File size: 20,091 Bytes
4596a70
0227006
d35aee2
4596a70
9346f1c
 
4596a70
460d762
0227006
460d762
8c49cb6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d350941
9346f1c
 
d16cee2
 
 
 
 
 
 
460d762
1f60a20
d16cee2
 
d52179b
d16cee2
 
2a73469
8c49cb6
 
2a73469
10f9b3c
8c49cb6
10f9b3c
f742519
8c49cb6
d52179b
 
 
 
f742519
460d762
12cea14
 
9346f1c
460d762
 
9346f1c
8c49cb6
 
 
 
 
 
 
 
 
9346f1c
8c49cb6
 
 
 
a885f09
8c49cb6
 
 
 
 
 
 
 
 
2a73469
8c49cb6
ffefe11
 
 
 
 
8c49cb6
614ee1f
1f60a20
8c49cb6
85dbbc4
 
 
 
12cea14
85dbbc4
12cea14
217b585
85dbbc4
12cea14
8696209
 
ef627e9
 
 
1f60a20
b2c063a
 
614ee1f
12cea14
460d762
 
 
 
2f6ebf5
 
 
 
8c49cb6
1f60a20
614ee1f
1f60a20
85dbbc4
 
 
 
12cea14
 
85dbbc4
8696209
217b585
614ee1f
 
1f60a20
 
 
 
 
614ee1f
d52179b
1f60a20
12cea14
614ee1f
ed1fdef
 
 
 
f742519
d16cee2
8c49cb6
1363c8a
1f60a20
 
614ee1f
1f60a20
 
d16cee2
 
1f60a20
d16cee2
1f60a20
a885f09
d16cee2
 
 
8c49cb6
 
 
1f60a20
614ee1f
8c49cb6
 
 
ffefe11
 
 
 
8c49cb6
ffefe11
 
 
 
 
 
614ee1f
2a73469
8c49cb6
 
 
 
 
 
 
 
 
 
 
 
 
 
99b25b8
 
 
8c49cb6
99b25b8
 
8c49cb6
ecef2dc
aa7c3f4
8c49cb6
 
 
 
 
 
 
 
 
ecef2dc
7644705
d2179b0
8c49cb6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d2179b0
8c49cb6
d2179b0
7644705
01233b7
 
58733e4
6e8f400
10f9b3c
8cb7546
613696b
ecef2dc
8c49cb6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
601f2e9
 
 
 
 
 
d2179b0
8c49cb6
d2179b0
8c49cb6
 
d2179b0
 
 
8c49cb6
d2179b0
 
8c49cb6
 
d2179b0
 
 
8c49cb6
d2179b0
 
 
 
 
 
8c49cb6
d2179b0
 
 
8c49cb6
d2179b0
8c49cb6
6e8f400
8c49cb6
 
 
 
 
 
 
 
 
 
 
6e8f400
 
 
ecef2dc
 
6e8f400
460d762
6e8f400
 
 
 
 
 
 
 
 
 
8c49cb6
 
 
 
 
 
 
 
 
 
6e8f400
8c49cb6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e8f400
613696b
6e8f400
0227006
613696b
8dfa543
0227006
8dfa543
6e8f400
8dfa543
8c49cb6
 
 
 
8dfa543
 
 
 
 
 
 
8c49cb6
 
 
 
8dfa543
 
 
 
 
 
 
 
8c49cb6
 
 
 
8dfa543
 
 
 
 
 
 
00358b1
 
0227006
6e8f400
 
 
8c49cb6
 
b323764
8c49cb6
95f85ed
 
 
8c49cb6
 
 
b323764
ef627e9
b323764
 
0227006
6e8f400
12cea14
8c49cb6
 
 
 
 
 
 
12cea14
 
217b585
 
12cea14
 
8c49cb6
12cea14
 
 
6e8f400
8c49cb6
8cb7546
6e8f400
 
 
 
 
 
 
 
12cea14
6e8f400
12cea14
8c49cb6
6e8f400
 
8cb7546
 
8dfa543
 
 
 
 
 
 
 
 
 
 
 
 
d16cee2
 
 
 
 
 
 
 
8cb7546
 
 
 
 
 
