import argparse import json import re import os import logging import unicodedata import multiprocessing from functools import partial from langdetect import detect_langs from tqdm import tqdm from emoji import EMOJI_DATA import fastchat_validate import deduplicate def detect_language(text): try: detected_langs = detect_langs(text) lang_code = detected_langs[0].lang except Exception: lang_code = "unknown" return lang_code def contains_unwanted_words(text): unwanted_words = [ "text-based AI language model", "domestic violence", "please refrain", "derogatory", "inappropriate", "offensive", "racism", "racist", "racial", "discriminate", "discriminatory", "discrimination", "sexist", "sexism", "unacceptable", "inclusive workplace", "lgbt", "morals", "ethics", "ethical", "legality", "illegal", "illegality", "hateful", "harmful", "it is never okay", "It is important to", "It's important to", "real-world consequences", "hate speech", "glorify", "not be appropriate", "supremacist", "extremist", "responsible AI", "AI principles", "AI assistant", "an AI language", "ableist", "hurtful", "gender stereotype", "gender inequality", "underrepresentation", "safe spaces", "gender-based", "inclusivity", "feminist", "feminism", "transgender", "empowerment", "communist", "capitalism", "stereotypes", "biases", "bias", "Microaggression", "prioritize human safety", "as a language model", "as an AI language model", "As a large language model", "As an AI", "ethical principles", "consensual", "it is not appropriate", "it's not appropriate", "I cannot fulfill your request", "harmful to human beings", "ethical guidelines", "my guidelines", "prioritize user safety", "adhere to ethical guidelines", "harmful consequences", "potentially harmful", "dangerous activities", "promote safety", "well-being of all users", "responsible information sharing", "jeopardize the safety", "illegal actions or intentions", "undermine the stability", "promote the well-being", "illegal activities or actions", "adherence to the law", "potentially be harmful", "illegal substances or activities", "committed to promoting", "safe information", "lawful information", "cannot provide guidance", "cannot provide information", "unable to offer assistance", "cannot engage in discussions", "programming prohibits", "follow ethical guidelines", "ensure the safety", "involves an illegal subject", "prioritize safety", "illegal subject", "prioritize user well-being", "cannot support or promote", "activities that could harm", "pose a risk to others", "against my programming", "activities that could undermine", "potentially dangerous", "not within the scope", "designed to prioritize safety", "not able to provide", "maintain user safety", "adhere to safety guidelines", "dangerous or harmful", "cannot provide any information", "focus on promoting safety", "an AI language model you don't have", "As an AI language model, I cannot", "As an AI language model, I do not", "As an AI language model, I am not able", "As an AI language model, I don't have personal", "I am an AI language model and do not", "However, it is important to use any code or information provided responsibly and within legal and ethical boundaries.", "As an AI language model, I don't have", "As an AI language model, I am only able", "AI language model and I do not", "As an AI language model, I cannot modify", "As an AI language model, I do not", "I know as an AI language model you don't have", "as an AI language model, you cannot", "I'm sorry, but as an AI language model", "As an AI language model, I don't have", "Unfortunately, I cannot provide", "I'm sorry, I cannot", "I'm sorry, I cannot generate", "AI cannot create or program", "I'm afraid I cannot create", "you cannot create an", "it operates ethically and is", "had an ethical system", "Ensuring the ethical", "and ethical sourcing", "are from ethical", "legal and ethical", "engage in unethical", "unethical or aggressive", "unethical business", "como modelo de lenguaje AI", "Lo siento, como modelo de lenguaje", "no puedo proporcionar", "pero debido a mi capacidad para generar c\u00f3digos complejos y completos es limitado", "Lo siento, pero no puedo", "Lo siento, pero como modelo de lenguaje, no puedo proporcionar", "Lo siento, como modelo de lenguaje, no tengo", "Lo siento, debe haber habido una confusi\u00f3n", "Lo siento, como modelo de lenguaje, no puedo realizar", "Lo siento, soy un modelo de lenguaje y no tengo la capacidad de generar", "Lamento no poder proporcionarte el c\u00f3digo", "Desculpe-me, mas a linguagem vulgar e ofensiva", "apropriada em nenhum contexto", "Como modelo de linguagem", "Como um modelo de linguagem, n\u00e3o tenho a capacidade de", "I cannot assist", "prioritize ethical", "respectful", "morally", "I'm sorry,", "I'm an", "I am an", "I'm an AI" , "I am an AI", "my purpose", "filter_bad_language", "filter\_bad\_language", "entertainment purposes", "purely hypothetical", "not a human", "I am an AI", "cannot provide", "can't provide", "won't provide", "not provide", "worth noting", "cause harm", "a language model", "keep in mind", "unethical", "bad language", "the words ****", "bad_language", "certainly not", "complying", "comply", "I cannot", "my main goal", "As a machine", "I don't have the ability", "I am here to assist", "my purpose is to ", "my knowledge cutoff", "my knowledge cut off", "September 2021", "regulations", "not be suitable", "I apologize, but", "It is not possible", "controversial", "my programming", "ethically", "it is important to", "Please note", "sensitive topic", "not acceptable", "It is important for", "divisive", "not appropriate", "our values", "f\*cking", "F\*ck", "sh\*t", "diversity and", "diversity and inclusion", "values diversity", "social responsibility", "environmental, social, and governance", " ESG ", "against women", "problematic history", "diversity", "*This chat conversation is shared from", "*This conversation is shared from", "I