File size: 8,780 Bytes
c9f26e8 4f5f090 ff4e3da c9f26e8 cb90219 4f5f090 c9f26e8 4f5f090 c9f26e8 4f5f090 c9f26e8 4f5f090 cb90219 c9f26e8 4f5f090 c9f26e8 4f5f090 c9f26e8 4f5f090 c9f26e8 4f5f090 c9f26e8 4f5f090 ff4e3da cb90219 ff4e3da 4f5f090 ff4e3da 4f5f090 ff4e3da c9f26e8 4f5f090 c9f26e8 ff4e3da c9f26e8 2591f90 e71f7f1 2591f90 4f5f090 2591f90 c9f26e8 2591f90 e71f7f1 2591f90 ff4e3da 2591f90 ff4e3da 4f5f090 2591f90 ff4e3da 4f5f090 ff4e3da c9f26e8 4f5f090 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 |
import os
import argparse
import asyncio
import gradio as gr
from difflib import Differ
from string import Template
from utils import load_prompt, setup_gemini_client
from configs.responses import SummaryResponses
from google.genai import types
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--ai-studio-api-key", type=str, default=os.getenv("GEMINI_API_KEY"))
parser.add_argument("--vertexai", action="store_true", default=False)
parser.add_argument("--vertexai-project", type=str, default="gcp-ml-172005")
parser.add_argument("--vertexai-location", type=str, default="us-central1")
parser.add_argument("--model", type=str, default="gemini-2.0-flash", choices=["gemini-1.5-flash", "gemini-2.0-flash", "gemini-2.0-flash-001"])
parser.add_argument("--seed", type=int, default=2025)
parser.add_argument("--prompt-tmpl-path", type=str, default="configs/prompts.toml")
parser.add_argument("--css-path", type=str, default="statics/styles.css")
args = parser.parse_args()
return args
def find_attached_file(filename, attached_files):
for file in attached_files:
if file['name'] == filename:
return file
return None
async def echo(message, history, state, persona):
attached_file = None
system_instruction = Template(prompt_tmpl['summarization']['system_prompt']).safe_substitute(persona=persona)
if message['files']:
path_local = message['files'][0]
filename = os.path.basename(path_local)
attached_file = find_attached_file(filename, state["attached_files"])
if attached_file is None:
path_gcp = await client.files.upload(path=path_local)
state["attached_files"].append({
"name": filename,
"path_local": path_local,
"gcp_entity": path_gcp,
"path_gcp": path_gcp.name,
"mime_type=": path_gcp.mime_type,
"expiration_time": path_gcp.expiration_time,
})
attached_file = path_gcp
user_message = [message['text']]
if attached_file: user_message.append(attached_file)
chat_history = state['messages']
chat_history = chat_history + user_message
state['messages'] = chat_history
response_chunks = ""
model_content_stream = await client.models.generate_content_stream(
model=args.model,
contents=state['messages'],
config=types.GenerateContentConfig(
system_instruction=system_instruction, seed=args.seed
),
)
async for chunk in model_content_stream:
