-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
1270 lines (1196 loc) · 110 KB
/
index.html
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
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html>
<head>
<script async src="https://www.googletagmanager.com/gtag/js?id=G-C1CRWDNJ1J"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-C1CRWDNJ1J');
</script>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0"><title>HF. 13 papers. March 6.</title>
<link rel="icon" href="favicon.svg" sizes="any" type="image/svg+xml">
<link href="https://fonts.googleapis.com/css2?family=Roboto:wght@300;400;700&display=swap" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Roboto+Slab:[email protected]&family=Tiny5&display=swap" rel="stylesheet">
<style>
:root {
--primary-color: cornflowerblue;
--primary-color-dark: #fffd87cf;
--secondary-color: #fff;
--background-color: #eee;
--text-color: #333333;
--header-color: cornflowerblue;
--body-color: #eee;
--menu-color: #002370;
}
.background-digit {
position: absolute;
font-family: 'Tiny5';
bottom: -20px;
right: -10px;
font-size: 8em;
font-weight: 400;
color: #0989ea22;
z-index: 2;
line-height: 1;
}
.dark-theme .background-digit {
color: #e9e78f3d;
}
body {
font-family: 'Roboto Slab', sans-serif;
line-height: 1.6;
color: var(--text-color);
margin: 0;
padding: 0;
min-height: 100vh;
display: flex;
flex-direction: column;
}
.container {
max-width: 1500px;
margin: 0 auto;
flex: 1 0 auto;
width: 100%
}
.a-clean {
color: var(--secondary-color);
text-decoration: none;
}
.a-clean:hover {
color: #fff;
}
header {
padding: 3.6em 0 2.4em 0;
text-align: center;
}
footer {
background-color: var(--primary-color);
color: white;
text-align: center;
margin-top: 2em;
flex-shrink: 0;
padding: 20px;
}
h1 {
font-size: 2.4em;
margin: 0;
font-weight: 700;
}
.article-title-cont {
margin: -21px -21px 0px -21px;
padding: 10px 20px;
background: cornflowerblue;
display: table;
min-height: 5.9em;
}
.dark-theme .article-title-cont {
background: #444444;
}
.article-title {
color: white;
}
.article-title h2 {
margin: 0px;
padding: 0px;
font-weight: 400;
text-align:center;
}
h2 {
# color: var(--primary-color);
font-size: 1.2em;
margin-top: 0;
margin-bottom: 0.5em;
}
header p {
font-size: 1.2em;
margin-top: 0.5em;
font-weight: 300;
}
main {
display: grid;
grid-template-columns: repeat(auto-fit, minmax(400px, 1fr));
gap: 1.5em;
padding: 10px 20px 20px 20px;
}
body.dark-tmeme>header {
background-color: background-color: #333333;
color: white;
}
body.dark-theme>div>main>article>div.article-content>p.meta {
color: #fff;
}
body.light-theme>div>main>article>div.article-content>p.meta {
color: #555;
}
body.dark-theme>div>main>article>div.article-content>p.pub-date {
color: #ccc;
}
body.light-theme>div>main>article>div.article-content>p.pub-date {
color: #555;
}
body.dark-theme>div>main>article>div.article-content>div.tags {
color: #ccc;
}
body.light-theme>div>main>article>div.article-content>div.tags {
color: #fff;
}
body.light-theme>header {
background-color: var(--header-color);
color: white;
}
article {
display: flex;
flex-direction: row;
justify-content: center;
}
.article-content {
border-radius: 5px;
border: 1px solid #ddd;
overflow: hidden;
transition: background-color 0.2s ease;
padding: 1.3em;
flex-grow: 1;
display: flex;
flex-direction: column;
position: relative;
z-index: 1;
cursor: pointer;
max-width: 800px;
position: relative;
}
body.dark-theme>div>main>article>div.article-content {
background-color: #444;
border: none;
}
body.light-theme>div>main>article>div.article-content {
background-color: #fff;
}
body.dark-theme>div>main>article>div.article-content:hover {
background-color: #414141;
}
body.light-theme>div>main>article>div.article-content:hover {
background-color: #fafafa;
}
.meta {
font-size: 0.9em;
margin-bottom: 0em;
font-weight: 500;
margin: 20px 0 0px 0;
padding-bottom: 20px;
border-bottom: 1px solid #ddd;
}
.pub-date {
font-size: 0.8em;
margin-bottom: 0.8em;
font-weight: 400;
text-align: right;
font-family: Roboto;
}
.tags {
font-size: 0.9em;
margin-bottom: 0;
position: absolute;
bottom: 0px;
font-weight: 300;
font-family: 'Roboto Slab';
background: #555;
left: 0;
width: 100%;
padding: 10px 20px;
}
.abstract {
position: relative;
max-height: 170px;
overflow: hidden;
transition: max-height 0.3s ease;
cursor: pointer;
}
.abstract.expanded {
