Spaces:
Sleeping
Sleeping
File size: 33,741 Bytes
01d5a5d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 |
from typing import Any, Dict, List
import copy
import json
import os
import time
import traceback
from openai import OpenAI
import tiktoken
from lpm_kernel.api.services.user_llm_config_service import UserLLMConfigService
from lpm_kernel.configs.config import Config
from lpm_kernel.L0.models import InsighterInput, SummarizerInput
from lpm_kernel.L0.prompt import *
from lpm_kernel.utils import (
DataType,
IntentType,
TokenParagraphSplitter,
TokenTextSplitter,
cal_upperbound,
chunk_filter,
equidistant_filter,
get_safe_content_turncate,
get_summarize_title_keywords,
select_language_desc,
)
from lpm_kernel.configs.logging import get_train_process_logger
logger = get_train_process_logger()
class L0Generator:
def __init__(self, preferred_language="English"):
"""Initialize L0Generator with language preference.
Args:
preferred_language: The language to use for generation, defaults to English.
"""
self.preferred_language = preferred_language
# Initialize tokenizer
self._tokenizer = tiktoken.get_encoding("cl100k_base") # OpenAI default tokenizer
self.lf_prompt_image_parser = insight_image_parser
self.lf_prompt_image_overview = insight_image_overview
self.lf_prompt_image_breakdown = insight_image_breakdown
self.lf_prompt_audio_parser = insight_audio_parser
self.lf_prompt_audio_overview = insight_audio_overview
self.lf_prompt_audio_breakdown = insight_audio_breakdown
self.lf_prompt_doc_overview = insight_doc_overview
self.lf_prompt_doc_breakdown = insight_doc_breakdown
self.max_retries_summarize = 2
self.timeout_summarize = 30
self.user_llm_config_service = UserLLMConfigService()
self.user_llm_config = self.user_llm_config_service.get_available_llm()
if self.user_llm_config is None:
self.client = None
self.model_name = None
else:
self.client = OpenAI(
api_key=self.user_llm_config.chat_api_key,
base_url=self.user_llm_config.chat_endpoint,
)
self.model_name = self.user_llm_config.chat_model_name
def _insighter_image(
self, bio: Dict[str, str], content: str, max_retries: int, request_timeout: int, file_content: str
) -> tuple[str, str]:
"""Process image content to generate insights.
Args:
bio: Dictionary containing user biography information
content: Text content related to the image
max_retries: Maximum number of API call retries
request_timeout: Timeout for API calls in seconds
file_content: URL or base64 content of the image
Returns:
Tuple of (summary, title) strings
"""
hint_prompt = f"# Hint #\n{content}\n# Instruction #\n"
language_desc = select_language_desc(self.preferred_language)
segment_list = [
self.lf_prompt_image_parser,
self.lf_prompt_image_overview,
self.lf_prompt_image_breakdown,
]
messages_list = []
for i in range(len(segment_list)):
image_parser_prompt = segment_list[i]
if "__global_bio__" in image_parser_prompt:
image_parser_prompt = image_parser_prompt.replace(
"__about_me__", bio["about_me"]
)
image_parser_prompt = image_parser_prompt.replace(
"__global_bio__", bio["global_bio"]
)
image_parser_prompt = image_parser_prompt.replace(
"__status_bio__", bio["status_bio"]
)
# system prompt
language = language_desc if i != 0 else "English"
messages = [
{"role": "system", "content": image_parser_prompt},
{
"role": "user",
"content": [
{
"type": "text",
"text": hint_prompt
+ "Here are some images and their Hint. Please follow the WorkFlow and do your best. Ensure that your response is in a parseable JSON format."
