File size: 5,374 Bytes
8163dc5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import ast
import argparse
import glob
import pickle

import gradio as gr
import numpy as np
import pandas as pd
def model_hyperlink(model_name, link):
    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
def load_leaderboard_table_csv(filename, add_hyperlink=True):
    lines = open(filename).readlines()
    heads = [v.strip() for v in lines[0].split(",")]
    rows = []
    for i in range(1, len(lines)):
        row = [v.strip() for v in lines[i].split(",")]
        for j in range(len(heads)):
            item = {}
            for h, v in zip(heads, row):
                if "Score" in h:
                    item[h] = float(v)
                elif h != "Model" and h != "Parameters (B)" and h != "Repo" and h != "Quantization" and h != "Link":
                    item[h] = int(v)
                else:
                    item[h] = v
            if add_hyperlink:
                item["Repo"] = model_hyperlink(item["Repo"], item["Link"])
        rows.append(item)
    return rows

def get_arena_table(model_table_df):
    # sort by rating
    model_table_df = model_table_df.sort_values(by=["Final Score"], ascending=False)
    values = []
    for i in range(len(model_table_df)):
        row = []
        model_key = model_table_df.index[i]
        model_name = model_table_df["Model"].values[model_key]
        # rank
        row.append(i + 1)
        # model display name
        row.append(model_name)

        row.append(
            model_table_df["Parameters (B)"].values[model_key]
        )
        row.append(
            model_table_df["Repo"].values[model_key]
        )
        row.append(
            model_table_df["Quantization"].values[model_key]
        )
        row.append(
            model_table_df["Final Score"].values[model_key]
        )
        row.append(
            model_table_df["Strict Prompt Score"].values[model_key]
        )
        row.append(
            model_table_df["Strict Inst Score"].values[model_key]
        )
        row.append(
            model_table_df["Loose Prompt Score"].values[model_key]
        )
        row.append(
            model_table_df["Loose Inst Score"].values[model_key]
        )
        values.append(row)
    return values

def build_leaderboard_tab(leaderboard_table_file, show_plot=False):
    if leaderboard_table_file:
        data = load_leaderboard_table_csv(leaderboard_table_file)
        model_table_df = pd.DataFrame(data)
        md_head = f"""
        # 🏆 IFEval Leaderboard
        """
        gr.Markdown(md_head, elem_id="leaderboard_markdown")
        with gr.Tabs() as tabs:
            # arena table
            arena_table_vals = get_arena_table(model_table_df)
            with gr.Tab("IFEval", id=0):
                md = "Leaderboard for various Large Language Models measured with IFEval benchmark.\n\n[IFEval](https://github.com/google-research/google-research/tree/master/instruction_following_eval) is a straightforward and easy-to-reproduce evaluation benchmark. It focuses on a set of \"verifiable instructions\" such as \"write in more than 400 words\" and \"mention the keyword of AI at least 3 times\". We identified 25 types of those verifiable instructions and constructed around 500 prompts, with each prompt containing one or more verifiable instructions. \n\nTest ran with `lm-evaluation-harness`. Raw results can be found in the `results` directory. Made by [Kristian Polso](https://polso.info)."
                gr.Markdown(md, elem_id="leaderboard_markdown")
                gr.Dataframe(
                    headers=[
                        "Rank",
                        "Model",
                        "Parameters (B)",
                        "Repo",
                        "Quantization",
                        "Final Score",
                        "Strict Prompt Score",
                        "Strict Inst Score",
                        "Loose Prompt Score",
                        "Loose Inst Score"
                    ],
                    datatype=[
                        "number",
                        "str",
                        "number",
                        "markdown",
                        "str",
                        "number",
                        "number",
                        "number",
                        "number",
                        "number"
                    ],
                    value=arena_table_vals,
                    elem_id="arena_leaderboard_dataframe",
                    height=700,
                    column_widths=[50, 150, 100, 150, 100, 100, 100, 100, 100, 100],
                    wrap=True,
                )

    else:
        pass

def build_demo(leaderboard_table_file):
    text_size = gr.themes.sizes.text_lg

    with gr.Blocks(
        title="IFEval Leaderboard",
        theme=gr.themes.Base(text_size=text_size),
    ) as demo:
        leader_components = build_leaderboard_tab(
            leaderboard_table_file, show_plot=True
        )
    return demo

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--share", action="store_true")
    parser.add_argument("--IFEval_file", type=str, default="./IFEval.csv")
    args = parser.parse_args()

    demo = build_demo(args.IFEval_file)
    demo.launch()