max-height: 1000px;
}
.abstract-toggle {
position: absolute;
bottom: 4px;
right: 0;
cursor: pointer;
color: var(--primary-color);
float: right;
font-weight: 400;
}
.explanation {
background-color: #e8f5e9;
border-left: 4px solid var(--secondary-color);
padding: 1em;
margin-top: 1.5em;
}
.links {
margin-top: 1.5em;
margin-bottom: 20px;
}
.affiliations {
margin-bottom: 50px;
padding:10px;
font-size: 0.9em;
text-align: center
}
a {
color: var(--primary-color);
text-decoration: none;
font-weight: 500;
transition: color 0.3s ease;
}
.dark-theme a {
color: var(--primary-color-dark);
}
a:hover {
color: #e73838;
}
.light-theme {
background-color: var(--body-color);
color: #333333;
}
.dark-theme {
background-color: #333333;
color: #ffffff;
}
.theme-switch {
position: absolute;
top: 20px;
right: 20px;
display: flex;
align-items: center;
}
.switch {
position: relative;
display: inline-block;
width: 50px;
height: 30px;
}
.switch input {
opacity: 0;
width: 0;
height: 0;
}
.slider {
position: absolute;
cursor: pointer;
top: 0;
left: 0;
right: 0;
bottom: 0;
background-color: #ccc;
transition: .4s;
border-radius: 30px;
}
.slider:before {
position: absolute;
content: "";
height: 24px;
width: 24px;
left: 3px;
bottom: 3px;
background-color: white;
transition: .4s;
border-radius: 50%;
}
input:checked + .slider {
background-color: var(--primary-color);
}
input:checked + .slider:before {
transform: translateX(20px);
}
.switch-label {
margin-right: 10px;
}
.sub-header-container {
display: flex;
justify-content: space-between;
align-items: center;
flex-wrap: wrap;
gap: 15px;
margin-top: 7px;
padding: 0 20px;
}
.sub-header-container-2 {
display: flex;
justify-content: left;
align-items: center;
flex-wrap: wrap;
gap: 15px;
margin: 0 auto;
padding: 0 20px;
}
.update-info-container {
margin-top: 15px;
margin-bottom: 0px;
text-align: left;
flex: 1;
}
.sort-container {
margin-top: 15px;
margin-bottom: 0px;
text-align: right;
flex: 2;
}
.category-toggle-container {
display: inline-block;
margin-top: 15px;
margin-bottom: 10px;
cursor: pointer;
}
.category-option-container {
margin-top: 15px;
margin-bottom: 10px;
display: none;
margin-left: auto;
}
.category-option-container.expanded {
display: block;
}
.sort-dropdown {
padding: 5px 10px;
font-size: 16px;
border-radius: 5px;
border: 1px solid #ccc;
background-color: white;
color: var(--text-color);
font-family: 'Roboto Slab', sans-serif;
}
.sort-label {
margin-right: 10px;
font-size: 1.0em !important;
}
.dark-theme .sort-dropdown {
background-color: #444;
color: white;
border-color: var(--text-color);
}
.title-sign {
display: inline-block;
transition: all 0.5s ease;
}
.rotate {
transform: rotate(45deg) translateY(-6px);
transform-origin: center;
}
.title-text {
display: inline;
padding-left: 10px;
}
.summary_title {
font-size: 1.2em;
font-weight: bold;
color: #222;
margin-bottom: 5px;
}
.summary_text {
}
.summary_image {
max-height: 500px;
max-width: 100%;
align: center;
margin-top: 40px;
margin-bottom: 60px;
}
.category-filters {
margin-top: 20px;
margin-bottom: 20px;
text-align: center;
display: none;
}
.category-filters.expanded {
display: block;
margin-top: 10px;
}
.category-button {
display: inline-block;
margin: 5px;
padding: 5px 10px;
border-radius: 15px;
background-color: #f0f0f0;
color: #333;
cursor: pointer;
transition: background-color 0.3s ease;
}
.category-button.active {
background-color: var(--primary-color);
color: white;
}
.category-button.inactive:not(.active) {
color: #ccc;
}
.dark-theme .category-button {
background-color: #555;
color: #fff;
}
.dark-theme .category-button.active {
background-color: var(--primary-color);
}
.dark-theme .category-button.inactive:not(.active) {
color: #888;
}
.clear-categories {
display: inline-block;
margin: 5px;
padding: 5px 10px;
border-radius: 15px;
background-color: #f0f0f0;
color: #333;
cursor: pointer;
transition: background-color 0.3s ease;
}
.clear-categories:hover {
background-color: #bbb;
}
.svg-container {
display: inline-block;
position: relative;
overflow: hidden;
}
.svg-container span {
position: relative;
z-index: 1;
}
.svg-container svg {
position: absolute;
bottom: 0;
left: 0;
z-index: 0;
}
.nav-menu {
background-color: var(--menu-color);
padding: 2px 0 2px 0;
display: inline-block;
position: relative;
overflow: hidden;
width: 100%;
}
.nav-container {
max-width: 1500px;
margin: 0 auto;
