File size: 12,074 Bytes
dfb91d7
 
 
 
 
 
 
 
1f93c1a
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6654d7b
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
af20151
dfb91d7
 
 
 
 
 
 
 
 
ad10431
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1f93c1a
 
dfb91d7
 
 
1f93c1a
dfb91d7
 
9a49194
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1f93c1a
 
dfb91d7
af20151
1f93c1a
 
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
f5878c6
dfb91d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import io
import re
from typing import *
import pandas as pd
import streamlit as st
from pandas.api.types import is_bool_dtype, is_datetime64_any_dtype, is_numeric_dtype

GITHUB_URL = "https://github.com/RSTLess-research/"
NON_BENCHMARK_COLS = ["Publisher"]


def extract_table_and_format_from_markdown_text(markdown_table: str) -> pd.DataFrame:
    """Extracts a table from a markdown text and formats it as a pandas DataFrame.

    Args:
        text (str): Markdown text containing a table.

    Returns:
        pd.DataFrame: Table as pandas DataFrame.
    """
    df = (
        pd.read_table(io.StringIO(markdown_table), sep="|", header=0, index_col=1)
        .dropna(axis=1, how="all")  # drop empty columns
        .iloc[1:]  # drop first row which is the "----" separator of the original markdown table
        .sort_index(ascending=True)
        .apply(lambda x: x.str.strip() if x.dtype == "object" else x)
        .replace("", float("NaN"))
        .astype(float, errors="ignore")
    )

    # remove whitespace from column names and index
    df.columns = df.columns.str.strip()
    df.index = df.index.str.strip()
    df.index.name = df.index.name.strip()

    return df


def extract_markdown_table_from_multiline(multiline: str, table_headline: str, next_headline_start: str = "#") -> str:
    """Extracts the markdown table from a multiline string.

    Args:
        multiline (str): content of README.md file.
        table_headline (str): Headline of the table in the README.md file.
        next_headline_start (str, optional): Start of the next headline. Defaults to "#".

    Returns:
        str: Markdown table.

    Raises:
        ValueError: If the table could not be found.
    """
    # extract everything between the table headline and the next headline
    table = []
    start = False
    for line in multiline.split("\n"):
        if line.startswith(table_headline):
            start = True
        elif line.startswith(next_headline_start):
            start = False
        elif start:
            table.append(line + "\n")

    if len(table) == 0:
        raise ValueError(f"Could not find table with headline '{table_headline}'")

    return "".join(table)


def remove_markdown_links(text: str) -> str:
    """Modifies a markdown text to remove all markdown links.
    Example: [DISPLAY](LINK) to DISPLAY
    First find all markdown links with regex.
    Then replace them with: $1
    Args:
        text (str): Markdown text containing markdown links
    Returns:
        str: Markdown text without markdown links.
    """

    # find all markdown links
    markdown_links = re.findall(r"\[([^\]]+)\]\(([^)]+)\)", text)

    # remove link keep display text
    for display, link in markdown_links:
        text = text.replace(f"[{display}]({link})", display)

    return text

def filter_dataframe_by_model_type(df: pd.DataFrame, model_type_column: str = 'Lang.', ignore_columns: List[str] = None) -> pd.DataFrame:
    """
    Filter dataframe by the rows based on model type and by user-selected columns.

    This function provides a user interface to select model types and columns for filtering a DataFrame.
    Model types are dynamically derived from the column specified as 'model_type_column'.

    Args:
        df (pd.DataFrame): Original dataframe.
        model_type_column (str): Column name that contains model types for filtering.
        ignore_columns (list[str], optional): Columns to ignore when showing in column selection. Defaults to None.

