Spaces:
Running
on
Zero
Running
on
Zero
File size: 18,559 Bytes
d6bc023 6d74e51 d6bc023 6d74e51 d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 8f2263e d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 1b2e2ca d6bc023 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 |
import argparse
import hashlib
import json
import os
import time
from threading import Thread
import logging
import gradio as gr
import torch
from tinychart.model.builder import load_pretrained_model
from tinychart.mm_utils import (
KeywordsStoppingCriteria,
load_image_from_base64,
process_images,
tokenizer_image_token,
get_model_name_from_path,
)
from PIL import Image
from io import BytesIO
import base64
import torch
from transformers import StoppingCriteria
from tinychart.constants import (
DEFAULT_IM_END_TOKEN,
DEFAULT_IM_START_TOKEN,
DEFAULT_IMAGE_TOKEN,
IMAGE_TOKEN_INDEX,
)
from tinychart.conversation import SeparatorStyle, conv_templates, default_conversation
from tinychart.eval.eval_metric import parse_model_output, evaluate_cmds
from transformers import TextIteratorStreamer
from pathlib import Path
import spaces
DEFAULT_MODEL_PATH = "mPLUG/TinyChart-3B-768"
DEFAULT_MODEL_NAME = "TinyChart-3B-768"
block_css = """
#buttons button {
min-width: min(120px,100%);
}
"""
title_markdown = """
# TinyChart: Efficient Chart Understanding with Visual Token Merging and Program-of-Thoughts Learning
π [[Code](https://github.com/X-PLUG/mPLUG-DocOwl/tree/main/TinyChart)] | π [[Paper](https://arxiv.org/abs/2404.16635)]
**Note:**
1. Currently, this demo only supports English chart understanding and may not work well with other languages.
2. To use Program-of-Thoughts answer, please append "Answer with detailed steps." to your question.
"""
tos_markdown = """
### Terms of use
By using this service, users are required to agree to the following terms:
The service is a research preview intended for non-commercial use only. It only provides limited safety measures and may generate offensive content. It must not be used for any illegal, harmful, violent, racist, or sexual purposes.
For an optimal experience, please use desktop computers for this demo, as mobile devices may compromise its quality.
"""
def regenerate(state, image_process_mode):
state.messages[-1][-1] = None
prev_human_msg = state.messages[-2]
if type(prev_human_msg[1]) in (tuple, list):
prev_human_msg[1] = (*prev_human_msg[1][:2], image_process_mode)
state.skip_next = False
return (state, state.to_gradio_chatbot(), "", None)
def clear_history():
state = default_conversation.copy()
return (state, state.to_gradio_chatbot(), "", None)
def add_text(state, text, image, image_process_mode):
if len(text) <= 0 and image is None:
state.skip_next = True
return (state, state.to_gradio_chatbot(), "", None)
text = text[:1536] # Hard cut-off
if image is not None:
text = text[:1200] # Hard cut-off for images
if "<image>" not in text:
# text = '<Image><image></Image>' + text
# text = text + "\n<image>"
text = "<image>\n"+text
text = (text, image, image_process_mode)
if len(state.get_images(return_pil=True)) > 0:
state = default_conversation.copy()
state.append_message(state.roles[0], text)
state.append_message(state.roles[1], None)
state.skip_next = False
return (state, state.to_gradio_chatbot(), "", None)
def load_demo():
state = default_conversation.copy()
return state
def is_float(value):
try:
float(value)
return True
except ValueError:
return False
@torch.inference_mode()
@spaces.GPU
def get_response(params):
prompt = params["prompt"]
ori_prompt = prompt
images = params.get("images", None)
num_image_tokens = 0
if images is not None and len(images) > 0:
if len(images) > 0:
if len(images) != prompt.count(DEFAULT_IMAGE_TOKEN):
raise ValueError(
"Number of images does not match number of <image> tokens in prompt"
)
images = [load_image_from_base64(image) for image in images]
images = process_images(images, image_processor, model.config)
if type(images) is list:
images = [
image.to(model.device, dtype=torch.float32) for image in images
]
else:
images = images.to(model.device, dtype=torch.float32)
replace_token = DEFAULT_IMAGE_TOKEN
if getattr(model.config, "mm_use_im_start_end", False):
replace_token = (
DEFAULT_IM_START_TOKEN + replace_token + DEFAULT_IM_END_TOKEN
)
prompt = prompt.replace(DEFAULT_IMAGE_TOKEN, replace_token)
if hasattr(model.get_vision_tower().config, "tome_r"):
num_image_tokens = (
prompt.count(replace_token) * model.get_vision_tower().num_patches - 26 * model.get_vision_tower().config.tome_r
)
else:
num_image_tokens = (
prompt.count(replace_token) * model.get_vision_tower().num_patches
)
else:
images = None
image_args = {"images": images}
else:
images = None
image_args = {}
temperature = float(params.get("temperature", 1.0))
top_p = float(params.get("top_p", 1.0))
max_context_length = getattr(model.config, "max_position_embeddings", 2048)
max_new_tokens = min(int(params.get("max_new_tokens", 256)), 1024)
stop_str = params.get("stop", None)
do_sample = True if temperature > 0.001 else False
logger.info(prompt)
input_ids = (
tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt")
.unsqueeze(0)
.to(model.device)
)
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
streamer = TextIteratorStreamer(
tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=1500
)
max_new_tokens = min(