display: flex;
justify-content: left;
gap: 3em;
}
.nav-container span a {
color: white;
}
.nav-item {
color: white;
padding: 3px 0px;
cursor: pointer;
font-weight: 400;
}
.nav-prev {
margin-left: 20px;
}
.nav-item:hover {
background-color: rgba(255, 255, 255, 0.1);
border-color: rgba(255, 255, 255, 0.3);
}
.language-flags {
display: flex;
gap: 7px;
padding: 5px 20px 0 0;
margin-left: auto;
}
.flag-svg {
width: 22px;
height: 22px;
cursor: pointer;
opacity: 0.4;
transition: opacity 0.3s ease;
border-radius: 2px;
}
.flag-svg.active {
opacity: 1;
}
.flag-svg:hover {
opacity: 0.8;
}
.dark-theme .nav-menu {
background-color: #333;
}
.dark-theme .nav-item {
color: white;
}
.dark-theme .nav-item:hover {
background-color: rgba(255, 255, 255, 0.05);
}
.pointer { cursor: pointer; }
.article-pdf-title-img {
max-width: 100%;
max-height: 400px;
display: inline-block;
margin-top: 10px;
margin-bottom: 10px;
border-radius: 5px;
}
.article-pdf-title-img-cont {
text-align: center;
}
.dark-theme .article-pdf-title-img {
opacity: 0.8;
filter: grayscale(1);
}
@media (max-width: 600px) {
.nav-container {
flex-direction: row;
gap: 1.5em;
}
.nav-item {
padding: 3px 0px;
}
}
@media (max-width: 768px) {
.category-filters {
display: none;
}
.category-toggle {
display: inline-block;
width: 100%;
text-align: left;
}
.category-filters.expanded {
display: block;
margin-top: 10px;
}
}
@media (max-width: 600px) {
.sub-header-container {
flex-direction: column;
align-items: flex-start;
}
.sort-container {
width: 100%;
display: flex;
justify-content: left;
margin: 0 auto;
}
.sort-dropdown {
margin-left: auto;
}
.sort-label {
margin-top: 5px;
float: left;
}
.sub-header-container-2 {
flex-direction: row;
align-items: flex-start;
}
.update-info-container {
text-align: left;
width: 100%;
margin-bottom: 0px;
}
.category-toggle-container {
margin-top: 15px;
text-align: left;
margin-bottom: 10px;
}
.category-option-container {
margin-top: 15px;
text-align: center;
margin-bottom: 10px;
}
main {
grid-template-columns: repeat(auto-fit);
gap: 0em;
padding: 10px 0 20px 0;
}
footer {
margin-top: -20px;
}
article>div.article-content {
border-radius: 0px;
}
}
</style>
<script>
function toggleAbstract(id) {
var abstract = document.getElementById('abstract-' + id);
var toggle = document.getElementById('toggle-' + id);
if (abstract.classList.contains('expanded')) {
abstract.classList.remove('expanded');
toggle.textContent = '...';
} else {
abstract.classList.add('expanded');
toggle.textContent = '';
}
}
function getTimeDiff(dateString, lang='ru') {
const timeUnits = {
ru: {
minute: ["минуту", "минуты", "минут"],
hour: ["час", "часа", "часов"],
day: ["день", "дня", "дней"],
justNow: "только что",
ago: "назад"
},
en: {
minute: ["minute", "minutes", "minutes"],
hour: ["hour", "hours", "hours"],
day: ["day", "days", "days"],
justNow: "just now",
ago: "ago"
},
zh: {
minute: ["分钟", "分钟", "分钟"],
hour: ["小时", "小时", "小时"],
day: ["天", "天", "天"],
justNow: "刚刚",
ago: "前"
}
};
function getPlural(number, words, lang) {
if (lang === 'ru') {
if (number % 10 === 1 && number % 100 !== 11) {
return words[0];
} else if (number % 10 >= 2 && number % 10 <= 4 && (number % 100 < 10 || number % 100 >= 20)) {
return words[1];
} else {
return words[2];
}
} else if (lang === 'en') {
return number === 1 ? words[0] : words[1];
} else {
// Chinese doesn't need plural forms
return words[0];
}
}
function formatTimeDiff(number, unit, lang) {
const unitWord = getPlural(number, timeUnits[lang][unit], lang);
if (lang === 'zh') {
return `${number}${unitWord}${timeUnits[lang].ago}`;
} else {
return `${number} ${unitWord} ${timeUnits[lang].ago}`;
}
}
if (!['ru', 'en', 'zh'].includes(lang)) {
throw new Error('Unsupported language. Supported languages are: ru, en, zh');
}
const pastDate = new Date(dateString.replace(" ", "T") + ":00Z");
const currentDate = new Date();
const diffInSeconds = Math.floor((currentDate - pastDate) / 1000);
const minutes = Math.floor(diffInSeconds / 60);
const hours = Math.floor(diffInSeconds / 3600);
const days = Math.floor(diffInSeconds / 86400);
if (minutes === 0) {
return timeUnits[lang].justNow;
} else if (minutes < 60) {
return formatTimeDiff(minutes, 'minute', lang);
} else if (hours < 24) {
return formatTimeDiff(hours, 'hour', lang);
} else {
return formatTimeDiff(days, 'day', lang);
}
}
function isToday(dateString) {
