File size: 3,859 Bytes
dbafbbf
15bf463
b1dd47e
dbafbbf
b1dd47e
15bf463
b1dd47e
 
dbafbbf
 
 
15bf463
 
dbafbbf
15bf463
b1dd47e
15bf463
 
 
dbafbbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1dd47e
 
 
 
 
 
 
 
 
dbafbbf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1dd47e
 
 
 
 
 
 
 
 
 
 
d3a17fc
b1dd47e
dbafbbf
 
 
 
 
 
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
from typing import Generator, Set, Union, List

import requests
from bs4 import BeautifulSoup, Tag, NavigableString, PageElement
from concurrent.futures import ThreadPoolExecutor, as_completed

SUPPORTED_MODEL_NAME_PAGES_FORMAT = "https://huggingface.co/models?pipeline_tag=text-generation&library=pytorch"
MAX_WORKERS = 10
BLACKLISTED_MODEL_NAMES = {"ykilcher/gpt-4chan"}
MIN_NUMBER_OF_DOWNLOADS = 100
MIN_NUMBER_OF_LIKES = 20


def get_model_name(model_card: Tag) -> str:
    h4_class = "text-md truncate font-mono text-black dark:group-hover:text-yellow-500 group-hover:text-indigo-600"
    h4_tag = model_card.find("h4", class_=h4_class)
    return h4_tag.text


def is_a_number(s: PageElement) -> bool:
    s = s.text.strip().lower().replace("k", "").replace("m", "").replace(",", "").replace(".", "").replace("b", "")
    try:
        float(s)
        return True
    except ValueError:
        return False


def get_numeric_contents(model_card):
    div: Union[Tag | NavigableString] = model_card.find(
        "div",
        class_="mr-1 flex items-center overflow-hidden whitespace-nowrap text-sm leading-tight text-gray-400",
        recursive=True
    )
    contents: List[PageElement] = div.contents
    contents_without_tags: List[PageElement] = [content for content in contents if not isinstance(content, Tag)]
    number_contents: List[PageElement] = [content for content in contents_without_tags if is_a_number(content)]
    return number_contents


def convert_to_int(element: PageElement) -> int:
    element_str = element.text.strip().lower()
    if element_str.endswith("k"):
        return int(float(element_str[:-1]) * 1_000)
    elif element_str.endswith("m"):
        return int(float(element_str[:-1]) * 1_000_000)
    elif element_str.endswith("b"):
        return int(float(element_str[:-1]) * 1_000_000_000)
    else:
        return int(element_str)


def get_page(page_index: int):
    curr_page_url = f"{SUPPORTED_MODEL_NAME_PAGES_FORMAT}&p={page_index}"
    response = requests.get(curr_page_url)
    if response.status_code == 200:
        soup = BeautifulSoup(response.content, "html.parser")
        return soup
    return None


def card_filter(model_card: Tag, model_name: str) -> bool:
    if model_name in BLACKLISTED_MODEL_NAMES:
        return False
    numeric_contents = get_numeric_contents(model_card)
    if len(numeric_contents) < 2:
        # If the model card doesn't have at least 2 numeric contents,
        # It means that he doesn't have any downloads/likes, so it's not a valid model card.
        return False
    number_of_downloads = convert_to_int(numeric_contents[0])
    if number_of_downloads < MIN_NUMBER_OF_DOWNLOADS:
        return False
    number_of_likes = convert_to_int(numeric_contents[1])
    if number_of_likes < MIN_NUMBER_OF_LIKES:
        return False
    return True


def get_model_names(soup: BeautifulSoup):
    model_cards: List[Tag] = soup.find_all("article", class_="overview-card-wrapper group", recursive=True)
    for model_card in model_cards:
        model_name = get_model_name(model_card)
        if card_filter(model_card, model_name):
            yield model_name


def generate_supported_model_names() -> Generator[str, None, None]:
    with ThreadPoolExecutor(max_workers=MAX_WORKERS) as executor:
        future_to_index = {executor.submit(get_page, index): index for index in range(100)}
        for future in as_completed(future_to_index):
            soup = future.result()
            if soup:
                yield from get_model_names(soup)


def get_supported_model_names() -> Set[str]:
    return set(generate_supported_model_names())


if __name__ == "__main__":
    supported_model_names = get_supported_model_names()
    print(f"Number of supported model names: {len(supported_model_names)}")
    print(f"Supported model names: {supported_model_names}")