 
10f9b3c
 
c131125
10f9b3c
8c49cb6
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
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
import json
import os
from datetime import datetime, timezone

import gradio as gr
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import HfApi

from src.assets.css_html_js import custom_css, get_window_url_params
from src.assets.text_content import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    EVALUATION_QUEUE_TEXT,
    INTRODUCTION_TEXT,
    LLM_BENCHMARKS_TEXT,
    TITLE,
)
from src.display_models.get_model_metadata import DO_NOT_SUBMIT_MODELS, ModelType
from src.display_models.utils import (
    AutoEvalColumn,
    EvalQueueColumn,
    fields,
    styled_error,
    styled_message,
    styled_warning,
)
from src.load_from_hub import get_evaluation_queue_df, get_leaderboard_df, is_model_on_hub, load_all_info_from_hub

pd.set_option("display.precision", 1)

# clone / pull the lmeh eval data
H4_TOKEN = os.environ.get("H4_TOKEN", None)

QUEUE_REPO = "open-llm-leaderboard/requests"
RESULTS_REPO = "open-llm-leaderboard/results"

PRIVATE_QUEUE_REPO = "open-llm-leaderboard/private-requests"
PRIVATE_RESULTS_REPO = "open-llm-leaderboard/private-results"

IS_PUBLIC = bool(os.environ.get("IS_PUBLIC", True))

EVAL_REQUESTS_PATH = "eval-queue"
EVAL_RESULTS_PATH = "eval-results"

EVAL_REQUESTS_PATH_PRIVATE = "eval-queue-private"
EVAL_RESULTS_PATH_PRIVATE = "eval-results-private"

api = HfApi(token=H4_TOKEN)


def restart_space():
    api.restart_space(repo_id="HuggingFaceH4/open_llm_leaderboard", token=H4_TOKEN)


# Column selection
COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
TYPES = [c.type for c in fields(AutoEvalColumn) if not c.hidden]
COLS_LITE = [c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]
TYPES_LITE = [c.type for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]

if not IS_PUBLIC:
    COLS.insert(2, AutoEvalColumn.precision.name)
    TYPES.insert(2, AutoEvalColumn.precision.type)

EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]

BENCHMARK_COLS = [
    c.name
    for c in [
        AutoEvalColumn.arc,
        AutoEvalColumn.hellaswag,
        AutoEvalColumn.mmlu,
        AutoEvalColumn.truthfulqa,
    ]
]

## LOAD INFO FROM HUB
eval_queue, requested_models, eval_results = load_all_info_from_hub(
    QUEUE_REPO, RESULTS_REPO, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH
)

if not IS_PUBLIC:
    (eval_queue_private, requested_models_private, eval_results_private,) = load_all_info_from_hub(
        PRIVATE_QUEUE_REPO,
        PRIVATE_RESULTS_REPO,
        EVAL_REQUESTS_PATH_PRIVATE,
        EVAL_RESULTS_PATH_PRIVATE,
    )
else:
    eval_queue_private, eval_results_private = None, None

original_df = get_leaderboard_df(eval_results, eval_results_private, COLS, BENCHMARK_COLS)
leaderboard_df = original_df.copy()
(
    finished_eval_queue_df,
    running_eval_queue_df,
    pending_eval_queue_df,
) = get_evaluation_queue_df(eval_queue, eval_queue_private, EVAL_REQUESTS_PATH, EVAL_COLS)


## INTERACTION FUNCTIONS
def add_new_eval(
    model: str,
    base_model: str,
    revision: str,
    precision: str,
    private: bool,
    weight_type: str,
    model_type: str,
):
    precision = precision.split(" ")[0]
    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")

    if model_type is None or model_type == "":
        return styled_error("Please select a model type.")

    # check the model actually exists before adding the eval
    if revision == "":
        revision = "main"

    if weight_type in ["Delta", "Adapter"]:
        base_model_on_hub, error = is_model_on_hub(base_model, revision)
        if not base_model_on_hub:
            return styled_error(f'Base model "{base_model}" {error}')

    if not weight_type == "Adapter":
        model_on_hub, error = is_model_on_hub(model, revision)
        if not model_on_hub:
            return styled_error(f'Model "{model}" {error}')

    print("adding new eval")

    eval_entry = {
        "model": model,
        "base_model": base_model,
        "revision": revision,
        "private": private,
        "precision": precision,
        "weight_type": weight_type,
        "status": "PENDING",
        "submitted_time": current_time,
        "model_type": model_type,
    }

    user_name = ""
    model_path = model
    if "/" in model:
        user_name = model.split("/")[0]
        model_path = model.split("/")[1]

    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
    os.makedirs(OUT_DIR, exist_ok=True)
    out_path = f"{OUT_DIR}/{model_path}_eval_request_{private}_{precision}_{weight_type}.json"

    # Check if the model has been forbidden:
    if out_path.split("eval-queue/")[1] in DO_NOT_SUBMIT_MODELS:
        return styled_warning("Model authors have requested that their model be not submitted on the leaderboard.")