const inputDate = new Date(dateString);
const today = new Date();
return (
inputDate.getFullYear() === today.getFullYear() &&
inputDate.getMonth() === today.getMonth() &&
inputDate.getDate() === today.getDate()
);
}
function isCurrentMonth(dateString) {
const inputDate = new Date(dateString);
const today = new Date();
return (
inputDate.getFullYear() === today.getFullYear() &&
inputDate.getMonth() === today.getMonth()
);
}
function formatArticlesTitle(number, lang='ru') {
const lastDigit = number % 10;
const lastTwoDigits = number % 100;
let word;
if (!['ru', 'en', 'zh'].includes(lang)) {
throw new Error('Unsupported language. Supported languages are: ru, en, zh');
}
if (lang === 'ru') {
if (lastTwoDigits >= 11 && lastTwoDigits <= 14) {
word = "статей";
} else if (lastDigit === 1) {
word = "статья";
} else if (lastDigit >= 2 && lastDigit <= 4) {
word = "статьи";
} else {
word = "статей";
}
} else if (lang === 'en') {
if (number === 1) {
word = 'paper'
} else {
word = 'papers'
}
} else if (lang === 'zh') {
word = "篇论文"
}
if (lang === 'zh') {
return `${number}${word}`;
} else {
return `${number} ${word}`;
}
}
</script>
</head>
<body class="light-theme">
<header>
<div class="container">
<a href="https://hfday.ru" class="a-clean"><h1 class="title-sign" id="doomgrad-icon">🔺</h1><h1 class="title-text" id="doomgrad">hf daily</h1></a>
<p><span id="title-date">6 марта</span> | <span id="title-articles-count">13 papers</span></p>
</div>
<div class="theme-switch">
<label class="switch">
<input type="checkbox" id="theme-toggle">
<span class="slider"></span>
</label>
</div>
</header>
<div class="nav-menu">
<div class="nav-container">
<span class="nav-item nav-prev" id="nav-prev"><a href="/d/2025-03-05.html">⬅️ <span id="prev-date">05.03</span></a></span>
<span class="nav-item" id="nav-next"><a href="/d/2025-03-07.html">➡️ <span id="next-date">07.03</span></a></span>
<span class="nav-item" id="nav-monthly"><a href="/m/2025-03.html">📈 <span id='top-month-label'>Месяц</span></a></span>
<div class="language-flags">
<svg class="flag-svg" data-lang="ru" xmlns="http://www.w3.org/2000/svg" width="32" height="32" viewBox="0 0 32 32"><path fill="#1435a1" d="M1 11H31V21H1z"></path><path d="M5,4H27c2.208,0,4,1.792,4,4v4H1v-4c0-2.208,1.792-4,4-4Z" fill="#fff"></path><path d="M5,20H27c2.208,0,4,1.792,4,4v4H1v-4c0-2.208,1.792-4,4-4Z" transform="rotate(180 16 24)" fill="#c53a28"></path><path d="M27,4H5c-2.209,0-4,1.791-4,4V24c0,2.209,1.791,4,4,4H27c2.209,0,4-1.791,4-4V8c0-2.209-1.791-4-4-4Zm3,20c0,1.654-1.346,3-3,3H5c-1.654,0-3-1.346-3-3V8c0-1.654,1.346-3,3-3H27c1.654,0,3,1.346,3,3V24Z" opacity=".15"></path><path d="M27,5H5c-1.657,0-3,1.343-3,3v1c0-1.657,1.343-3,3-3H27c1.657,0,3,1.343,3,3v-1c0-1.657-1.343-3-3-3Z" fill="#fff" opacity=".2"></path></svg>
<svg class="flag-svg" data-lang="zh" xmlns="http://www.w3.org/2000/svg" width="32" height="32" viewBox="0 0 32 32"><rect x="1" y="4" width="30" height="24" rx="4" ry="4" fill="#db362f"></rect><path d="M27,4H5c-2.209,0-4,1.791-4,4V24c0,2.209,1.791,4,4,4H27c2.209,0,4-1.791,4-4V8c0-2.209-1.791-4-4-4Zm3,20c0,1.654-1.346,3-3,3H5c-1.654,0-3-1.346-3-3V8c0-1.654,1.346-3,3-3H27c1.654,0,3,1.346,3,3V24Z" opacity=".15"></path><path fill="#ff0" d="M7.958 10.152L7.19 7.786 6.421 10.152 3.934 10.152 5.946 11.614 5.177 13.979 7.19 12.517 9.202 13.979 8.433 11.614 10.446 10.152 7.958 10.152z"></path><path fill="#ff0" d="M12.725 8.187L13.152 8.898 13.224 8.072 14.032 7.886 13.269 7.562 13.342 6.736 12.798 7.361 12.035 7.037 12.461 7.748 11.917 8.373 12.725 8.187z"></path><path fill="#ff0" d="M14.865 10.372L14.982 11.193 15.37 10.46 16.187 10.602 15.61 10.007 15.997 9.274 15.253 9.639 14.675 9.044 14.793 9.865 14.048 10.23 14.865 10.372z"></path><path fill="#ff0" d="M15.597 13.612L16.25 13.101 15.421 13.13 15.137 12.352 14.909 13.149 14.081 13.179 14.769 13.642 14.541 14.439 15.194 13.928 15.881 14.391 15.597 13.612z"></path><path fill="#ff0" d="M13.26 15.535L13.298 14.707 12.78 15.354 12.005 15.062 12.46 15.754 11.942 16.402 12.742 16.182 13.198 16.875 13.236 16.047 14.036 15.827 13.26 15.535z"></path><path d="M27,5H5c-1.657,0-3,1.343-3,3v1c0-1.657,1.343-3,3-3H27c1.657,0,3,1.343,3,3v-1c0-1.657-1.343-3-3-3Z" fill="#fff" opacity=".2"></path></svg>
