File size: 14,617 Bytes
7dfe065
 
e394fbd
 
 
 
a8ede2f
 
7dfe065
a8ede2f
855cd65
a8ede2f
 
 
 
 
 
 
 
d40f223
fb33b22
855cd65
a8ede2f
855cd65
a8ede2f
7dfe065
a8ede2f
 
 
 
 
 
 
 
 
 
018441b
 
a8ede2f
7dfe065
a8ede2f
 
 
d40f223
a8ede2f
018441b
7dfe065
 
 
 
855cd65
7dfe065
 
669da77
855cd65
 
a8ede2f
 
855cd65
 
 
 
e394fbd
 
a8ede2f
855cd65
 
 
 
 
 
 
 
f69201c
a8ede2f
 
855cd65
 
 
 
 
 
 
a8ede2f
 
 
 
 
 
 
 
 
 
855cd65
 
 
 
 
a8ede2f
 
855cd65
 
a8ede2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
855cd65
 
a8ede2f
 
 
855cd65
 
 
 
a8ede2f
855cd65
a8ede2f
 
 
 
 
 
 
 
 
 
 
 
 
855cd65
 
 
 
 
 
a8ede2f
 
 
 
 
 
855cd65
 
 
a8ede2f
 
 
855cd65
 
 
a8ede2f
 
 
 
 
 
 
 
 
 
 
 
 
 
855cd65
7dfe065
a8ede2f
 
 
 
 
 
855cd65
 
a8ede2f
 
018441b
 
a8ede2f
855cd65
 
a8ede2f
 
 
 
 
855cd65
a8ede2f
bef4eff
 
a8ede2f
 
855cd65
6bcbc2b
a8ede2f
 
 
 
3c8d224
a8ede2f
 
 
3bc2f22
a8ede2f
 
855cd65
 
a8ede2f
 
 
 
 
 
 
 
 
 
855cd65
 
 
 
 
7dfe065
a8ede2f
 
 
 
 
 
 
 
 
 
 
855cd65
a8ede2f
7148b21
a8ede2f
bef4eff
d40f223
 
 
 
 
 
 
bef4eff
 
d40f223
fb33b22
a8ede2f
855cd65
a8ede2f
 
 
 
 
855cd65
a8ede2f
 
 
 
 
855cd65
 
 
a8ede2f
 
 
 
 
855cd65
a8ede2f
855cd65
a8ede2f
 
 
 
 
855cd65
 
a8ede2f
855cd65
a8ede2f
 
 
 
 
 
 
018441b
a8ede2f
 
 
855cd65
a8ede2f
 
 
018441b
a8ede2f
 
e504efd
855cd65
 
a8ede2f
018441b
a8ede2f
 
 
855cd65
 
a8ede2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
855cd65
a8ede2f
 
855cd65
a8ede2f
 
 
 
 
855cd65
a8ede2f
 
81b5773
c0945b2
81b5773
 
 
 
b3198c3
 
 
81b5773
 
c8ae03b
81b5773
a8ede2f
 
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
358
359
360
361
362
363
364
365
366
367
#!/usr/bin/env python

import os
import datetime
import socket

import gradio as gr
import pandas as pd

from apscheduler.schedulers.background import BackgroundScheduler

from huggingface_hub import snapshot_download

from src.display.about import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    EVALUATION_QUEUE_TEXT,
    INTRODUCTION_TEXT,
    LLM_BENCHMARKS_TEXT,
    LLM_BENCHMARKS_DETAILS,
    FAQ_TEXT,
    TITLE
)

from src.display.css_html_js import custom_css

from src.display.utils import (
    BENCHMARK_COLS,
    COLS,
    EVAL_COLS,
    EVAL_TYPES,
    NUMERIC_INTERVALS,
    TYPES,
    AutoEvalColumn,
    ModelType,
    fields,
    WeightType,
    Precision
)

from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, H4_TOKEN, IS_PUBLIC, QUEUE_REPO, REPO_ID, RESULTS_REPO
from src.populate import get_evaluation_queue_df, get_leaderboard_df
from src.submission.submit import add_new_eval
from src.utils import get_dataset_summary_table


def ui_snapshot_download(repo_id, local_dir, repo_type, tqdm_class, etag_timeout):
    try:
        print(local_dir)
        snapshot_download(repo_id=repo_id, local_dir=local_dir, repo_type=repo_type, tqdm_class=tqdm_class, etag_timeout=etag_timeout)
    except Exception as e:
        restart_space()


def restart_space():
    API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)


def init_space():
    dataset_df = get_dataset_summary_table(file_path='blog/Hallucination-Leaderboard-Summary.csv')

    if socket.gethostname() not in {'neuromancer'}:
            ui_snapshot_download(repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30)
            ui_snapshot_download(repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30)

    raw_data, original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)

    finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
    return dataset_df, original_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df


dataset_df, original_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = init_space()
leaderboard_df = original_df.copy()