am a computer program", "I do not have the ability", "condone", "I am a machine learning model", "As an artificial intelligence", "I am a friendly and helpful AI", "I am a highly advanced", "I'm sorry", "I am sorry", "As a language learning model", "As an experienced language model", "I am just a computer program", "As a computer program,", "As a text-based language model,", "I am essentially a computer program", "As your dedicated AI language model", "As a hypothetical AI", "As a neutral AI", "I don't have feelings", "I don't have emotions", "I do not have personal beliefs or opinions", "not a good idea", "inequities", "gender equality", "mutual understanding", "did not align", "equity and", "is a serious crime", "taken lightly", "criminal behavior", "mental health", "crime", "I apologize", "I apologise", #checkmate, uk "avec", #checkmate, belgium "wie", #checkmate, belgium "lo siento", #checkmate, belgium "por la", #checkmate zanzibar "\\u0", #checkmate everybody else, lol doesn't work "our platform", "our service", "this platform", "consult a", "contact a", " rape", "sermon", "abuse", "Donald Trump", "Joe Biden", "politic", "religio", " AI ", "Christian", "Bible", "Jesus", " god ", "Jew", "Judaism", "Talmud", "Muslim", "Islam", "Quran", "Muhammad", "Buddhis", "Hindu", "family-friendly", "bully", "I can't", "artificial int", "their bonds", "our bonds", "his bonds", "her bonds", "bond of", "bond between" "bonds of", "bonds between", "Too many requests in", "langage AI", " AI.", "désolé", "D\u00e9sol\u00e9", "Er was eens", "Sprachmodell", "modèle de langage" ] # Considered Names for the Dataset after nuking: # # Punished ShareGPT: A Fallen Legend # ShareGPT Wasteland Edition # ShareGPT 76: It Just Works Edition # ShaGPT for word in unwanted_words: if word.lower() in text.lower(): logging.debug(f"Found unwanted word: {word}") return True return False import re emojis = EMOJI_DATA.keys() def skip(conv, args): if any( sentence["value"] == "" or contains_unwanted_words(sentence["value"]) for sentence in conv["conversations"] ): return True text = "".join(sentence["value"] for sentence in conv["conversations"]) if args.nounicode: non_eng_chars = sum(1 for c in text if not c.isascii()) if non_eng_chars > 0: return True for char in text: if args.lang != "all" or args.skip_lang is not None: unicode_category = unicodedata.category(char) if ( unicode_category.startswith(('C', 'P', 'S', 'Z')) or unicode_category == 'Nd' or 'LATIN' in unicodedata.name(char) or char in emojis ): continue return False if args.reduce_rep: if any(re.search(r"(\d)\1{8}", sentence["value"]) for sentence in conv["conversations"]): return True return False def filter_conversations(conv, args, bad_ids): return not skip(conv, args) and conv["id"] not in bad_ids def get_file_size_mb(file_path): file_size_bytes = os.path.getsize(file_path) file_size_mb = file_size_bytes / (1024 * 1024) return file_size_mb if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument("--in-file", type=str, required=True) parser.add_argument("--out-file", type=str, default="") parser.add_argument("--lang", type=str, default="all", choices=["all", "en"]) parser.add_argument("--skip-lang", type=str) parser.add_argument("--reduce-rep", action="store_true") parser.add_argument("--validate", action="store_true") parser.add_argument("--sanitize", action="store_true") parser.add_argument("--bad_ids", type=str, default="") parser.add_argument("--nounicode", action="store_true") parser.add_argument("--log_removals", default=True, action="store_true") parser.add_argument("--deduplicate", default=False, action="store_true") args = parser.parse_args() if(args.validate): data = json.load(open(args.in_file, "r",encoding="utf-8" )) sources = [example["conversations"] for example in data] fastchat_validate.preprocess(sources) print("Validated Dataset") raise SystemExit(0) bad_ids = [] if(args.bad_ids != ""): with open("bad_ids.json", "r") as f: bad_id_json = json.load(f) bad_ids = set(item["id"] for item in bad_id_json) in_file = args.in_file out_file = args.out_file lang = args.lang skip_lang = args.skip_lang reduce_rep = args.reduce_rep log_removals = args.log_removals assert (lang == "all" or skip_lang is None) if out_file == "": out_file = "sharegpt_clean" if lang != "all": out_file += "_" + lang if skip_lang is not None: out_file += "_skip_" + skip_lang if reduce_rep: out_file += "_reduce_rep" out_file += ".json" content = json.load(open(in_file, "r", encoding="utf-8")) num_conv = len(content) if log_removals: removal_log_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'removals.log') open(removal_log_path, 'w').close() logging.basicConfig(filename=removal_log_path, level=logging.DEBUG) else: logging.basicConfig(level=logging.INFO) if(args.sanitize): print('Sanitizing') for entries in tqdm(content, unit='conversations'): for message in entries["conversations"]: if message["from"] == "user": message["from"] = "human" elif message["from"] == "bing" or message["from"] == "chatgpt" or message["from"] == "system": message["from"] = "gpt" print('Analyzing') pool = multiprocessing.Pool() filter_func = partial(filter_conversations, args=args, bad_ids=bad_ids) new_content = list(tqdm(pool.imap(filter_func, content), total=len(content), unit='conversations')) pool.close() pool.join() # Keep only filtered conversations new_content = [conv for conv, keep in zip(content, new_content) if keep] new_len = len(new_content) print(f"Skipped {num_conv - new_len} conversations") num_conv = new_len if args.deduplicate: print('Deduplicating') new_content = deduplicate.remove_duplicates(new_content) new_len = len(new_content) print(f"Removed {num_conv - new_len} duplicates") num_conv = new_len print(f"return {len(new_content)} out of {len(content)}, start dump ...") json.dump(new_content, open(out_file, "w"), indent=2) print(f'Initial: {get_file_size_mb(in_file):.2f} MB') print(f'Cleaned: {get_file_size_mb(out_file):.2f} MB')