<svg class="flag-svg" data-lang="en" xmlns="http://www.w3.org/2000/svg" width="32" height="32" viewBox="0 0 32 32"><rect x="1" y="4" width="30" height="24" rx="4" ry="4" fill="#fff"></rect><path d="M1.638,5.846H30.362c-.711-1.108-1.947-1.846-3.362-1.846H5c-1.414,0-2.65,.738-3.362,1.846Z" fill="#a62842"></path><path d="M2.03,7.692c-.008,.103-.03,.202-.03,.308v1.539H31v-1.539c0-.105-.022-.204-.03-.308H2.03Z" fill="#a62842"></path><path fill="#a62842" d="M2 11.385H31V13.231H2z"></path><path fill="#a62842" d="M2 15.077H31V16.923000000000002H2z"></path><path fill="#a62842" d="M1 18.769H31V20.615H1z"></path><path d="M1,24c0,.105,.023,.204,.031,.308H30.969c.008-.103,.031-.202,.031-.308v-1.539H1v1.539Z" fill="#a62842"></path><path d="M30.362,26.154H1.638c.711,1.108,1.947,1.846,3.362,1.846H27c1.414,0,2.65-.738,3.362-1.846Z" fill="#a62842"></path><path d="M5,4h11v12.923H1V8c0-2.208,1.792-4,4-4Z" fill="#102d5e"></path><path d="M27,4H5c-2.209,0-4,1.791-4,4V24c0,2.209,1.791,4,4,4H27c2.209,0,4-1.791,4-4V8c0-2.209-1.791-4-4-4Zm3,20c0,1.654-1.346,3-3,3H5c-1.654,0-3-1.346-3-3V8c0-1.654,1.346-3,3-3H27c1.654,0,3,1.346,3,3V24Z" opacity=".15"></path><path d="M27,5H5c-1.657,0-3,1.343-3,3v1c0-1.657,1.343-3,3-3H27c1.657,0,3,1.343,3,3v-1c0-1.657-1.343-3-3-3Z" fill="#fff" opacity=".2"></path><path fill="#fff" d="M4.601 7.463L5.193 7.033 4.462 7.033 4.236 6.338 4.01 7.033 3.279 7.033 3.87 7.463 3.644 8.158 4.236 7.729 4.827 8.158 4.601 7.463z"></path><path fill="#fff" d="M7.58 7.463L8.172 7.033 7.441 7.033 7.215 6.338 6.989 7.033 6.258 7.033 6.849 7.463 6.623 8.158 7.215 7.729 7.806 8.158 7.58 7.463z"></path><path fill="#fff" d="M10.56 7.463L11.151 7.033 10.42 7.033 10.194 6.338 9.968 7.033 9.237 7.033 9.828 7.463 9.603 8.158 10.194 7.729 10.785 8.158 10.56 7.463z"></path><path fill="#fff" d="M6.066 9.283L6.658 8.854 5.927 8.854 5.701 8.158 5.475 8.854 4.744 8.854 5.335 9.283 5.109 9.979 5.701 9.549 6.292 9.979 6.066 9.283z"></path><path fill="#fff" d="M9.046 9.283L9.637 8.854 8.906 8.854 8.68 8.158 8.454 8.854 7.723 8.854 8.314 9.283 8.089 9.979 8.68 9.549 9.271 9.979 9.046 9.283z"></path><path fill="#fff" d="M12.025 9.283L12.616 8.854 11.885 8.854 11.659 8.158 11.433 8.854 10.702 8.854 11.294 9.283 11.068 9.979 11.659 9.549 12.251 9.979 12.025 9.283z"></path><path fill="#fff" d="M6.066 12.924L6.658 12.494 5.927 12.494 5.701 11.799 5.475 12.494 4.744 12.494 5.335 12.924 5.109 13.619 5.701 13.19 6.292 13.619 6.066 12.924z"></path><path fill="#fff" d="M9.046 12.924L9.637 12.494 8.906 12.494 8.68 11.799 8.454 12.494 7.723 12.494 8.314 12.924 8.089 13.619 8.68 13.19 9.271 13.619 9.046 12.924z"></path><path fill="#fff" d="M12.025 12.924L12.616 12.494 11.885 12.494 11.659 11.799 11.433 12.494 10.702 12.494 11.294 12.924 11.068 13.619 11.659 13.19 12.251 13.619 12.025 12.924z"></path><path fill="#fff" d="M13.539 7.463L14.13 7.033 13.399 7.033 13.173 6.338 12.947 7.033 12.216 7.033 12.808 7.463 12.582 8.158 13.173 7.729 13.765 8.158 13.539 7.463z"></path><path fill="#fff" d="M4.601 11.104L5.193 10.674 4.462 10.674 4.236 9.979 4.01 10.674 3.279 10.674 3.87 11.104 3.644 11.799 4.236 11.369 4.827 11.799 4.601 11.104z"></path><path fill="#fff" d="M7.58 11.104L8.172 10.674 7.441 10.674 7.215 9.979 6.989 10.674 6.258 10.674 6.849 11.104 6.623 11.799 7.215 11.369 7.806 11.799 7.58 11.104z"></path><path fill="#fff" d="M10.56 11.104L11.151 10.674 10.42 10.674 10.194 9.979 9.968 10.674 9.237 10.674 9.828 11.104 9.603 11.799 10.194 11.369 10.785 11.799 10.56 11.104z"></path><path fill="#fff" d="M13.539 11.104L14.13 10.674 13.399 10.674 13.173 9.979 12.947 10.674 12.216 10.674 12.808 11.104 12.582 11.799 13.173 11.369 13.765 11.799 13.539 11.104z"></path><path fill="#fff" d="M4.601 14.744L5.193 14.315 4.462 14.315 4.236 13.619 4.01 14.315 3.279 14.315 3.87 14.744 3.644 15.44 4.236 15.01 4.827 15.44 4.601 14.744z"></path><path fill="#fff" d="M7.58 14.744L8.172 14.315 7.441 14.315 7.215 13.619 6.989 14.315 6.258 14.315 6.849 14.744 6.623 15.44 7.215 15.01 7.806 15.44 7.58 14.744z"></path><path fill="#fff" d="M10.56 14.744L11.151 14.315 10.42 14.315 10.194 13.619 9.968 14.315 9.237 14.315 9.828 14.744 9.603 15.44 10.194 15.01 10.785 15.44 10.56 14.744z"></path><path fill="#fff" d="M13.539 14.744L14.13 14.315 13.399 14.315 13.173 13.619 12.947 14.315 12.216 14.315 12.808 14.744 12.582 15.44 13.173 15.01 13.765 15.44 13.539 14.744z"></path></svg>
</div>
</div>
</div>
<div class="container">
<div class="sub-header-container">