    Returns:
        pd.DataFrame: Filtered dataframe.
    """
    df = df.copy()

    if ignore_columns is None:
        ignore_columns = []

    # Streamlit UI Container
    modification_container = st.container()

    with modification_container:
        # Selection for model types
        unique_model_types = sorted(df[model_type_column].unique())
        selected_model_types = st.multiselect("Filter by model type:", unique_model_types)

        # Filter dataframe by selected model types
        if selected_model_types:
            df = df[df[model_type_column].isin(selected_model_types)]

        # Column selection excluding the model type column and any specified ignore columns
        valid_columns = sorted(set(df.columns) - set(ignore_columns) - {model_type_column})
        selected_columns = st.multiselect("Filter by columns:", valid_columns)

        # Filter dataframe to include only the selected columns plus the model type column
        if selected_columns:
            df = pd.DataFrame(df[[model_type_column] + selected_columns])

    return df

def filter_dataframe_by_row_and_columns(df: pd.DataFrame, ignore_columns: List[str] = None) -> pd.DataFrame:
    """
    Filter dataframe by the rows and columns to display.

    This does not select based on the values in the dataframe, but rather on the index and columns.
    Modified from https://blog.streamlit.io/auto-generate-a-dataframe-filtering-ui-in-streamlit-with-filter_dataframe/

    Args:
        df (pd.DataFrame): Original dataframe
        ignore_columns (list[str], optional): Columns to ignore. Defaults to None.

    Returns:
        pd.DataFrame: Filtered dataframe
    """
    df = df.copy()

    if ignore_columns is None:
        ignore_columns = []

    modification_container = st.container()

    with modification_container:
        to_filter_index = st.multiselect("Filter by model:", sorted(df.index))
        if to_filter_index:
            df = pd.DataFrame(df.loc[to_filter_index])

        to_filter_columns = st.multiselect(
            "Filter by benchmark:", sorted([c for c in df.columns if c not in ignore_columns])
        )
        if to_filter_columns:
            df = pd.DataFrame(df[ignore_columns + to_filter_columns])

    return df


def filter_dataframe_by_column_values(df: pd.DataFrame) -> pd.DataFrame:
    """
    Filter dataframe by the values in the dataframe.

    Modified from https://blog.streamlit.io/auto-generate-a-dataframe-filtering-ui-in-streamlit-with-filter_dataframe/

    Args:
        df (pd.DataFrame): Original dataframe

    Returns:
        pd.DataFrame: Filtered dataframe
    """
    df = df.copy()

    modification_container = st.container()

    with modification_container:
        to_filter_columns = st.multiselect("Filter results on:", df.columns)
        left, right = st.columns((1, 20))

        for column in to_filter_columns:
            if is_bool_dtype(df[column]):
                user_bool_input = right.checkbox(f"{column}", value=True)
                df = df[df[column] == user_bool_input]

            elif is_numeric_dtype(df[column]):
                _min = float(df[column].min())
                _max = float(df[column].max())

                if (_min != _max) and pd.notna(_min) and pd.notna(_max):
                    step = 0.01
                    user_num_input = right.slider(
                        f"Values for {column}:",
                        min_value=round(_min - step, 2),
                        max_value=round(_max + step, 2),
                        value=(_min, _max),
                        step=step,
                    )
                    df = df[df[column].between(*user_num_input)]

            elif is_datetime64_any_dtype(df[column]):
                user_date_input = right.date_input(
                    f"Values for {column}:",
                    value=(
                        df[column].min(),
                        df[column].max(),
                    ),
                )
                if isinstance(user_date_input, Iterable) and len(user_date_input) == 2:
                    user_date_input_datetime = tuple(map(pd.to_datetime, user_date_input))
                    start_date, end_date = user_date_input_datetime
                    df = df.loc[df[column].between(start_date, end_date)]

            else:
                selected_values = right.multiselect(
                    f"Values for {column}:",
                    sorted(df[column].unique()),
                )

                if selected_values:
                    df = df[df[column].isin(selected_values)]

    return df


def setup_basic():
    title = "๐Ÿ† Italian LLM-Leaderboard ๐Ÿ‡ฎ๐Ÿ‡น๐ŸคŒ"

    st.set_page_config(
        page_title=title,
        page_icon="๐Ÿ†๐Ÿ‡ฎ๐Ÿ‡น๐ŸคŒ",
        layout="wide",
    )
    st.title(title)

