Spaces:
Runtime error
Runtime error
File size: 4,377 Bytes
2febf52 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 |
"""
Credit to Derek Thomas, [email protected]
"""
import subprocess
# subprocess.run(["pip", "install", "--upgrade", "transformers[torch,sentencepiece]==4.34.1"])
import logging
from pathlib import Path
from time import perf_counter
import gradio as gr
import numpy as np
from jinja2 import Environment, FileSystemLoader
from backend.query_llm import generate_hf, generate_openai
from backend.semantic_search import table, retriever, ranker
VECTOR_COLUMN_NAME = "vector"
TEXT_COLUMN_NAME = "text"
proj_dir = Path(__file__).parent
# Setting up the logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Set up the template environment with the templates directory
env = Environment(loader=FileSystemLoader(proj_dir / 'templates'))
# Load the templates directly from the environment
template = env.get_template('template.j2')
template_html = env.get_template('template_html.j2')
# Examples
examples = ['What is the capital of China?',
'Why is the sky blue?',
'Who won the mens world cup in 2014?', ]
def add_text(history, text):
history = [] if history is None else history
history = history + [(text, None)]
return history, gr.Textbox(value="", interactive=False)
def bot(history, api_kind):
top_k_retreive = 50
top_k_rank = 5
query = history[-1][0]
if not query:
gr.Warning("Please submit a non-empty string as a prompt")
raise ValueError("Empty string was submitted")
logger.warning('Retrieving documents...')
# Retrieve documents relevant to query
document_start = perf_counter()
query_vec = retriever.encode(query)
documents = table.search(query_vec, vector_column_name=VECTOR_COLUMN_NAME).limit(top_k_retreive).to_list()
documents = [doc[TEXT_COLUMN_NAME] for doc in documents]
document_time = perf_counter() - document_start
logger.warning(f'Finished Retrieving documents in {round(document_time, 2)} seconds...')
scores = ranker.predict([(query, doc) for doc in documents])
scores = np.argsort(scores)[::-1][:top_k_rank]
documents = [documents[idx] for idx in scores]
# Create Prompt
prompt = template.render(documents=documents, query=query)
prompt_html = template_html.render(documents=documents, query=query)
if api_kind == "HuggingFace":
generate_fn = generate_hf
elif api_kind == "OpenAI":
generate_fn = generate_openai
elif api_kind is None:
gr.Warning("API name was not provided")
raise ValueError("API name was not provided")
else:
gr.Warning(f"API {api_kind} is not supported")
raise ValueError(f"API {api_kind} is not supported")
history[-1][1] = ""
for character in generate_fn(prompt, history[:-1]):
history[-1][1] = character
yield history, prompt_html
with gr.Blocks() as demo:
chatbot = gr.Chatbot(
[],
elem_id="chatbot",
avatar_images=('https://aui.atlassian.com/aui/8.8/docs/images/avatar-person.svg',
'https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.svg'),
bubble_full_width=False,
show_copy_button=True,
show_share_button=True,
)
with gr.Row():
txt = gr.Textbox(
scale=3,
show_label=False,
placeholder="Enter text and press enter",
container=False,
)
txt_btn = gr.Button(value="Submit text", scale=1)
api_kind = gr.Radio(choices=["HuggingFace", "OpenAI"], value="HuggingFace")
prompt_html = gr.HTML()
# Turn off interactivity while generating if you click
txt_msg = txt_btn.click(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
bot, [chatbot, api_kind], [chatbot, prompt_html])
# Turn it back on
txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)
# Turn off interactivity while generating if you hit enter
txt_msg = txt.submit(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
bot, [chatbot, api_kind], [chatbot, prompt_html])
# Turn it back on
txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False)
# Examples
gr.Examples(examples, txt)
demo.queue()
demo.launch(debug=True)
|