<div class="update-info-container">
<label class="update-info-label" id="timeDiff"></label>
</div>
<div class="sort-container">
<label class="sort-label">🔀 <span id="sort-label-text">Сортировка по</span></label>
<select id="sort-dropdown" class="sort-dropdown">
<option value="default">рейтингу</option>
<option value="pub_date">дате публикации</option>
<option value="issue_id">добавлению на HF</option>
</select>
</div>
</div>
<div class="sub-header-container-2">
<div class="category-toggle-container">
<div class="svg-container">
<span id="category-toggle">🏷️ Фильтр</span>
<svg height="3" width="200">
<line x1="0" y1="0" x2="200" y2="0"
stroke="black"
stroke-width="2"
stroke-dasharray="3, 3" />
</svg>
</div>
</div>
<div class="category-option-container" id="category-options">
<label class="pointer" for="filter-logic-or"><input type="radio" id="filter-logic-or" name="filter-logic" value="or"> A∪B</label>
<label class="pointer" for="filter-logic-and"><input type="radio" id="filter-logic-and" name="filter-logic" value="and"> A∩B</label>
</div>
</div>
<div class="category-filters" id="category-filters">
<span class="clear-categories" id="clear-categories">🧹</span>
<!-- Categories -->
</div>
<main id="articles-container">
<!-- Articles -->
</main>
</div>
<footer>
<div class="container">
<p><a style="color:white;" href="https://t.me/doomgrad">doomgrad</a> ✖️ <a style="color:white;" href="https://huggingface.co/papers">hugging face</a></p>
</div>
</footer>
<script>
// Language handling
let currentLang = localStorage.getItem('selectedLang') || 'en';
let feedDate = {'ru': '6 марта', 'en': 'March 6', 'zh': '3月6日'};
let feedDateNext = {'ru': '07.03', 'en': '03/07', 'zh': '3月7日'};
let feedDatePrev = {'ru': '05.03', 'en': '03/05', 'zh': '3月5日'};
let filterLabel = {'ru': 'Фильтр', 'en': 'Topics', 'zh': '主题筛选'}
let publishedLabel = {'ru': 'статья от ', 'en': 'published on ', 'zh': '发表于'}
let sortLabel = {'ru': 'Сортировка по', 'en': 'Sort by', 'zh': '排序方式'}
let paperLabel = {'ru': 'Статья', 'en': 'Paper', 'zh': '论文'}
let topMonthLabel = {'ru': 'Месяц', 'en': 'Month', 'zh': '月度论文'}
let topDayLabel = {'ru': 'День', 'en': 'Day', 'zh': '日度论文'}
function initializeLanguageFlags() {
const flags = document.querySelectorAll('.flag-svg');
flags.forEach(flag => {
if (flag.dataset.lang === currentLang) {
flag.classList.add('active');
}
flag.addEventListener('click', () => {
flags.forEach(f => f.classList.remove('active'));
flag.classList.add('active');
currentLang = flag.dataset.lang;
localStorage.setItem('selectedLang', currentLang);
updateTimeDiffs();
updateLocalization();
filterAndRenderArticles();
});
});
}
function toggleTheme() {
const body = document.body;
body.classList.toggle('light-theme');
body.classList.toggle('dark-theme');
const isDarkMode = body.classList.contains('dark-theme');
localStorage.setItem('darkMode', isDarkMode);
if (isDarkMode) {
const title = document.getElementById('doomgrad');
title.innerHTML = "hf nightly";
const titleSign = document.getElementById('doomgrad-icon');
titleSign.classList.add('rotate');
} else {
const title = document.getElementById('doomgrad');
title.innerHTML = "hf daily";
const titleSign = document.getElementById('doomgrad-icon');
titleSign.classList.remove('rotate');
}
}
const articlesData = [{'id': 'https://huggingface.co/papers/2503.00865', 'title': 'Babel: Open Multilingual Large Language Models Serving Over 90% of Global Speakers', 'url': 'https://huggingface.co/papers/2503.00865', 'abstract': "Large language models (LLMs) have revolutionized natural language processing (NLP), yet open-source multilingual LLMs remain scarce, with existing models often limited in language coverage. Such models typically prioritize well-resourced languages, while widely spoken but under-resourced languages are often overlooked. To address this disparity, we introduce Babel, an open multilingual LLM that covers the top 25 languages by number of speakers, supports over 90% of the global population, and includes many languages neglected by other open multilingual LLMs. Unlike traditional continue pretraining approaches, Babel expands its parameter count through a layer extension technique that elevates Babel's performance ceiling. We introduce two variants: Babel-9B, designed for efficient inference and fine-tuning, and Babel-83B, which sets a new standard for open multilingual