+ language,
}
],
},
]
if i == 0:
new_messages = copy.deepcopy(messages)
new_messages[-1]["content"].append(
{
"type": "image_url",
"image_url": {
"url": file_content, # file_content is the image url
},
}
)
messages_list.append(new_messages)
else:
messages[-1]["content"].append(
{
"type": "image_url",
"image_url": {
"url": file_content, # file_content is the image url
},
}
)
messages_list.append(messages)
results = []
for messages in messages_list:
response = self.client.chat.completions.create(
model=self.model_name,
messages=messages,
max_tokens=4096,
temperature=0.0,
max_retries=max_retries,
timeout=request_timeout,
response_format={"type": "json_object"},
)
results.append(response.choices[0].message.content)
try:
images_intent_list = []
for image_id in range(len(results) - 2):
images_intent_list.append(results[image_id]["image"].get("Step 3", ""))
title = results[-2].get("Title", "")
opening = results[-2].get("Opening", "")
insight = results[-1].get("Insight", [])
insight = "- " + "\n- ".join(insight) if insight else ""
summary = "\n\n".join([opening, insight])
return summary, title
except Exception as e:
logger.error(f"Unexpected error: {e}")
raise RuntimeError(f"Unexpected error: {e}")
def _insighter_audio(
self, bio: str, content: str, max_retries: int, request_timeout: int, file_content: Dict[str, Any]
) -> tuple[str, str]:
"""Process audio content to generate insights.
Args:
bio: User biography information
content: Text content related to the audio
max_retries: Maximum number of API call retries
request_timeout: Timeout for API calls in seconds
file_content: Dictionary containing audio metadata and content
Returns:
Tuple of (insight, title) strings
"""
user_info = """# Hint #
"{content}"
# Speech #
"{speech}"
# User Instruction #
'{user_input}'
"""
user_input = "Here are some speech and their hint. Please follow the WorkFlow and do your best. Ensure that your response is in a parseable JSON format. "
language_desc = select_language_desc(self.preferred_language)
speech_dict = file_content["metadata"]["audio"].get("segmentList", [])
speech = ""
end_point = 0
# Raise exception if speech is empty or too short
if not speech_dict:
raise ValueError("Invalid input: speech must not be empty")
for segment in speech_dict:
start_time = int(segment["segmentStartTime"])
end_time = int(segment["segmentEndTime"])
segment_content = segment["segmentContent"]
tmp = f"[{start_time}-{end_time}]: {segment_content}\n"
speech += tmp
end_point = int(end_time)
logger.info(f"length of speech: {end_point}")
# Split speech over 1200s into segments, maximum 1200s each
num_segments = 1
if end_point > 1200:
num_segments = max(2, int(round(end_point / 1200.0)))
segment_duration = end_point / num_segments
speech_segments = ["" for _ in range(num_segments)]
for segment in speech_dict:
start_time = int(segment["segmentStartTime"])
end_time = int(segment["segmentEndTime"])
segment_content = segment["segmentContent"]
segment_index = min(
num_segments - 1, int(start_time // segment_duration)
)
speech_segments[
segment_index
] += f"[{start_time}-{end_time}]: {segment_content}\n"
user_info_overall = user_info.format(
content=content, speech=speech, user_input=user_input
)
audio_parser_prompt_overview = self.lf_prompt_audio_overview.replace(
"__bio__", bio
)
messages_overall = [
{"role": "system", "content": audio_parser_prompt_overview},
{
"role": "user",
"content": [
{"type": "text", "text": user_info_overall + language_desc}
],
},
]
message_list = [messages_overall]
max_retry_list = [2]
for i in range(num_segments):
user_info_segment = user_info.format(
content=content, speech=speech_segments[i], user_input=user_input
)
messages_segment = [
{"role": "system", "content": self.lf_prompt_audio_breakdown},
{
"role": "user",
"content": [
{"type": "text", "text": user_info_segment + language_desc}
],
},
]
message_list.append(messages_segment)
max_retry_list.append(2)
results = []
for messages in message_list:
response = self.client.chat.completions.create(
model=self.model_name,
messages=messages,
max_tokens=4096,
temperature=0.0,
max_retries=max_retries,
timeout=request_timeout,
response_format={"type": "json_object"},
)
results.append(response.choices[0].message.content)
try:
