs=qa_chatbot)
example_1.click(
load_customer_support_example,
[],
[input_context, qa_input_question],
)
example_2.click(
load_databricks_doc_example,
[],
[input_context, qa_input_question],
)
if __name__ == "__main__":
demo.queue(concurrency_count=1,max_size=100).launch(max_threads=5,debug=True)