make_task_files.py

import os
from collections import defaultdict

def stat_size(fnames):
    cnt = 0
    for fname in fnames:
        with open(fname, 'r') as f:
            for line in f:
                cnt += 1
    return cnt


dataset2fname = defaultdict(lambda :defaultdict(list))
final_task_files = []
srcs = [
    'bge-m3', 
    'medi',
    'mteb-Classification',
    'mteb-Clustering',
    'mteb-PairClassification',
    'mteb-Reranking',
    'mteb-Retrieval',
    'mteb-Retrieval_aug',
    'mteb-STS',
]

for src in srcs:
    src_path = f'/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/{src}'
    for fname in os.listdir(src_path):
        if not fname.endswith('.jsonl'):
            continue
        _, dataset, lang = fname[:-6].split('_')
        if src.startswith('mteb') and lang != 'default' and not lang.startswith('en'):
            continue
            # if lang == 'default' or lang.startswith('en'):
        fname = os.path.join(src_path, fname)
        dataset2fname[dataset][src].append(fname)

for dataset, item in dataset2fname.items():
    if len(item) == 1:
        fnames = item[list(item.keys())[0]]
        final_task_files.extend(fnames)
    else:
        max_size = -1
        max_size_src = None
        for src, fnames in item.items():
            size = stat_size(fnames)
            if size > max_size:
                max_size = size
                max_size_src = src
        fnames = item[max_size_src]
        final_task_files.extend(fnames)

# with open('/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/task_files.txt', 'w') as f:
#     for task_file in final_task_files:
#         if 'Classification' in task_file:
#             line = f'{task_file}\tclassification'
#         else:
#             line = f'{task_file}\tdefault'
#         f.write(line + '\n')

# from tqdm import tqdm
# cnt = 0
# for task_file in tqdm(final_task_files):
#     with open(task_file, 'r') as f:
#         for line in f:
#             cnt += 1

# fnames = []
# with open('/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/task_files.txt', 'r') as f:
#     for line in f:
#         fname, task_type = line.strip().split('\t')
#         fnames.append(fname)

with open('/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/p+a_aug_en_task_files.txt', 'w') as f:
    for fname in final_task_files:
        if 'classification' in fname.lower():
            task_type = 'classification'
        elif 'medi_task' in fname.lower():
            task_type = 'super-NI'
        elif 'clustering' in fname.lower():
            task_type = 'clustering'
        elif 'sts' in fname.lower():
            task_type = 'sts'
        else:
            task_type = 'default'
        # f.write(f'{fname}\t{task_type}\t-1\n')
        f.write(f'{fname}\t-1\n')


# def stat_size(fnames):
#     cnt = 0
#     for fname in fnames:
#         with open(fname, 'r') as f:
#             for line in f:
#                 cnt += 1
#     return cnt

# f_in = open('/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/p+a_task_files.txt', 'r')
# f_out = open('/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/p+a_task_files_en.txt', 'w')
# for line in f_in:
#     file, type_, size = line.strip().split('\t')
#     lang = file[:-6].split('_')[-1]
#     if lang == 'default' or lang.startswith('en'):
#         f_out.write(line)
# f_in.close()
# f_out.close()