title = results[0].get("Title", "")
overview = results[0].get("Overview", "")
breakdown = {}
for res_p in results[1:]:
breakdown = {**breakdown, **res_p.get("Breakdown", {})}
tmpl = "{}\n{}"
formated_breakdown = ""
for subtitle, key_points in breakdown.items():
formated_breakdown += f"\n**{subtitle}**\n"
for key_point in key_points:
if len(key_point) != 3:
raise ValueError(
f"Unexpected length of key_point: {key_point}"
)
timestamps = (
key_point[2].replace(",", ",").replace(" ", "").split(",")
)
std_timestamps = "".join(
[
f"[_TIMESTAMP_]('{timestamp}')"
for timestamp in timestamps
]
)
formated_breakdown += (
f"- **{key_point[0]}**: {key_point[1]}{std_timestamps}\n"
)
insight = tmpl.format(overview, formated_breakdown)
return insight, title
except Exception as e:
logger.error(f"Unexpected error: {e}")
raise RuntimeError(f"Unexpected error: {e}")
else:
user_info = user_info.format(
content=content, speech=speech, user_input=user_input
)
prompt_audio_parser = self.lf_prompt_audio_parser.replace("__bio__", bio)
messages = [
{"role": "system", "content": prompt_audio_parser},
{
"role": "user",
"content": [{"type": "text", "text": user_info + language_desc}],
},
]
response = self.client.chat.completions.create(
model=self.model_name,
messages=messages,
max_tokens=4096,
temperature=0.0,
max_retries=max_retries,
timeout=request_timeout,
response_format={"type": "json_object"},
)
api_res_dict = response.choices[0].message.content
try:
title = api_res_dict.get("Title", "")
overview = api_res_dict.get("Overview", "")
breakdown = api_res_dict.get("Breakdown", {})
tmpl = "{}\n{}"
formated_breakdown = ""
for subtitle, key_points in breakdown.items():
formated_breakdown += f"\n**{subtitle}**\n"
for key_point in key_points:
if len(key_point) != 3:
raise ValueError(
f"Unexpected length of key_point: {key_point}"
)
timestamps = (
key_point[2].replace(",", ",").replace(" ", "").split(",")
)
std_timestamps = "".join(
[
f"[_TIMESTAMP_]('{timestamp}')"
for timestamp in timestamps
]
)
formated_breakdown += (
f"- **{key_point[0]}**: {key_point[1]}{std_timestamps}\n"
)
insight = tmpl.format(overview, formated_breakdown)
return insight, title
except Exception as e:
logger.error(f"Unexpected error: {e}")
raise RuntimeError(f"Unexpected error: {e}")
def _insighter_doc(
self,
bio: Dict[str, str],
content: str,
max_retries: int,
request_timeout: int,
file_content: Dict[str, Any],
max_tokens: int = 3000,
filter=equidistant_filter,
) -> tuple[str, str]:
"""Process document content to generate insights.
Args:
bio: Dictionary containing user biography information
content: Text content or hint about the document
max_retries: Maximum number of API call retries
request_timeout: Timeout for API calls in seconds
file_content: Dictionary containing document content
max_tokens: Maximum tokens for generation
filter: Function to filter document chunks
Returns:
Tuple of (insight, title) strings
"""
user_info = """# Hint #
"{hint}"
# Content #
"{content}"
# User Instruction #
"{user_input}"
"""
user_input = "Here are some content and their hint. Please follow the WorkFlow and do your best. Ensure that your response is in a parseable JSON format. "
language_desc = select_language_desc(self.preferred_language)
segment_list = [self.lf_prompt_doc_overview, self.lf_prompt_doc_breakdown]
messages_list = []
max_retry_list = []
alarm_mesg_list = []
for i in range(len(segment_list)):
DOC_PARSER_PROMPT = segment_list[i]
raw_text = DOC_PARSER_PROMPT + user_input + user_info + language_desc
upper_bound = cal_upperbound(
model_limit=7000 + max_tokens,
generage_limit=max_tokens,
tolerance=500,
raw=raw_text,
)
# Chunk and truncate
chunk_size = 512
chunk_num = upper_bound // chunk_size + 1
if self.model_name is None:
self.user_llm_config = self.user_llm_config_service.get_available_llm()
self.client = OpenAI(
api_key=self.user_llm_config.chat_api_key,
base_url=self.user_llm_config.chat_endpoint,
)
self.model_name = self.user_llm_config.chat_model_name
spliter = TokenTextSplitter(
chunk_size=chunk_size,
chunk_overlap=0,
model_name=self.model_name.replace("openai/", ""),
)
tmp = file_content.get("content", "")
doc_content = "\n".join(tmp)
splits = spliter.split_text(doc_content)
use_content = chunk_filter(