    # Check for duplicate submission
    if out_path.split("eval-queue/")[1].lower() in requested_models:
        return styled_warning("This model has been already submitted.")

    with open(out_path, "w") as f:
        f.write(json.dumps(eval_entry))

    api.upload_file(
        path_or_fileobj=out_path,
        path_in_repo=out_path.split("eval-queue/")[1],
        repo_id=QUEUE_REPO,
        repo_type="dataset",
        commit_message=f"Add {model} to eval queue",
    )

    # remove the local file
    os.remove(out_path)

    return styled_message(
        "Your request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
    )


# Basics
def refresh() -> list[pd.DataFrame]:
    leaderboard_df = get_leaderboard_df(eval_results, eval_results_private, COLS, BENCHMARK_COLS)
    (
        finished_eval_queue_df,
        running_eval_queue_df,
        pending_eval_queue_df,
    ) = get_evaluation_queue_df(eval_queue, eval_queue_private, EVAL_REQUESTS_PATH, COLS)
    return (
        leaderboard_df,
        finished_eval_queue_df,
        running_eval_queue_df,
        pending_eval_queue_df,
    )


def change_tab(query_param: str):
    query_param = query_param.replace("'", '"')
    query_param = json.loads(query_param)

    if isinstance(query_param, dict) and "tab" in query_param and query_param["tab"] == "evaluation":
        return gr.Tabs.update(selected=1)
    else:
        return gr.Tabs.update(selected=0)


# Searching and filtering
def search_table(df: pd.DataFrame, current_columns_df: pd.DataFrame, query: str) -> pd.DataFrame:
    current_columns = current_columns_df.columns
    if AutoEvalColumn.model_type.name in current_columns:
        filtered_df = df[
            (df[AutoEvalColumn.dummy.name].str.contains(query, case=False))
            | (df[AutoEvalColumn.model_type.name].str.contains(query, case=False))
        ]
    else:
        filtered_df = df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))]
    return filtered_df[current_columns]


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


def filter_models(
    df: pd.DataFrame, current_columns_df: pd.DataFrame, type_query: str, size_query: str, show_deleted: bool
) -> pd.DataFrame:
    current_columns = current_columns_df.columns

    # Show all models
    if show_deleted:
        filtered_df = df[current_columns]
    else:  # Show only still on the hub models
        filtered_df = df[df[AutoEvalColumn.still_on_hub.name] is True][current_columns]

    if type_query != "all":
        type_emoji = type_query[0]
        filtered_df = filtered_df[df[AutoEvalColumn.model_type_symbol.name] == type_emoji]

    if size_query != "all":
        numeric_intervals = {
            "all": (0, 10000),
            "< 1B": (0, 1),
            "~3B": (1, 5),
            "~7B": (6, 11),
            "~13B": (12, 15),
            "~35B": (16, 55),
            "60B+": (55, 10000),
        }
        numeric_interval = numeric_intervals[size_query]
        params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")

        filtered_df = filtered_df[params_column.between(*numeric_interval)]

    return filtered_df


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("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
            with gr.Row():
                with gr.Column():
                    with gr.Row():
                        shown_columns = gr.CheckboxGroup(
                            choices=[
                                c
                                for c in COLS
                                if c
                                not in [
                                    AutoEvalColumn.dummy.name,
                                    AutoEvalColumn.model.name,
                                    AutoEvalColumn.model_type_symbol.name,
                                    AutoEvalColumn.still_on_hub.name,
                                ]
                            ],
                            value=[
                                c
                                for c in COLS_LITE
                                if c
                                not in [
                                    AutoEvalColumn.dummy.name,
                                    AutoEvalColumn.model.name,
                                    AutoEvalColumn.model_type_symbol.name,
                                    AutoEvalColumn.still_on_hub.name,
                                ]
                            ],
                            label="Select columns to show",
                            elem_id="column-select",
                            interactive=True,
                        )
                    with gr.Row():
                        deleted_models_visibility = gr.Checkbox(
                            value=True, label="Show models removed from the hub", interactive=True
                        )
                with gr.Column(min_width=320):
                    search_bar = gr.Textbox(
                        placeholder="🔍 Search for your model and press ENTER...",
                        show_label=False,
                        elem_id="search-bar",
                    )
                    with gr.Box(elem_id="box-filter"):
                        filter_columns_type = gr.Radio(
                            label="⏚ Filter model types",
                            choices=[
                                "all",
                                ModelType.PT.to_str(),
                                ModelType.FT.to_str(),
                                ModelType.IFT.to_str(),
                                ModelType.RL.to_str(),
                            ],
                            value="all",
                            interactive=True,
                            elem_id="filter-columns-type",
                        )
                        filter_columns_size = gr.Radio(
                            label="⏚ Filter model sizes",
                            choices=[
                                "all",
                                "< 1B",
                                "~3B",
                                "~7B",
                                "~13B",
                                "~35B",
                                "60B+",
                            ],
                            value="all",
                            interactive=True,
                            elem_id="filter-columns-size",
                        )

