Spaces:
Running
on
TPU v5e
Running
on
TPU v5e
File size: 11,264 Bytes
2ca0c5e 40912b5 2ca0c5e 80e88ee 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e a2b7758 ba64cc5 2ca0c5e b637f0b ba64cc5 4ef1969 ba64cc5 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 80e88ee 2ca0c5e 80e88ee 2ca0c5e 80e88ee 2ca0c5e 4ef1969 80e88ee 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 40912b5 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e d78f59f 061118b 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 061118b 4ef1969 2ca0c5e 4ef1969 c16851d 4ef1969 2ca0c5e 061118b 4ef1969 2ca0c5e 4ef1969 061118b 4ef1969 2ca0c5e d78f59f 2ca0c5e 40912b5 2ca0c5e 64919f8 2ca0c5e 3a9ff40 2ca0c5e 4ef1969 d78f59f 2ca0c5e 4ef1969 061118b 4ef1969 061118b 4ef1969 061118b 4ef1969 061118b 4ef1969 061118b 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 4ef1969 2ca0c5e 80e88ee |
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 |
import os
# Questions for Gradio
# - Chat share button is enabled by default but thrown an error when clicked.
# - How to add local images in HTML? (https://github.com/gradio-app/gradio/issues/884)
# - How to allow Chatbot to fill the vertical space? (https://github.com/gradio-app/gradio/issues/4001)
# TODO:
# - Add the 1MB models, keras/gemma_1.1_instruct_7b_en
# - Add retry button, for each model individually
# - Add ability to route a message to a single model only.
# - log_applied_layout_map: make it work for Llama3CausalLM and LlamaCausalLM (vicuna)
# - display context length
os.environ["KERAS_BACKEND"] = "jax"
import gradio as gr
from gradio import ChatMessage
import keras_hub
from chatstate import ChatState
from enum import Enum
from models import (
model_presets,
load_model,
model_labels,
preset_to_website_url,
get_appropriate_chat_template,
)
class TextRoute(Enum):
LEFT = 0
RIGHT = 1
BOTH = 2
model_labels_list = list(model_labels)
# load and warm up (compile) all the models
models = []
for preset in model_presets:
model = load_model(preset)
chat_template = get_appropriate_chat_template(preset)
chat_state = ChatState(model, "", chat_template)
prompt, response = chat_state.send_message("Hello")
print("model " + preset + " loaded and initialized.")
print("The model responded: " + response)
models.append(model)
# For local debugging
# model = keras_hub.models.Llama3CausalLM.from_preset(
# # "hf://meta-llama/Llama-3.2-1B-Instruct", dtype="bfloat16"
# "../misc-code/ari_tiny_llama3"
# )
# models = [model, model, model, model, model]
def chat_turn_assistant(
message,
sel,
history,
system_message,
# max_tokens,
# temperature,
# top_p,
):
model = models[sel]
preset = model_presets[sel]
chat_template = get_appropriate_chat_template(preset)
chat_state = ChatState(model, system_message, chat_template)
for msg in history:
msg = ChatMessage(**msg)
if msg.role == "user":
chat_state.add_to_history_as_user(msg.content)
elif msg.role == "assistant":
chat_state.add_to_history_as_model(msg.content)
prompt, response = chat_state.send_message(message)
history.append(ChatMessage(role="assistant", content=response))
return history
def chat_turn_both_assistant(
message, sel1, sel2, history1, history2, system_message
):
return (
chat_turn_assistant(message, sel1, history1, system_message),
chat_turn_assistant(message, sel2, history2, system_message),
)
def chat_turn_user(message, history):
history.append(ChatMessage(role="user", content=message))
return history
def chat_turn_both_user(message, history1, history2):
return (
chat_turn_user(message, history1),
chat_turn_user(message, history2),
)
def bot_icon_select(model_name):
if "gemma" in model_name:
return "img/gemma.png"
elif "llama" in model_name:
return "img/meta.png"
elif "vicuna" in model_name:
return "img/vicuna.png"
elif "mistral" in model_name:
return "img/mistral.png"
# default
return "img/bot.png"
def instantiate_select_box(sel, model_labels):
return gr.Dropdown(
choices=[(name, i) for i, name in enumerate(model_labels)],
show_label=False,
value=sel,
info="<span style='color:black'>Selected model:</span> <a href='"
+ preset_to_website_url(model_presets[sel])
+ "'>"
+ preset_to_website_url(model_presets[sel])
+ "</a>",
)
def instantiate_chatbot(sel, key):
model_name = model_presets[sel]
return gr.Chatbot(
type="messages",
key=key,
show_label=False,
show_share_button=False,
show_copy_all_button=True,
avatar_images=("img/usr.png", bot_icon_select(model_name)),
)
def instantiate_arrow_button(route, text_route):
icons = {
TextRoute.LEFT: "img/arrowL.png",
TextRoute.RIGHT: "img/arrowR.png",
TextRoute.BOTH: "img/arrowRL.png",
}
button = gr.Button(
"",
size="sm",
scale=0,
min_width=40,
icon=icons[route],
)
button.click(lambda: route, outputs=[text_route])
return button
def instantiate_retry_button(route):
return gr.Button(
"",
size="sm",
scale=0,
min_width=40,
icon="img/retry.png",
)
def instantiate_trash_button():
return gr.Button(
"",
size="sm",
scale=0,