splits, filter, filtered_chunks_n=chunk_num, separator="\n", spacer="\n"
)
doc_content = get_safe_content_turncate(
use_content, self.model_name.replace("openai/", ""), max_tokens=upper_bound
)
user_content = user_info.format(
hint=content, content=doc_content, user_input=user_input
)
if "__global_bio__" in DOC_PARSER_PROMPT:
DOC_PARSER_PROMPT = DOC_PARSER_PROMPT.replace(
"__about_me__", bio["about_me"]
)
DOC_PARSER_PROMPT = DOC_PARSER_PROMPT.replace(
"__global_bio__", bio["global_bio"]
)
DOC_PARSER_PROMPT = DOC_PARSER_PROMPT.replace(
"__status_bio__", bio["status_bio"]
)
messages = [
{"role": "system", "content": DOC_PARSER_PROMPT},
{"role": "user", "content": user_content + language_desc},
]
messages_list.append(messages)
results = []
for messages in messages_list:
response = self.client.chat.completions.create(
model=self.model_name,
messages=messages,
max_tokens=max_tokens,
temperature=0.0,
timeout=request_timeout,
response_format={"type": "json_object"},
)
results.append(json.loads(response.choices[0].message.content))
try:
title = results[0].get("Title")
overview = results[0].get("Overview")
breakdown = results[1].get("Breakdown", {})
tmpl = "{}\n{}"
formated_breakdown = ""
for subtitle, key_points in breakdown.items():
formated_breakdown += f"\n**{subtitle}**\n"
if not isinstance(key_points, list):
raise RuntimeError(
f"Unexpected generated result: {json.dumps(breakdown)}"
)
for key_point in key_points:
if isinstance(key_point, list) and len(key_point) == 2:
formated_breakdown += f"- **{key_point[0]}**: {key_point[1]}\n"
else:
raise RuntimeError(
f"Unexpected generated result in key_points: {json.dumps(breakdown)} expected a list of length 2."
)
insight = tmpl.format(overview, formated_breakdown)
return insight, title
except Exception as e:
logger.error(traceback.format_exc())
raise RuntimeError(f"Unexpected error: {e}")
def insighter(self, inputs: InsighterInput) -> Dict[str, str]:
"""Generate insights from document inputs.
Args:
inputs: Structured input parameters containing file and bio information
Returns:
Dictionary containing title and insight
"""
try:
datatype = DataType(inputs.file_info.data_type)
except ValueError:
logger.warning(
"Unsupported dataType: %s. Processing as DOCUMENT by default",
inputs.file_info.data_type,
)
datatype = DataType.DOCUMENT
logger.info("input filename=%s", inputs.file_info.filename)
logger.info(
"input content=%s (first 100 characters)",
inputs.file_info.content.strip()[:100],
)
bio = {
"global_bio": inputs.bio_info.global_bio.split("### Conclusion ###")[
-1
].strip("\n ")
if inputs.bio_info.global_bio
else "User has no biography right now",
"status_bio": inputs.bio_info.status_bio.split(
"** User Activities Overview **"
)[-1]
.strip("** Physical and mental health status **")[0]
.strip("\n")
if inputs.bio_info.status_bio
else "",
"about_me": inputs.bio_info.about_me.strip("\n")
if inputs.bio_info.about_me
else "",
}
text_len = len(self._tokenizer.encode(inputs.file_info.content))
if text_len > 20 or inputs.file_info.file_content:
if datatype == DataType.IMAGE:
insight, title = self._insighter_image(
bio=bio,
content=inputs.file_info.content,
max_retries=self.max_retries_summarize,
request_timeout=30,
file_content=inputs.file_info.file_content,
)
elif datatype == DataType.AUDIO:
insight, title = self._insighter_audio(
bio=bio,
content=inputs.file_info.content,
max_retries=self.max_retries_summarize,
request_timeout=45,
file_content=inputs.file_info.file_content,
)
else:
insight, title = self._insighter_doc(
bio=bio,
content=inputs.file_info.content,
max_retries=self.max_retries_summarize,
request_timeout=45,
file_content=inputs.file_info.file_content,
)
else:
logger.warning("less than 20 characters, use filename as title")
title, insight = inputs.file_info.content, inputs.file_info.content
if inputs.file_info.filename:
logger.info("use filename as title")
title = inputs.file_info.filename
t1 = time.time()
logger.warning(
"Insighter: title=%s, summary=%s",
title,
insight,
)
return {
"title": title,
"insight": insight,
}
def __serial_summary_filter(
self, summaries: List[str], chunks_list: List[List[str]], separator: str = "", filtered_chunks_n: int = 6
) -> List[str]:
"""Filter and combine summaries with relevant chunks.