# Searching and filtering
def update_table(hidden_df: pd.DataFrame,
                 columns: list,
                 type_query: list,
                 precision_query: list,
                 size_query: list,
                 query: str):
    filtered_df = filter_models(hidden_df, type_query, size_query, precision_query)
    filtered_df = filter_queries(query, filtered_df)
    df = select_columns(filtered_df, columns)
    return df


def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
    return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))]


def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
    # always_here_cols = [AutoEvalColumn.model_type_symbol.name, AutoEvalColumn.model.name]

    always_here_cols = [c.name for c in fields(AutoEvalColumn) if c.never_hidden]
    dummy_col = [AutoEvalColumn.dummy.name]

    # We use COLS to maintain sorting
    filtered_df = df[
        # always_here_cols + [c for c in COLS if c in df.columns and c in columns] + [AutoEvalColumn.dummy.name]
        always_here_cols + [c for c in COLS if c in df.columns and c in columns] + dummy_col
    ]
    return filtered_df


def filter_queries(query: str, filtered_df: pd.DataFrame):
    final_df = []
    if query != "":
        queries = [q.strip() for q in query.split(";")]
        for _q in queries:
            _q = _q.strip()
            if _q != "":
                temp_filtered_df = search_table(filtered_df, _q)
                if len(temp_filtered_df) > 0:
                    final_df.append(temp_filtered_df)
        if len(final_df) > 0:
            filtered_df = pd.concat(final_df)
            subset = [AutoEvalColumn.model.name, AutoEvalColumn.precision.name, AutoEvalColumn.revision.name]
            filtered_df = filtered_df.drop_duplicates(subset=subset)
    return filtered_df


def filter_models(df: pd.DataFrame,
                  type_query: list,
                  size_query: list,
                  precision_query: list) -> pd.DataFrame:
    # Show all models
    filtered_df = df

    type_emoji = [t[0] for t in type_query]
    filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
    filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]

    numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
    params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
    mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
    filtered_df = filtered_df.loc[mask]

    return filtered_df


# triggered only once at startup => read query parameter if it exists
def load_query(request: gr.Request):
    query = request.query_params.get("query") or ""
    return query


demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("Hallucinations Benchmark",
                        elem_id="llm-benchmark-tab-table",
                        id=0):
            with gr.Row():
                with gr.Column():
                    with gr.Row():
                        search_bar = gr.Textbox(placeholder=" πŸ” Model search (separate multiple queries with `;`)",
                                                show_label=False,
                                                elem_id="search-bar")
                    with gr.Row():
                        shown_columns = gr.CheckboxGroup(
                            choices=[
                                c.name
                                for c in fields(AutoEvalColumn)
                                if not c.hidden and not c.never_hidden and not c.dummy
                            ],
                            value=[
                                c.name
                                for c in fields(AutoEvalColumn)
                                if c.displayed_by_default and not c.hidden and not c.never_hidden
                            ],
                            label="Select columns to show",
                            elem_id="column-select",
                            interactive=True)

                with gr.Column(min_width=320):
                    filter_columns_type = gr.CheckboxGroup(
                        label="Model types",
                        choices=[t.to_str() for t in ModelType],
                        value=[t.to_str() for t in ModelType],
                        interactive=True,
                        elem_id="filter-columns-type")

                    filter_columns_precision = gr.CheckboxGroup(
                        label="Precision",
                        choices=[i.value.name for i in Precision],
                        value=[i.value.name for i in Precision],
                        interactive=True,
                        elem_id="filter-columns-precision")

                    filter_columns_size = gr.CheckboxGroup(
                        label="Model sizes (in billions of parameters)",
                        choices=list(NUMERIC_INTERVALS.keys()),
                        value=list(NUMERIC_INTERVALS.keys()),
                        interactive=True,
                        elem_id="filter-columns-size")

            # breakpoint()

            leaderboard_table = gr.components.Dataframe(
                value=leaderboard_df[
                    [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value + [AutoEvalColumn.dummy.name]
                ] if leaderboard_df.empty is False else leaderboard_df,
                headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
                datatype=TYPES,
                elem_id="leaderboard-table",
                interactive=False,
                visible=True)  # column_widths=["2%", "20%"]

