File size: 6,445 Bytes
2b62c4c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import gradio as gr

import core as core
from style import CSS, T_SYMBOLS, TITLE

demo = gr.Blocks(css=CSS)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(
        "This is a (WIP) collection of multilingual evaluation results obtained using our fork of the LM-evaluation-harness (https://github.com/OpenGPTX/lm-evaluation-harness), based on https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard.\
                Note that currently, not all benchmarks are available in all languages, results are averaged over those languages under the selected ones for which the benchmark is available.",
        elem_classes="markdown-text",
    )

    with gr.Column():
        with gr.Row():
            with gr.Column():
                with gr.Row():
                    search_bar = gr.Textbox(
                        label="Search models",
                        placeholder=" πŸ” Separate multiple queries with ';' and press ENTER...",
                        show_label=True,
                        elem_id="search-bar",
                    )

                    model_types = gr.CheckboxGroup(
                        label="Select model type",
                        choices=[
                            (
                                f"Pretrained {T_SYMBOLS['pretrained']}",
                                T_SYMBOLS["pretrained"],
                            ),
                            (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
                        ],
                        value=list(T_SYMBOLS.values()),
                    )
                with gr.Row():
                    langs_bar = gr.CheckboxGroup(
                        choices=core.languages_list,
                        value=core.languages_list,
                        label="Select languages to average over",
                        elem_id="column-select",
                        interactive=True,
                        scale=6,
                    )
                    with gr.Column(scale=1):
                        clear = gr.ClearButton(
                            langs_bar,
                            value="Deselect all languages",
                            size="sm",
                            scale=1,
                        )
                        select = gr.Button(
                            value="Select all languages", size="sm", scale=1
                        )

                        def update_bar():
                            langs_bar = gr.CheckboxGroup(
                                choices=core.languages_list,
                                value=core.languages_list,
                                label="Select languages to average over",
                                elem_id="column-select",
                                interactive=True,
                            )
                            return langs_bar

                        select.click(update_bar, inputs=[], outputs=langs_bar)

                with gr.Row():
                    acc_task_group_names = core.task_groups_with_task_type("accuracy")
                    shown_tasks = gr.CheckboxGroup(
                        choices=acc_task_group_names,
                        value=acc_task_group_names,
                        label="Select tasks to show",
                        elem_id="column-select",
                        interactive=True,
                        scale=50,
                    )
                    fewshot = gr.Radio(
                        choices=[("0-Shot", False), ("Few-shot", True)],
                        value=True,
                        label="Select evaluation type",
                        interactive=True,
                        scale=29,
                    )
                    fewshot.change(
                        core.fix_zeroshot, [shown_tasks, fewshot], shown_tasks
                    )
                    clear = gr.ClearButton(
                        shown_tasks, value="Deselect all tasks", size="sm", scale=21
                    )

        with gr.Tabs(elem_classes="tab-buttons") as tabs:
            with gr.TabItem(
                "πŸ… LLM accuracy benchmark", elem_id="llm-benchmark-tab-table-acc", id=0
            ) as acc:
                leaderboard_table = gr.Dataframe()
            with gr.TabItem(
                "🌐 LLM translation benchmark",
                elem_id="llm-benchmark-tab-table-misc",
                id=1,
            ) as misc:
                leaderboard_table_misc = gr.Dataframe()
            with gr.TabItem("Plots", elem_id="llm-plot-tab", id=2) as plot:
                leaderboard_plot = gr.Plot(elem_id="plot")
            acc.select(
                lambda x: core.update_tab_tasks(0, x),
                inputs=fewshot,
                outputs=[shown_tasks, fewshot],
            )
            misc.select(
                lambda x: core.update_tab_tasks(1, x),
                inputs=fewshot,
                outputs=[shown_tasks, fewshot],
            )
            for comp, fn in [
                (search_bar, "submit"),
                (langs_bar, "change"),
                (shown_tasks, "change"),
                (fewshot, "change"),
                (model_types, "change"),
            ]:
                getattr(comp, fn)(
                    core.update_df,
                    [shown_tasks, search_bar, langs_bar, model_types, fewshot],
                    leaderboard_table,
                )
                getattr(comp, fn)(
                    core.update_df,
                    [shown_tasks, search_bar, langs_bar, model_types, fewshot],
                    leaderboard_table_misc,
                )
                getattr(comp, fn)(
                    core.update_plot,
                    [shown_tasks, search_bar, langs_bar, model_types, fewshot],
                    leaderboard_plot,
                )

    gr.Blocks.load(
        block=demo,
        fn=core.update_df,
        inputs=[shown_tasks, search_bar, langs_bar, model_types, fewshot],
        outputs=leaderboard_table,
    )

    gr.Blocks.load(
        block=demo,
        fn=core.update_df,
        inputs=[shown_tasks, search_bar, langs_bar, model_types, fewshot],
        outputs=leaderboard_table_misc,
    )

    gr.Blocks.load(
        block=demo,
        fn=core.update_plot,
        inputs=[shown_tasks, search_bar, langs_bar, model_types, fewshot],
        outputs=leaderboard_plot,
    )

demo.launch()