    st.markdown(
        "The Italian Open LLM Leaderboard published along with the paper _DanteLLM: Let's Push Italian LLM Research Forward!_ ๐ŸคŒ๐Ÿ‡ฎ๐Ÿ‡น๐Ÿ† (To be presented at: LREC-COLING 2024, May 20th-25th) \n"
    )

def setup_leaderboard(readme: str):
    leaderboard_table = extract_markdown_table_from_multiline(readme, table_headline="## Leaderboard")
    leaderboard_table = remove_markdown_links(leaderboard_table)
    df_leaderboard = extract_table_and_format_from_markdown_text(leaderboard_table)

    st.markdown("## Leaderboard")
    modify = st.checkbox("Add filters")

    if modify:
        df_leaderboard = filter_dataframe_by_row_and_columns(df_leaderboard, ignore_columns=NON_BENCHMARK_COLS)
        df_leaderboard = filter_dataframe_by_column_values(df_leaderboard)
        df_leaderboard = filter_dataframe_by_model_type(df_leaderboard)

    df_leaderboard = df_leaderboard.sort_values(by=['Avg.'], ascending=False)
    df_leaderboard["Rank"] = df_leaderboard["Avg."].rank(ascending=False)
    # move rank at 0-th column
    # Ensure 'Rank' is the first column
    cols = ['Rank'] + [col for col in df_leaderboard.columns if col != 'Rank']

    df_leaderboard = df_leaderboard[cols]

    #print(df_leaderboard.columns)
    #df_leaderboard.reset_index(drop=True, inplace=True)

    st.dataframe(df_leaderboard)

    st.download_button(
        "Download leaderboard as .html",
        df_leaderboard.to_html().encode("utf-8"),
        "leaderboard.html",
        "text/html",
        key="download-html",
    )

    st.download_button(
        "Download leaderboard as .csv",
        df_leaderboard.to_csv().encode("utf-8"),
        "leaderboard.csv",
        "text/csv",
        key="download-csv",
    )


def setup_benchmarks(readme: str):
    benchmarks_table = extract_markdown_table_from_multiline(readme, table_headline="## Benchmarks")
    df_benchmarks = extract_table_and_format_from_markdown_text(benchmarks_table)

    st.markdown("## Covered Benchmarks")

    selected_benchmark = st.selectbox("Select a benchmark to learn more:", df_benchmarks.index.unique())
    df_selected = df_benchmarks.loc[selected_benchmark]
    text = [
        f"Name: {selected_benchmark}",
    ]
    for key in df_selected.keys():
        text.append(f"{key}: {df_selected[key]} ")
    st.markdown("  \n".join(text))


def setup_sources():
    st.markdown("## Sources")
    st.markdown(
        "The results of this leaderboard are collected from the individual papers and published results of the model "
        "authors. If you are interested in the sources of each individual reported model value, please visit the "
        f"[llm-leaderboard]({GITHUB_URL}) repository."
    )


def setup_disclaimer():

    st.markdown("## Authors")
    st.markdown(
"""
- [Andrea Bacciu](https://www.linkedin.com/in/andreabacciu/)* (Work done prior joining Amazon)
- [Cesare Campagnano](https://www.linkedin.com/in/caesar-one/)*
- [Giovanni Trappolini](https://www.linkedin.com/in/giovanni-trappolini/)
- [Prof. Fabrizio Silvestri](https://www.linkedin.com/in/fabrizio-silvestri-a6b0391/)

\*Equal contribution
"""
    )

    st.markdown("## Ack")
    st.markdown(
        f"Special thanks to [llm-leaderboard](https://github.com/LudwigStumpp/llm-leaderboard) for the initial inspiration and codebase"
    )


def setup_footer():
    st.markdown(
        """
        ---
        Made with โค๏ธ by the awesome open-source Italian community ๐ŸคŒ๐Ÿ‡ฎ๐Ÿ‡น.
        """
    )


def main():
    setup_basic()

    with open("README.md", "r") as f:
        readme = f.read()

    setup_leaderboard(readme)
    setup_benchmarks(readme)
    setup_sources()
    setup_disclaimer()
    setup_footer()


if __name__ == "__main__":
    main()