            # Dummy leaderboard for handling the case when the user uses backspace key
            hidden_leaderboard_table_for_search = gr.components.Dataframe(
                value=original_df[COLS] if original_df.empty is False else original_df,
                headers=COLS,
                datatype=TYPES,
                visible=False)

            search_bar.submit(
                update_table,
                [
                    hidden_leaderboard_table_for_search,
                    shown_columns,
                    filter_columns_type,
                    filter_columns_precision,
                    filter_columns_size,
                    search_bar,
                ],
                leaderboard_table)

            # Check query parameter once at startup and update search bar
            demo.load(load_query, inputs=[], outputs=[search_bar])

            for selector in [shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size]:
                selector.change(
                    update_table,
                    [
                        hidden_leaderboard_table_for_search,
                        shown_columns,
                        filter_columns_type,
                        filter_columns_precision,
                        filter_columns_size,
                        search_bar,
                    ],
                    leaderboard_table,
                    queue=True)

        with gr.TabItem("About", elem_id="llm-benchmark-tab-table", id=2):
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

            dataset_table = gr.components.Dataframe(
                value=dataset_df,
                headers=list(dataset_df.columns),
                datatype=['str', 'markdown', 'str', 'str', 'str'],
                elem_id="dataset-table",
                interactive=False,
                visible=True,
                column_widths=["15%", "20%"])

            gr.Markdown(LLM_BENCHMARKS_DETAILS, elem_classes="markdown-text")
            gr.Markdown(FAQ_TEXT, elem_classes="markdown-text")

        with gr.TabItem("Submit a model ", elem_id="llm-benchmark-tab-table", id=3):
            with gr.Column():
                with gr.Row():
                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")

                with gr.Column():
                    with gr.Accordion(f"βœ… Finished Evaluations ({len(finished_eval_queue_df)})", open=False):
                        with gr.Row():
                            finished_eval_table = gr.components.Dataframe(
                                value=finished_eval_queue_df,
                                headers=EVAL_COLS,
                                datatype=EVAL_TYPES,
                                row_count=5)

                    with gr.Accordion(f"πŸ”„ Running Evaluation Queue ({len(running_eval_queue_df)})", open=False):
                        with gr.Row():
                            running_eval_table = gr.components.Dataframe(
                                value=running_eval_queue_df,
                                headers=EVAL_COLS,
                                datatype=EVAL_TYPES,
                                row_count=5)

                    with gr.Accordion(f"⏳ Scheduled Evaluation Queue ({len(pending_eval_queue_df)})", open=False):
                        with gr.Row():
                            pending_eval_table = gr.components.Dataframe(
                                value=pending_eval_queue_df,
                                headers=EVAL_COLS,
                                datatype=EVAL_TYPES,
                                row_count=5)

            with gr.Row():
                gr.Markdown("# Submit your model here", elem_classes="markdown-text")

            with gr.Row():
                with gr.Column():
                    model_name_textbox = gr.Textbox(label="Model name")
                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
                    private = gr.Checkbox(False, label="Private", visible=not IS_PUBLIC)
                    model_type = gr.Dropdown(
                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
                        label="Model type",
                        multiselect=False,
                        value=None,
                        interactive=True)

                with gr.Column():
                    precision = gr.Dropdown(
                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
                        label="Precision",
                        multiselect=False,
                        value="float32",
                        interactive=True)

                    weight_type = gr.Dropdown(
                        choices=[i.value.name for i in WeightType],
                        label="Weights type",
                        multiselect=False,
                        value="Original",
                        interactive=True)

                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")

            submit_button = gr.Button("Submit Eval")
            submission_result = gr.Markdown()
            submit_button.click(
                add_new_eval,
                [
                    model_name_textbox,
                    base_model_name_textbox,
                    revision_name_textbox,
                    precision,
                    private,
                    weight_type,
                    model_type,
                ],
                submission_result)

    with gr.Row():
        with gr.Accordion("Citing this leaderboard", open=False):
            citation_button = gr.Textbox(
                value=CITATION_BUTTON_TEXT,
                label=CITATION_BUTTON_LABEL,
                lines=20,
                elem_id="citation-button",
                show_copy_button=True)

scheduler = BackgroundScheduler()

scheduler.add_job(restart_space, "interval", seconds=6 * 60 * 60)


def launch_backend():
    import subprocess
    from src.backend.envs import DEVICE
    if DEVICE not in {'cpu'}:
        _ = subprocess.run(["python", "backend-cli.py"])


# scheduler.add_job(launch_backend, "interval", seconds=120)

scheduler.start()
demo.queue(default_concurrency_limit=40).launch()