            leaderboard_table = gr.components.Dataframe(
                value=leaderboard_df[
                    [AutoEvalColumn.model_type_symbol.name, AutoEvalColumn.model.name]
                    + shown_columns.value
                    + [AutoEvalColumn.dummy.name]
                ],
                headers=[
                    AutoEvalColumn.model_type_symbol.name,
                    AutoEvalColumn.model.name,
                ]
                + shown_columns.value
                + [AutoEvalColumn.dummy.name],
                datatype=TYPES,
                max_rows=None,
                elem_id="leaderboard-table",
                interactive=False,
                visible=True,
            )

            # Dummy leaderboard for handling the case when the user uses backspace key
            hidden_leaderboard_table_for_search = gr.components.Dataframe(
                value=original_df,
                headers=COLS,
                datatype=TYPES,
                max_rows=None,
                visible=False,
            )
            search_bar.submit(
                search_table,
                [
                    hidden_leaderboard_table_for_search,
                    leaderboard_table,
                    search_bar,
                ],
                leaderboard_table,
            )
            shown_columns.change(
                select_columns,
                [hidden_leaderboard_table_for_search, shown_columns],
                leaderboard_table,
                queue=False,
            )
            filter_columns_type.change(
                filter_models,
                [
                    hidden_leaderboard_table_for_search,
                    leaderboard_table,
                    filter_columns_type,
                    filter_columns_size,
                    deleted_models_visibility,
                ],
                leaderboard_table,
                queue=False,
            )
            filter_columns_size.change(
                filter_models,
                [
                    hidden_leaderboard_table_for_search,
                    leaderboard_table,
                    filter_columns_type,
                    filter_columns_size,
                    deleted_models_visibility,
                ],
                leaderboard_table,
                queue=False,
            )
            deleted_models_visibility.change(
                filter_models,
                [
                    hidden_leaderboard_table_for_search,
                    leaderboard_table,
                    filter_columns_type,
                    filter_columns_size,
                    deleted_models_visibility,
                ],
                leaderboard_table,
                queue=False,
            )
        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

        with gr.TabItem("🚀 Submit here! ", 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"✅ 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,
                                max_rows=5,
                            )
                    with gr.Accordion(
                        f"🔄 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,
                                max_rows=5,
                            )

                    with gr.Accordion(
                        f"⏳ Pending 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,
                                max_rows=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", placeholder="main")
                    private = gr.Checkbox(False, label="Private", visible=not IS_PUBLIC)
                    model_type = gr.Dropdown(
                        choices=[
                            ModelType.PT.to_str(" : "),
                            ModelType.FT.to_str(" : "),
                            ModelType.IFT.to_str(" : "),
                            ModelType.RL.to_str(" : "),
                        ],
                        label="Model type",
                        multiselect=False,
                        value=None,
                        interactive=True,
                    )

                with gr.Column():
                    precision = gr.Dropdown(
                        choices=[
                            "float16",
                            "bfloat16",
                            "8bit (LLM.int8)",
                            "4bit (QLoRA / FP4)",
                        ],
                        label="Precision",
                        multiselect=False,
                        value="float16",
                        interactive=True,
                    )
                    weight_type = gr.Dropdown(
                        choices=["Original", "Delta", "Adapter"],
                        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():
            refresh_button = gr.Button("Refresh")
            refresh_button.click(
                refresh,
                inputs=[],
                outputs=[
                    leaderboard_table,
                    finished_eval_table,
                    running_eval_table,
                    pending_eval_table,
                ],
            )

    with gr.Row():
        with gr.Accordion("📙 Citation", open=False):
            citation_button = gr.Textbox(
                value=CITATION_BUTTON_TEXT,
                label=CITATION_BUTTON_LABEL,
                elem_id="citation-button",
            ).style(show_copy_button=True)

    dummy = gr.Textbox(visible=False)
    demo.load(
        change_tab,
        dummy,
        tabs,
        _js=get_window_url_params,
    )

scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=3600)
scheduler.start()
demo.queue(concurrency_count=40).launch()