max_new_tokens, max_context_length - input_ids.shape[-1] - num_image_tokens
)
if max_new_tokens < 1:
yield json.dumps(
{
"text": ori_prompt
+ "Exceeds max token length. Please start a new conversation, thanks.",
"error_code": 0,
}
).encode() + b"\0"
return
# local inference
# BUG: If stopping_criteria is set, an error occur:
# RuntimeError: The size of tensor a (2) must match the size of tensor b (3) at non-singleton dimension 0
generate_kwargs = dict(
inputs=input_ids,
do_sample=do_sample,
temperature=temperature,
top_p=top_p,
max_new_tokens=max_new_tokens,
streamer=streamer,
# stopping_criteria=[stopping_criteria],
use_cache=True,
**image_args,
)
thread = Thread(target=model.generate, kwargs=generate_kwargs)
thread.start()
logger.debug(ori_prompt)
logger.debug(generate_kwargs)
generated_text = ori_prompt
for new_text in streamer:
generated_text += new_text
if generated_text.endswith(stop_str):
generated_text = generated_text[: -len(stop_str)]
yield json.dumps({"text": generated_text, "error_code": 0}).encode()
if '<step>' in generated_text and '</step>' in generated_text and '<comment>' in generated_text and '</comment>' in generated_text:
program = generated_text
program = '<comment>#' + program.split('ASSISTANT: <comment>#')[-1]
print(program)
try:
execuate_result = evaluate_cmds(parse_model_output(program))
if is_float(execuate_result):
execuate_result = round(float(execuate_result), 4)
generated_text += f'\n\nExecute result: {execuate_result}'
yield json.dumps({"text": generated_text, "error_code": 0}).encode() + b"\0"
except:
# execuate_result = 'Failed.'
generated_text += f'\n\nIt seems the execution of the above code encounters bugs. I\'m trying to answer this question directly...'
ori_generated_text = generated_text + '\nDirect Answer: '
direct_prompt = ori_prompt.replace(' Answer with detailed steps.', '')
direct_input_ids = (
tokenizer_image_token(direct_prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt")
.unsqueeze(0)
.to(model.device)
)
generate_kwargs = dict(
inputs=direct_input_ids,
do_sample=do_sample,
temperature=temperature,
top_p=top_p,
max_new_tokens=max_new_tokens,
streamer=streamer,
use_cache=True,
**image_args,
)
thread = Thread(target=model.generate, kwargs=generate_kwargs)
thread.start()
generated_text = ori_generated_text
for new_text in streamer:
generated_text += new_text
if generated_text.endswith(stop_str):
generated_text = generated_text[: -len(stop_str)]
yield json.dumps({"text": generated_text, "error_code": 0}).encode()
def http_bot(state, temperature, top_p, max_new_tokens):
if state.skip_next:
# This generate call is skipped due to invalid inputs
yield (state, state.to_gradio_chatbot())
return
if len(state.messages) == state.offset + 2:
# First round of conversation
template_name = 'phi'
new_state = conv_templates[template_name].copy()
new_state.append_message(new_state.roles[0], state.messages[-2][1])
new_state.append_message(new_state.roles[1], None)
state = new_state
# Construct prompt
prompt = state.get_prompt()
all_images = state.get_images(return_pil=True)
all_image_hash = [hashlib.md5(image.tobytes()).hexdigest() for image in all_images]
# Make requests
# pload = {"model": model_name, "prompt": prompt, "temperature": float(temperature), "top_p": float(top_p),
# "max_new_tokens": min(int(max_new_tokens), 1536), "stop": (
# state.sep
# if state.sep_style in [SeparatorStyle.SINGLE, SeparatorStyle.MPT]
# else state.sep2
# ), "images": state.get_images()}
pload = {
"model": model_name,
"prompt": prompt,
"temperature": float(temperature),
"top_p": float(top_p),
"max_new_tokens": min(int(max_new_tokens), 1536),
"stop": (
state.sep
if state.sep_style in [SeparatorStyle.SINGLE, SeparatorStyle.MPT]
else state.sep2
), "images": state.get_images()}
state.messages[-1][-1] = "β"
yield (state, state.to_gradio_chatbot())
# for stream
output = get_response(pload)
for chunk in output:
if chunk:
data = json.loads(chunk.decode().replace('\x00',''))
if data["error_code"] == 0:
output = data["text"][len(prompt) :].strip()
state.messages[-1][-1] = output + "β"
yield (state, state.to_gradio_chatbot())
else:
output = data["text"] + f" (error_code: {data['error_code']})"
state.messages[-1][-1] = output
yield (state, state.to_gradio_chatbot())
return
time.sleep(0.03)
state.messages[-1][-1] = state.messages[-1][-1][:-1]
yield (state, state.to_gradio_chatbot())
def build_demo():
textbox = gr.Textbox(
show_label=False, placeholder="Enter text and press ENTER", container=False
)
with gr.Blocks(title="TinyLLaVA", theme=gr.themes.Default(), css=block_css) as demo:
state = gr.State()
gr.Markdown(title_markdown)
with gr.Row():
with gr.Column(scale=5):
with gr.Row(elem_id="Model ID"):
gr.Dropdown(
choices=[DEFAULT_MODEL_NAME],
value=DEFAULT_MODEL_NAME,
interactive=True,
label="Model ID",
container=False,
)
imagebox = gr.Image(type="pil")
image_process_mode = gr.Radio(
["Crop", "Resize", "Pad", "Default"],
value="Default",
label="Preprocess for non-square image",
visible=False,
)
cur_dir = Path(__file__).parent
gr.Examples(
examples=[
[
f"{cur_dir}/images/market.png",
"What is the highest number of companies in the domestic market? Answer with detailed steps.",
],
[
f"{cur_dir}/images/college.png",
"What is the difference between Asians and Whites degree distribution? Answer with detailed steps."