LLMs. Extensive evaluations on multilingual tasks demonstrate its superior performance compared to open LLMs of comparable size. In addition, using open-source supervised fine-tuning datasets, Babel achieves remarkable performance, with Babel-9B-Chat leading among 10B-sized LLMs and Babel-83B-Chat setting a new standard for multilingual tasks, reaching the same level of commercial models.", 'score': 26, 'issue_id': 2555, 'pub_date': '2025-03-02', 'pub_date_card': {'ru': '2 марта', 'en': 'March 2', 'zh': '3月2日'}, 'hash': 'bc2424e709a2dd78', 'authors': ['Yiran Zhao', 'Chaoqun Liu', 'Yue Deng', 'Jiahao Ying', 'Mahani Aljunied', 'Zhaodonghui Li', 'Lidong Bing', 'Hou Pong Chan', 'Yu Rong', 'Deli Zhao', 'Wenxuan Zhang'], 'affiliations': ['DAMO Academy, Alibaba Group'], 'pdf_title_img': 'assets/pdf/title_img/2503.00865.jpg', 'data': {'categories': ['#low_resource', '#architecture', '#open_source', '#training', '#multilingual'], 'emoji': '🌍', 'ru': {'title': 'Babel: революция в многоязычном машинном обучении', 'desc': 'Представлена новая многоязычная языковая модель Babel, охватывающая 25 самых распространенных языков мира. Модель использует технику расширения слоев для улучшения производительности. Предложены две версии: Babel-9B для эффективного вывода и дообучения, и Babel-83B, устанавливающая новый стандарт для открытых многоязычных моделей. Обе версии показывают превосходные результаты в многоязычных задачах по сравнению с аналогичными открытыми моделями.'}, 'en': {'title': 'Babel: Bridging the Language Gap with Open Multilingual LLMs', 'desc': "This paper presents Babel, an innovative open-source multilingual large language model (LLM) that addresses the lack of coverage for under-resourced languages in existing models. Babel supports the top 25 languages spoken globally, reaching over 90% of the world's population, and includes many languages that are often neglected. The model employs a unique layer extension technique to increase its parameter count, enhancing its performance beyond traditional pretraining methods. With two variants, Babel-9B and Babel-83B, the model demonstrates superior performance on multilingual tasks, outperforming other open LLMs of similar size and achieving results comparable to commercial models."}, 'zh': {'title': 'Babel：打破语言壁垒的多语言模型', 'desc': '大型语言模型（LLMs）在自然语言处理（NLP）领域带来了革命性的变化，但开源的多语言LLMs仍然稀缺，现有模型通常在语言覆盖上有限。许多模型优先考虑资源丰富的语言，而广泛使用但资源不足的语言常常被忽视。为了解决这一差距，我们推出了Babel，一个开放的多语言LLM，覆盖全球前25种语言，支持超过90%的人口，并包括许多其他开源多语言LLMs忽视的语言。Babel通过层扩展技术增加参数数量，提升了性能，并推出了两个变体：Babel-9B和Babel-83B，后者在多语言任务中设定了新的标准。'}}}, {'id': 'https://huggingface.co/papers/2503.00329', 'title': 'ABC: Achieving Better Control of Multimodal Embeddings using VLMs', 'url': 'https://huggingface.co/papers/2503.00329', 'abstract': 'Visual embedding models excel at zero-shot tasks like visual retrieval and classification. However, these models cannot be used for tasks that contain ambiguity or require user instruction. These tasks necessitate a multimodal embedding model, which outputs embeddings that combine visual and natural language input. Existing CLIP-based approaches embed images and text independently, and fuse the result. We find that this results in weak interactions between modalities, and poor user control over the representation. We introduce ABC, an open-source multimodal embedding model that uses a vision-language model backbone to deeply integrate image features with natural language instructions. ABC achieves bestfor-size performance on MSCOCO image-to-text retrieval and is the top performing model on classification and VQA tasks in the Massive Multimodal Embedding Benchmark. With a strongly unified vision-language representation, ABC can use natural language to solve subtle and potentially ambiguous visual retrieval problems. To evaluate this capability, we design CtrlBench, a benchmark that requires interleaving textual instructions with image content for correct retrieval. ABC advances the state of multimodal embeddings by offering high-quality representations and flexible natural language control. Our model and datasets are available at our project page.', 'score': 8, 'issue_id': 2555, 'pub_date': '2025-03-01', 'pub_date_card': {'ru': '1 марта', 'en': 'March 1', 'zh': '3月1日'}, 'hash': '0483c542c8885777', 'authors': ['Benjamin Schneider', 'Florian Kerschbaum', 'Wenhu