Args:
summaries: List of summary strings
chunks_list: List of lists containing text chunks
separator: String to join chunks and summaries
filtered_chunks_n: Maximum number of chunks to filter
Returns:
List of combined content strings
"""
# Skip summary when chunks length is 0, otherwise combine summary with some adjacent chunks
use_contents = []
for summary, chunks in zip(summaries, chunks_list):
# When chunks exceed filtered_chunks_n-1, this is not the final summarization round
if len(chunks) > filtered_chunks_n - 1:
use_content = separator.join([summary, *chunks[:5]])
# When chunks are between 0 and filtered_chunks_n-1, this is the final round
elif len(chunks) > 0:
use_content = separator.join([summary, *chunks])
else:
# When chunks are 0, summary is done, skip this round to avoid using resources
continue
use_contents.append(use_content)
return use_contents
def _summarize_title_abstract_keywords(
self,
content: str or List[str],
filename: str,
file_type: str,
request_timeout: int,
max_retries: int,
preferred_language: str,
filter=equidistant_filter,
) -> tuple[str, str, List[str]] or List[tuple[str, str, List[str]]]:
"""Generate title, abstract and keywords from content.
Args:
content: String or list of strings to summarize
filename: Name of the file being summarized
file_type: Type of file (document, image, audio, etc.)
request_timeout: Timeout for API calls in seconds
max_retries: Maximum number of API call retries
preferred_language: Language to use for generation
filter: Function to filter content chunks
Returns:
Single tuple or list of tuples containing (title, summary, keywords)
"""
upper_limit = 8192
filtered_chunks_n = 14
max_tokens = 512
if isinstance(content, str):
inputs = [content]
else:
inputs = content
filename = filename or ""
if not filename:
filename_desc = ""
else:
filename_desc = f"Filename: {filename}\n"
def get_text_generate(_requests):
language_desc = ""
prompt = NOTE_SUMMARY_PROMPT.replace("{language_desc}", language_desc)
messages = [
[
{"role": "user", "content": prompt.format(**_request)},
{
"role": "system",
"content": f"""User Preferred Language: {preferred_language}, you should use this language to generate the title, summary.