min_width=40,
icon="img/trash.png",
)
def instantiate_text_box():
return gr.Textbox(label="Your message:", submit_btn=True, key="msg")
def instantiate_additional_settings():
with gr.Accordion("Additional settings", open=False):
system_message = gr.Textbox(
label="Sytem prompt",
key="system_prompt",
value="You are a helpful assistant and your name is Eliza.",
)
return system_message
def retry_fn(history):
if len(history) >= 2:
msg = history.pop(-1) # assistant message
msg = history.pop(-1) # user message
return msg["content"], history
else:
return gr.skip(), gr.skip()
def retry_fn_both(history1, history2):
msg1, history1 = retry_fn(history1)
msg2, history2 = retry_fn(history2)
if isinstance(msg1, str) and isinstance(msg2, str):
if msg1 == msg2:
msg = msg1
else:
msg = msg1 + " / " + msg2
elif isinstance(msg1, str):
msg = msg1
elif isinstance(msg2, str):
msg = msg2
else:
msg = msg1
return msg, history1, history2
sel1 = instantiate_select_box(0, model_labels_list)
sel2 = instantiate_select_box(1, model_labels_list)
chatbot1 = instantiate_chatbot(sel1.value, "chat1")
chatbot2 = instantiate_chatbot(sel2.value, "chat2")
# to correctly align the left/right arrows
CSS = ".stick-to-the-right {align-items: end; justify-content: end}"
with gr.Blocks(fill_width=True, title="Keras demo", css=CSS) as demo:
# Where do messages go
text_route = gr.State(TextRoute.BOTH)
with gr.Row():
gr.Image(
"img/keras_logo_k.png",
width=80,
height=80,
min_width=80,
show_label=False,
show_download_button=False,
show_fullscreen_button=False,
show_share_button=False,
interactive=False,
scale=0,
container=False,
)
gr.HTML(
"<H2>Keras chatbot arena - running with JAX on TPU</H2>"
+ "All the models are loaded into the TPU memory. "
+ "You can call any of them and compare their answers. "
+ "The entire chat<br/>history is fed to the models at every submission. "
+ "This demo is runnig on a Google TPU v5e 2x4 (8 cores) in bfloat16 precision."
)
with gr.Row():
sel1.render(),
sel2.render(),
with gr.Row():
chatbot1.render()
chatbot2.render()
@gr.render(inputs=text_route)
def render_text_area(route):
if route == TextRoute.BOTH:
with gr.Row():
msg = instantiate_text_box()
with gr.Column(scale=0, min_width=100):
with gr.Row():
instantiate_arrow_button(TextRoute.LEFT, text_route)
retry = instantiate_retry_button(route)
with gr.Row():
instantiate_arrow_button(TextRoute.RIGHT, text_route)
trash = instantiate_trash_button()
retry.click(
retry_fn_both,
inputs=[chatbot1, chatbot2],
outputs=[msg, chatbot1, chatbot2],
)
trash.click(lambda: ("", [], []), outputs=[msg, chatbot1, chatbot2])
elif route == TextRoute.LEFT:
with gr.Row():
with gr.Column(scale=1):
msg = instantiate_text_box()
with gr.Column(scale=1):
with gr.Row():
instantiate_arrow_button(TextRoute.RIGHT, text_route)
retry = instantiate_retry_button(route)
with gr.Row():
instantiate_arrow_button(TextRoute.BOTH, text_route)
trash = instantiate_trash_button()
retry.click(retry_fn, inputs=[chatbot1], outputs=[msg, chatbot1])
trash.click(lambda: ("", []), outputs=[msg, chatbot1])
elif route == TextRoute.RIGHT:
with gr.Row():
with gr.Column(scale=1, elem_classes="stick-to-the-right"):
with gr.Row(elem_classes="stick-to-the-right"):
retry = instantiate_retry_button(route)
instantiate_arrow_button(TextRoute.LEFT, text_route)
with gr.Row(elem_classes="stick-to-the-right"):
trash = instantiate_trash_button()
instantiate_arrow_button(TextRoute.BOTH, text_route)
with gr.Column(scale=1):
msg = instantiate_text_box()
retry.click(retry_fn, inputs=[chatbot2], outputs=[msg, chatbot2])
trash.click(lambda: ("", []), outputs=[msg, chatbot2])
system_message = instantiate_additional_settings()
# Route the submitted message to the left, right or both chatbots
if route == TextRoute.LEFT:
submission = msg.submit(
chat_turn_user, inputs=[msg, chatbot1], outputs=[chatbot1]
).then(
chat_turn_assistant,
[msg, sel1, chatbot1, system_message],
outputs=[chatbot1],
)
elif route == TextRoute.RIGHT:
submission = msg.submit(
chat_turn_user, inputs=[msg, chatbot2], outputs=[chatbot2]
).then(
chat_turn_assistant,
[msg, sel2, chatbot2, system_message],
outputs=[chatbot2],
)
elif route == TextRoute.BOTH:
submission = msg.submit(
chat_turn_both_user,
inputs=[msg, chatbot1, chatbot2],
outputs=[chatbot1, chatbot2],
).then(
chat_turn_both_assistant,
[msg, sel1, sel2, chatbot1, chatbot2, system_message],
outputs=[chatbot1, chatbot2],
)
# In all cases reset text box after submission
submission.then(lambda: "", outputs=msg)
sel1.select(
lambda sel: instantiate_chatbot(sel, "chat1"),
inputs=[sel1],
outputs=[chatbot1],
).then(
lambda sel: instantiate_select_box(sel, model_labels_list),
inputs=[sel1],
outputs=[sel1],
)
sel2.select(
lambda sel: instantiate_chatbot(sel, "chat2"),
inputs=[sel2],
outputs=[chatbot2],
).then(
lambda sel: instantiate_select_box(sel, model_labels_list),
inputs=[sel2],
outputs=[sel2],
)
if __name__ == "__main__":
demo.launch()
|