Chen'], 'affiliations': ['Cheriton School of Computer Science, University of Waterloo', 'Vector Institute, Toronto'], 'pdf_title_img': 'assets/pdf/title_img/2503.00329.jpg', 'data': {'categories': ['#benchmark', '#multimodal', '#open_source', '#dataset'], 'emoji': '🧠', 'ru': {'title': 'ABC: Мультимодальные встраивания с гибким языковым контролем', 'desc': 'Статья представляет новую мультимодальную модель встраивания под названием ABC, которая объединяет визуальные и текстовые данные. В отличие от существующих подходов, ABC использует глубокую интеграцию изображений и естественного языка. Модель демонстрирует высокую производительность в задачах поиска изображений по тексту и классификации. ABC также позволяет использовать естественный язык для решения сложных задач визуального поиска с неоднозначностями.'}, 'en': {'title': 'ABC: Unifying Vision and Language for Enhanced Multimodal Understanding', 'desc': "This paper presents ABC, a new multimodal embedding model that integrates visual and natural language inputs more effectively than existing CLIP-based methods. Unlike previous models that treat images and text separately, ABC combines these modalities deeply, allowing for better interaction and user control. The model excels in zero-shot tasks, particularly in image-to-text retrieval and classification, outperforming others in the Massive Multimodal Embedding Benchmark. To assess its capabilities, the authors introduce CtrlBench, a benchmark designed to evaluate the model's performance in handling complex visual retrieval tasks with natural language instructions."}, 'zh': {'title': 'ABC：多模态嵌入的新突破', 'desc': '这篇论文介绍了一种名为ABC的多模态嵌入模型，旨在解决视觉检索和分类中的模糊性问题。与现有的CLIP方法不同，ABC通过深度整合图像特征和自然语言指令，提供更强的模态交互。ABC在MSCOCO图像到文本检索任务中表现出色，并在分类和视觉问答任务中取得了最佳性能。通过设计CtrlBench基准，评估了ABC在处理复杂视觉检索问题时的能力，展示了其高质量的表示和灵活的自然语言控制。'}}}, {'id': 'https://huggingface.co/papers/2503.03751', 'title': 'GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control', 'url': 'https://huggingface.co/papers/2503.03751', 'abstract': 'We present GEN3C, a generative video model with precise Camera Control and temporal 3D Consistency. Prior video models already generate realistic videos, but they tend to leverage little 3D information, leading to inconsistencies, such as objects popping in and out of existence. Camera control, if implemented at all, is imprecise, because camera parameters are mere inputs to the neural network which must then infer how the video depends on the camera. In contrast, GEN3C is guided by a 3D cache: point clouds obtained by predicting the pixel-wise depth of seed images or previously generated frames. When generating the next frames, GEN3C is conditioned on the 2D renderings of the 3D cache with the new camera trajectory provided by the user. Crucially, this means that GEN3C neither has to remember what it previously generated nor does it have to infer the image structure from the camera pose. The model, instead, can focus all its generative power on previously unobserved regions, as well as advancing the scene state to the next frame. Our results demonstrate more precise camera control than prior work, as well as state-of-the-art results in sparse-view novel view synthesis, even in challenging settings such as driving scenes and monocular dynamic video. Results are best viewed in videos. Check out our webpage! https://research.nvidia.com/labs/toronto-ai/GEN3C/', 'score': 6, 'issue_id': 2555, 'pub_date': '2025-03-05', 'pub_date_card': {'ru': '5 марта', 'en': 'March 5', 'zh': '3月5日'}, 'hash': '8f5f2ad910a260c0', 'authors': ['Xuanchi Ren', 'Tianchang Shen', 'Jiahui Huang', 'Huan Ling', 'Yifan Lu', 'Merlin Nimier-David', 'Thomas Müller', 'Alexander Keller', 'Sanja Fidler', 'Jun Gao'], 'affiliations': ['NVIDIA', 'University of Toronto', 'Vector Institute'], 'pdf_title_img': 'assets/pdf/title_img/2503.03751.jpg', 'data': {'categories': ['#3d', '#video'], 'emoji': '🎥', 'ru': {'title': 'Точный контроль камеры и 3D-согласованность в генерации видео', 'desc': 'GEN3C - это генеративная модель видео с точным контролем камеры и временной 3D-согласованностью. Она использует 3D-кэш в виде облаков точек, полученных из глубинных карт исходных изображений или ранее сгенерированных кадров. При генерации следующих кадров GEN3C опирается на 2D-рендеринг 3D-кэша с новой траекторией камеры, заданной пользователем. