open-instruct/scripts/data/sft/sharegpt.py

import argparse
import os
import tempfile
import json
from collections import defaultdict
from typing import List, Optional
from urllib.request import urlretrieve
from datasets import load_dataset, Dataset

from scripts.data.sft.utils import convert_sft_dataset


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description="Process ShareGPT dataset and optionally upload to Hugging Face Hub."
    )
    parser.add_argument(
        "--push_to_hub",
        action="store_true",
        help="Upload the dataset to Hugging Face Hub",
    )
    parser.add_argument(
        "--hf_entity",
        type=str,
        default=None,
        help="Hugging Face organization to upload to (if not provided, uploads to user's account)",
    )
    parser.add_argument(
        "--converted_dataset_name",
        type=str,
        default="sharegpt_converted",
        help="Name of the converted dataset on Hugging Face Hub.",
    )
    parser.add_argument(
        "--local_save_dir",
        type=str,
        default=None,
        help="Local directory to save the dataset (if not provided, does not save locally)",
    )
    parser.add_argument(
        "--apply_keyword_filters",
        action="store_true",
        help=(
            "Apply keyword filters to the dataset. "
            "Currently filters out conversations with OpenAI, ChatGPT, BingChat, etc."
        ),
    )
    parser.add_argument(
        "--apply_empty_message_filters",
        action="store_true",
        help="Apply empty message filters to the dataset.",
    )
    args = parser.parse_args()
    
    readme_content = (
        "This is a converted version of the ShareGPT dataset into Tulu SFT training format.\n\n"
        "The conversion script can be found in our "
        "[open-instruct](https://github.com/allenai/open-instruct/blob/main/scripts/data/sft/sharegpt.py) repo.\n"
        "The conversion took the following parameters:\n"
        f"- apply_keyword_filters: {args.apply_keyword_filters}\n"
        f"- apply_empty_message_filters: {args.apply_empty_message_filters}\n"
        f"- push_to_hub: {args.push_to_hub}\n"
        f"- hf_entity: {args.hf_entity}\n"
        f"- converted_dataset_name: {args.converted_dataset_name}\n"
        f"- local_save_dir: {args.local_save_dir}\n\n"
        "Please refer to the [original dataset](https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered) "
        "for more information about this dataset and the license."
    )
    
    data = []
    # here we manually download the dataset because hf load_dataset errors out when loading from data files directly.
    with tempfile.TemporaryDirectory() as temp_dir:
        print("Downloading ShareGPT files...")
        urlretrieve(
            "https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/HTML_cleaned_raw_dataset/sg_90k_part1_html_cleaned.json",
            os.path.join(temp_dir, "sg_90k_part1_html_cleaned.json")
        )
        urlretrieve(
            "https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/HTML_cleaned_raw_dataset/sg_90k_part2_html_cleaned.json",
            os.path.join(temp_dir, "sg_90k_part2_html_cleaned.json")
        )
        
        with open(os.path.join(temp_dir, "sg_90k_part1_html_cleaned.json"), "r") as f:
            data.extend(json.load(f))
        with open(os.path.join(temp_dir, "sg_90k_part2_html_cleaned.json"), "r") as f:
            data.extend(json.load(f))
    
    data = [d for d in data if "conversations" in d and len(d["conversations"]) > 0]
    
    role_map = {
        "user": "user",
        "human": "user",
        "bard": "assistant",
        "bing": "assistant",
        "chatgpt": "assistant",
        "gpt": "assistant",
        "system": "system"
    }
    data = [{
        "id": d["id"],
        "messages": [{"role": role_map[c["from"]], "content": c["value"]} for c in d["conversations"]]
    } for d in data]
    
    ds = Dataset.from_list(data)
    convert_sft_dataset(
        ds=ds,
        hf_dataset_id=None,
        convert_fn=None,
        apply_keyword_filters=args.apply_keyword_filters,
        apply_empty_message_filters=args.apply_empty_message_filters,
        push_to_hub=args.push_to_hub,
        hf_entity=args.hf_entity,
        converted_dataset_name=args.converted_dataset_name,
        local_save_dir=args.local_save_dir,
        readme_content=readme_content,
    )