File size: 3,753 Bytes
d60948f 1423dd4 d60948f 7a72052 d60948f 3555153 d60948f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 |
import cfg
import gradio as gr
import pandas as pd
from cfg import setup_buster
buster = setup_buster(cfg.buster_cfg)
def format_sources(matched_documents: pd.DataFrame) -> str:
if len(matched_documents) == 0:
return ""
matched_documents.similarity_to_answer = (
matched_documents.similarity_to_answer * 100
)
# print the page instead of the heading, more meaningful for hf docs
matched_documents["page"] = matched_documents.apply(
lambda x: x.url.split("/")[-1], axis=1
)
documents_answer_template: str = "📝 Here are the sources I used to answer your question:\n\n{documents}\n\n{footnote}"
document_template: str = "[🔗 {document.page}]({document.url}), relevance: {document.similarity_to_answer:2.1f} %"
documents = "\n".join(
[
document_template.format(document=document)
for _, document in matched_documents.iterrows()
]
)
footnote: str = "I'm a bot 🤖 and not always perfect."
return documents_answer_template.format(documents=documents, footnote=footnote)
def add_sources(history, completion):
if completion.answer_relevant:
formatted_sources = format_sources(completion.matched_documents)
history.append([None, formatted_sources])
return history
def user(user_input, history):
"""Adds user's question immediately to the chat."""
return "", history + [[user_input, None]]
def chat(history):
user_input = history[-1][0]
completion = buster.process_input(user_input)
print(completion)
history[-1][1] = ""
for token in completion.answer_generator:
history[-1][1] += token
yield history, completion
block = gr.Blocks()
with block:
gr.Markdown(
"""<h1><center>Buster 🤖: A Question-Answering Bot for your documentation</center></h1>"""
)
gr.Markdown(
"""
## Welcome to Buster!
This chatbot is designed to answer any questions related to the [huggingface transformers](https://huggingface.co/docs/transformers/index) library.
It uses ChatGPT + embeddings to search the docs for relevant sections and uses them to answer questions. It can then cite its sources back to you to verify the information.
Note that LLMs are prone to hallucination, so all outputs should always be vetted by users.
#### The Code is open-sourced and available on [Github](www.github.com/jerpint/buster)
"""
)
chatbot = gr.Chatbot()
with gr.Row():
with gr.Column(scale=4):
question = gr.Textbox(
label="What's your question?",
placeholder="Ask a question to AI stackoverflow here...",
lines=1,
)
submit = gr.Button(value="Send", variant="secondary")
examples = gr.Examples(
examples=[
"What kind of models should I use for images and text?",
"When should I finetune a model vs. training it form scratch?",
"Can you give me some python code to quickly finetune a model on my sentiment analysis dataset?",
],
inputs=question,
)
gr.HTML("️<center> Created with ❤️ by @jerpint and @hadrienbertrand.")
response = gr.State()
submit.click(user, [question, chatbot], [question, chatbot], queue=False).then(
chat, inputs=[chatbot], outputs=[chatbot, response]
).then(add_sources, inputs=[chatbot, response], outputs=[chatbot])
question.submit(user, [question, chatbot], [question, chatbot], queue=False).then(
chat, inputs=[chatbot], outputs=[chatbot, response]
).then(add_sources, inputs=[chatbot, response], outputs=[chatbot])
block.queue(concurrency_count=16)
block.launch(debug=True, share=False)
|