File size: 11,347 Bytes
bad215c 1005936 75612e3 cce2bc7 bad215c c5daedb bad215c c5ff273 bad215c 80419fc |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 |
import datetime
import os
import random
import re
from io import StringIO
import gradio as gr
import pandas as pd
from huggingface_hub import upload_file
from text_generation import Client
from dialogues import DialogueTemplate
HF_TOKEN = os.environ.get("HF_TOKEN", None)
API_TOKEN = os.environ.get("API_TOKEN", None)
model2endpoint = {
"zephyr-7b-beta": "https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-beta",
"mistral-7b-instruct-v0.2": "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2",
"mixtral-8x7b-instruct-v0.1": "https://api-inference.huggingface.co/models/mistralai/Mixtral-8x7B-Instruct-v0.1",
"gemma-7b-it": "https://api-inference.huggingface.co/models/google/gemma-7b-it",
"llama-7b-chat": "https://api-inference.huggingface.co/models/meta-llama/Llama-2-7b-chat-hf"
}
model_names = list(model2endpoint.keys())
def randomize_seed_generator():
seed = random.randint(0, 1000000)
return seed
def get_total_inputs(inputs, chatbot, preprompt, user_name, assistant_name, sep):
past = []
for data in chatbot:
user_data, model_data = data
if not user_data.startswith(user_name):
user_data = user_name + user_data
if not model_data.startswith(sep + assistant_name):
model_data = sep + assistant_name + model_data
past.append(user_data + model_data.rstrip() + sep)
if not inputs.startswith(user_name):
inputs = user_name + inputs
total_inputs = preprompt + "".join(past) + inputs + sep + assistant_name.rstrip()
return total_inputs
def wrap_html_code(text):
pattern = r"<.*?>"
matches = re.findall(pattern, text)
if len(matches) > 0:
return f"```{text}```"
else:
return text
def has_no_history(chatbot, history):
return not chatbot and not history
def generate(
RETRY_FLAG,
model_name,
system_message,
user_message,
chatbot,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save=True,
):
client = Client(
model2endpoint[model_name],
headers={"Authorization": f"Bearer {API_TOKEN}"},
timeout=60,
)
# Don't return meaningless message when the input is empty
if not user_message:
print("Empty input")
if not RETRY_FLAG:
history.append(user_message)
seed = 42
else:
seed = randomize_seed_generator()
past_messages = []
for data in chatbot:
user_data, model_data = data
past_messages.extend(
[{"role": "user", "content": user_data}, {"role": "assistant", "content": model_data.rstrip()}]
)
if len(past_messages) < 1:
dialogue_template = DialogueTemplate(
system=system_message, messages=[{"role": "user", "content": user_message}]
)
prompt = dialogue_template.get_inference_prompt()
else:
dialogue_template = DialogueTemplate(
system=system_message, messages=past_messages + [{"role": "user", "content": user_message}]
)
prompt = dialogue_template.get_inference_prompt()
generate_kwargs = {
"temperature": temperature,
"top_k": top_k,
"top_p": top_p,
"max_new_tokens": max_new_tokens,
}
temperature = float(temperature)
if temperature < 1e-2:
temperature = 1e-2
top_p = float(top_p)
generate_kwargs = dict(
temperature=temperature,
max_new_tokens=max_new_tokens,
top_p=top_p,
repetition_penalty=repetition_penalty,
do_sample=True,
truncate=4096,
seed=seed,
stop_sequences=["<|end|>"],
)
stream = client.generate_stream(
prompt,
**generate_kwargs,
)
output = ""
for idx, response in enumerate(stream):
if response.token.special:
continue
output += response.token.text
if idx == 0:
history.append(" " + output)
else:
history[-1] = output
chat = [
(wrap_html_code(history[i].strip()), wrap_html_code(history[i + 1].strip()))
for i in range(0, len(history) - 1, 2)
]
# chat = [(history[i].strip(), history[i + 1].strip()) for i in range(0, len(history) - 1, 2)]
yield chat, history, user_message, ""
return chat, history, user_message, ""
examples = [
"What are the signs and symptoms of community acquired pneumonia (CAP)?", "What is the treatment for recurrent otitis media?"