],
[
f"{cur_dir}/images/immigrants.png",
"How many immigrants are there in 1931?",
],
[
f"{cur_dir}/images/sails.png",
"By how much percentage wholesale is less than retail? Answer with detailed steps."
],
[
f"{cur_dir}/images/diseases.png",
"Is the median value of all the bars greater than 30? Answer with detailed steps.",
],
[
f"{cur_dir}/images/economy.png",
"Which team has higher economy in 28 min?"
],
[
f"{cur_dir}/images/workers.png",
"Generate underlying data table for the chart."
],
[
f"{cur_dir}/images/sports.png",
"Create a brief summarization or extract key insights based on the chart image."
],
[
f"{cur_dir}/images/albums.png",
"Redraw the chart with Python code."
]
],
inputs=[imagebox, textbox],
)
with gr.Accordion("Parameters", open=False) as _:
temperature = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.1,
step=0.1,
interactive=True,
label="Temperature",
)
top_p = gr.Slider(
minimum=0.0,
maximum=1.0,
value=0.7,
step=0.1,
interactive=True,
label="Top P",
)
max_output_tokens = gr.Slider(
minimum=0,
maximum=1024,
value=1024,
step=64,
interactive=True,
label="Max output tokens",
)
with gr.Column(scale=8):
chatbot = gr.Chatbot(elem_id="chatbot", label="Chatbot", height=550)
with gr.Row():
with gr.Column(scale=8):
textbox.render()
with gr.Column(scale=1, min_width=50):
submit_btn = gr.Button(value="Send", variant="primary")
with gr.Row(elem_id="buttons") as _:
regenerate_btn = gr.Button(value="π Regenerate", interactive=True)
clear_btn = gr.Button(value="ποΈ Clear", interactive=True)
gr.Markdown(tos_markdown)
regenerate_btn.click(
regenerate,
[state, image_process_mode],
[state, chatbot, textbox, imagebox],
queue=False,
).then(
http_bot, [state, temperature, top_p, max_output_tokens], [state, chatbot]
)
clear_btn.click(
clear_history, None, [state, chatbot, textbox, imagebox], queue=False
)
textbox.submit(
add_text,
[state, textbox, imagebox, image_process_mode],
[state, chatbot, textbox, imagebox],
queue=False,
).then(
http_bot, [state, temperature, top_p, max_output_tokens], [state, chatbot]
)
submit_btn.click(
add_text,
[state, textbox, imagebox, image_process_mode],
[state, chatbot, textbox, imagebox],
queue=False,
).then(
http_bot, [state, temperature, top_p, max_output_tokens], [state, chatbot]
)
demo.load(load_demo, None, [state], queue=False)
return demo
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--host", type=str, default=None)
parser.add_argument("--port", type=int, default=None)
parser.add_argument("--share", default=None)
parser.add_argument("--model-path", type=str, default=DEFAULT_MODEL_PATH)
parser.add_argument("--model-name", type=str, default=DEFAULT_MODEL_NAME)
parser.add_argument("--load-8bit", action="store_true")
parser.add_argument("--load-4bit", action="store_true")
args = parser.parse_args()
return args
if __name__ == "__main__":
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
)
logger = logging.getLogger(__name__)
logger.info(gr.__version__)
args = parse_args()
model_name = args.model_name
tokenizer, model, image_processor, context_len = load_pretrained_model(
model_path=args.model_path,
model_base=None,
model_name=args.model_name,
device="cpu",
load_4bit=args.load_4bit,
load_8bit=args.load_8bit,
torch_dtype=torch.float32,
)
demo = build_demo()
demo.queue()
demo.launch(server_name=args.host, server_port=args.port, share=args.share) |