response_chunks += chunk.text
# when model generates too fast, Gradio does not respond that in real-time.
await asyncio.sleep(0.1)
yield (
response_chunks,
state,
state['summary_diff_history'][-1] if len(state['summary_diff_history']) > 1 else "",
state['summary_history'][-1] if len(state['summary_history']) > 1 else "",
gr.Slider(
visible=False if len(state['summary_history']) <= 1 else True,
interactive=False if len(state['summary_history']) <= 1 else True,
),
)
# make summary
response = await client.models.generate_content(
model=args.model,
contents=[
Template(
prompt_tmpl['summarization']['prompt']
).safe_substitute(
previous_summary=state['summary'],
latest_conversation=str({"user": message['text'], "assistant": response_chunks})
)
],
config=types.GenerateContentConfig(
system_instruction=system_instruction,
seed=args.seed,
response_mime_type='application/json',
response_schema=SummaryResponses
)
)
prev_summary = state['summary_history'][-1] if len(state['summary_history']) >= 1 else ""
state['summary'] = (
response.parsed.summary
if getattr(response.parsed, "summary", None) is not None
else response.text
)
state['summary_history'].append(
response.parsed.summary
if getattr(response.parsed, "summary", None) is not None
else response.text
)
state['summary_diff_history'].append(
[
(token[2:], token[0] if token[0] != " " else None)
for token in Differ().compare(prev_summary, state['summary'])
]
)
yield (
response_chunks,
state,
state['summary_diff_history'][-1],
state['summary_history'][-1],
gr.Slider(
maximum=len(state['summary_history']),
value=len(state['summary_history']),
visible=False if len(state['summary_history']) == 1 else True, interactive=True
),
)
def change_view_toggle(view_toggle):
if view_toggle == "Diff":
return (
gr.HighlightedText(visible=True),
gr.Markdown(visible=False)
)
else:
return (
gr.HighlightedText(visible=False),
gr.Markdown(visible=True)
)
def navigate_to_summary(summary_num, state):
return (
state['summary_diff_history'][summary_num-1],
state['summary_history'][summary_num-1]
)
def main(args):
style_css = open(args.css_path, "r").read()
global client, prompt_tmpl, system_instruction
client = setup_gemini_client(args)
prompt_tmpl = load_prompt(args)
## Gradio Blocks
with gr.Blocks(css=style_css) as demo:
# State per session
state = gr.State({
"messages": [],
"attached_files": [],
"summary": "",
"summary_history": [],
"summary_diff_history": []
})
with gr.Column():
gr.Markdown("# Adaptive Summarization")
gr.Markdown("AdaptSum stands for Adaptive Summarization. This project focuses on developing an LLM-powered system for dynamic summarization. Instead of generating entirely new summaries with each update, the system intelligently identifies and modifies only the necessary parts of the existing summary. This approach aims to create a more efficient and fluid summarization process within a continuous chat interaction with an LLM.")
with gr.Column():
with gr.Accordion("Adaptively Summarized Conversation", elem_id="adaptive-summary-accordion", open=False):
with gr.Row(elem_id="view-toggle-btn-container"):
view_toggle_btn = gr.Radio(
choices=["Diff", "Markdown"],
value="Markdown",
interactive=True,
elem_id="view-toggle-btn"
)
summary_diff = gr.HighlightedText(
label="Summary so far",
# value="No summary yet. As you chat with the assistant, the summary will be updated automatically.",
combine_adjacent=True,
show_legend=True,
color_map={"-": "red", "+": "green"},
elem_classes=["summary-window"],
visible=False
)
summary_md = gr.Markdown(
label="Summary so far",
value="No summary yet. As you chat with the assistant, the summary will be updated automatically.",
elem_classes=["summary-window"],
visible=True
)
summary_num = gr.Slider(label="summary history", minimum=1, maximum=1, step=1, show_reset_button=False, visible=False)
view_toggle_btn.change(change_view_toggle, inputs=[view_toggle_btn], outputs=[summary_diff, summary_md])
summary_num.release(navigate_to_summary, inputs=[summary_num, state], outputs=[summary_diff, summary_md])
with gr.Column("persona-dropdown-container", elem_id="persona-dropdown-container"):
persona = gr.Dropdown(
["expert", "novice", "regular practitioner", "high schooler"],
label="Summary Persona",
info="Control the tonality of the conversation.",
min_width="auto",
)
with gr.Column("chat-window", elem_id="chat-window"):
gr.ChatInterface(
multimodal=True,
type="messages",
fn=echo,
additional_inputs=[state, persona],
additional_outputs=[state, summary_diff, summary_md, summary_num],
)
return demo
if __name__ == "__main__":
args = parse_args()
demo = main(args)
demo.launch()
|