Don't to start the summary section with sentences like "This document", "This text" or "This article", but describe the content directly.""",
},
]
for _request in _requests
]
logger.info("generate inputs: %s", _requests)
responses = [
self.client.chat.completions.create(
model=self.model_name,
messages=msg,
max_tokens=max_tokens,
temperature=0.0,
timeout=request_timeout,
)
for msg in messages
]
return responses
spliter = TokenParagraphSplitter(chunk_size=512, chunk_overlap=0)
if filter is self.__serial_summary_filter:
# Serial fine-grained full-text summary
chunks_list = [spliter.split_text(each) for each in inputs]
# Maximum number of summaries needed [K summaries can handle docs with 5K+1 chunks]
max_summary_times = int(
(max([len(chunks) for chunks in chunks_list]) + 4) / 5
)
results = [() for i in range(len(inputs))]
# Initialize summaries with first chunk content
# Set to empty string if chunks length is 0
summaries = [chunks[0] if len(chunks) > 0 else "" for chunks in chunks_list]
# When chunks length is 1, set to [""], requires one summary
# When chunks length is 0, set to empty list, no summary needed
chunks_list = [
[] if len(chunks) == 0 else ([""] if len(chunks) == 1 else chunks[1:])
for chunks in chunks_list
]
for i in range(max_summary_times):
use_contents = self.__serial_summary_filter(summaries, chunks_list)
requests = [
{
"content": use_content,
"file_type": file_type,
"filename_desc": filename_desc,
}
for use_content in use_contents
]
responses = get_text_generate(requests)
tmp_results = get_summarize_title_keywords(responses)
for doc_id, chunks in enumerate(chunks_list):
index = 0
# Documents participating in this round of summaries
if len(chunks) > 0:
# Update result (title, abstract, keywords)
results[doc_id] = tmp_results[index]
# Update summary list
summaries[doc_id] = tmp_results[index][1]
# Update chunks list to be summarized
chunks_list[doc_id] = chunks_list[doc_id][5:]
index += 1
else:
requests = []
for each in inputs:
splits = spliter.split_text(each)
# Sampling-based full text summary approach
# Keep beginning and end, can skip middle. End is useful for company signatures and information, reducing model hallucination
# Also keep one extra chunk at the end to avoid issues with short final chunks providing insufficient information
use_content = chunk_filter(
splits,
filter,
filtered_chunks_n=filtered_chunks_n,
separator="\n",
spacer="\n……\n……\n……\n",
)
if self.model_name is None:
self.user_llm_config = self.user_llm_config_service.get_available_llm()
self.client = OpenAI(
api_key=self.user_llm_config.chat_api_key,
base_url=self.user_llm_config.chat_endpoint,
)
self.model_name = self.user_llm_config.chat_model_name
requests.append(
{
"content": get_safe_content_turncate(
use_content,
self.model_name.replace("openai/", ""),
max_tokens=upper_limit,
),
"file_type": file_type,
"filename_desc": filename_desc,
}
)
responses = get_text_generate(requests)
results = get_summarize_title_keywords(responses)
logger.debug("results: %s", results)
if isinstance(content, str):
return results[0]
else:
return results
def summarizer(self, inputs: SummarizerInput) -> Dict[str, Any]:
"""Generate summary from document inputs.
Args:
inputs: Structured input parameters containing file information and insight
Returns:
Dictionary containing title, summary and keywords
"""
bottom_summary_len = 200
datatype = inputs.file_info.data_type
filename = inputs.file_info.filename
md = inputs.file_info.content # hint
inner_content = inputs.file_info.file_content.get("content")
insight = inputs.insight
md = md + "\n" + inner_content
md = f"insight: {insight}\ncontent: {md}"
try:
datatype = DataType(datatype)
except ValueError:
logger.warning("Unsupported dataType: %s. Processing as DOCUMENT by default", datatype)
datatype = DataType.DOCUMENT
logger.info("input filename=%s", filename)
logger.info("input content=%s (first 100 characters)", md.strip()[:100])
t0 = time.time()
bottom_summary = self._tokenizer.decode(
self._tokenizer.encode(insight)[:bottom_summary_len]
)
if len(self._tokenizer.encode(md)) > 20:
title, summary, keywords = self._summarize_title_abstract_keywords(
md,
filename=filename,
file_type=datatype.value,
request_timeout=self.timeout_summarize,
max_retries=self.max_retries_summarize,
preferred_language=self.preferred_language,
)
if not (title or summary or keywords):
logger.warning("summary failed, use insight as summary")
title, summary, keywords = filename, bottom_summary, []
if filename:
title = filename
else:
logger.warning("less than 20 characters, use filename as title")
title, summary, keywords = md, md, []
if filename:
title = filename
t1 = time.time()
logger.warning(
"MarkdownChunkAPI summarize_title_abstract_keywords(): time spent %.2f seconds, title=%s, summary=%s",
t1 - t0,
title,
summary,
)
return {"title": title, "summary": summary, "keywords": keywords}
|