]
def clear_chat():
return [], []
def delete_last_turn(chat, history):
if chat and history:
chat.pop(-1)
history.pop(-1)
history.pop(-1)
return chat, history
def process_example(args):
for [x, y] in generate(args):
pass
return [x, y]
# Regenerate response
def retry_last_answer(
selected_model,
system_message,
user_message,
chat,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save,
):
if chat and history:
# Removing the previous conversation from chat
chat.pop(-1)
# Removing bot response from the history
history.pop(-1)
# Setting up a flag to capture a retry
RETRY_FLAG = True
# Getting last message from user
user_message = history[-1]
yield from generate(
RETRY_FLAG,
selected_model,
system_message,
user_message,
chat,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save,
)
title = """<h1 align="center">LLM Playground π¬</h1>"""
custom_css = """
#banner-image {
display: block;
margin-left: auto;
margin-right: auto;
}
#chat-message {
font-size: 14px;
min-height: 300px;
}
"""
with gr.Blocks(analytics_enabled=False, css=custom_css) as demo:
gr.HTML(title)
with gr.Row():
# with gr.Column():
# gr.Image("thumbnail.png", elem_id="banner-image", show_label=False)
with gr.Column():
gr.Markdown(
"""
π» This demo showcases a few smaller open source models."""
)
with gr.Row():
selected_model = gr.Radio(choices=model_names, value=model_names[1], label="Select a model")
with gr.Accordion(label="System Prompt", open=False, elem_id="parameters-accordion"):
system_message = gr.Textbox(
elem_id="system-message",
placeholder="Below is a conversation between a medical student and a helpful AI medical assistant.",
show_label=False,
lines=10,
max_lines=100
)
with gr.Row():
with gr.Group():
output = gr.Markdown()
chatbot = gr.Chatbot(elem_id="chat-message", label="Chat")
with gr.Row():
with gr.Column(scale=3):
user_message = gr.Textbox(placeholder="Enter your message here", show_label=False, elem_id="q-input")
with gr.Row():
send_button = gr.Button("Send", elem_id="send-btn", visible=True)
regenerate_button = gr.Button("Regenerate", elem_id="retry-btn", visible=True)
delete_turn_button = gr.Button("Delete last turn", elem_id="delete-btn", visible=True)
clear_chat_button = gr.Button("Clear chat", elem_id="clear-btn", visible=True)
with gr.Accordion(label="Parameters", open=False, elem_id="parameters-accordion"):
temperature = gr.Slider(
label="Temperature",
value=0.2,
minimum=0.0,
maximum=1.0,
step=0.1,
interactive=True,
info="Higher values produce more diverse outputs",
)
top_k = gr.Slider(
label="Top-k",
value=50,
minimum=0.0,
maximum=100,
step=1,
interactive=True,
info="Sample from a shortlist of top-k tokens",
)
top_p = gr.Slider(
label="Top-p (nucleus sampling)",
value=0.95,
minimum=0.0,
maximum=1,
step=0.05,
interactive=True,
info="Higher values sample more low-probability tokens",
)
max_new_tokens = gr.Slider(
label="Max new tokens",
value=512,
minimum=0,
maximum=1024,
step=4,
interactive=True,
info="The maximum numbers of new tokens",
)
repetition_penalty = gr.Slider(
label="Repetition Penalty",
value=1.2,
minimum=0.0,
maximum=10,
step=0.1,
interactive=True,
info="The parameter for repetition penalty. 1.0 means no penalty.",
)
with gr.Row():
gr.Examples(
examples=examples,
inputs=[user_message],
cache_examples=False,
fn=process_example,
outputs=[output],
)
history = gr.State([])
RETRY_FLAG = gr.Checkbox(value=False, visible=False)
# To clear out "message" input textbox and use this to regenerate message
last_user_message = gr.State("")
user_message.submit(
generate,
inputs=[
RETRY_FLAG,
selected_model,
system_message,
user_message,
chatbot,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save,
],
outputs=[chatbot, history, last_user_message, user_message],
)
send_button.click(
generate,
inputs=[
RETRY_FLAG,
selected_model,
system_message,
user_message,
chatbot,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save,
],
outputs=[chatbot, history, last_user_message, user_message],
)
regenerate_button.click(
retry_last_answer,
inputs=[
selected_model,
system_message,
user_message,
chatbot,
history,
temperature,
top_k,
top_p,
max_new_tokens,
repetition_penalty,
# do_save,
],
outputs=[chatbot, history, last_user_message, user_message],
)
delete_turn_button.click(delete_last_turn, [chatbot, history], [chatbot, history])
clear_chat_button.click(clear_chat, outputs=[chatbot, history])
selected_model.change(clear_chat, outputs=[chatbot, history])
demo.queue().launch(debug=True) |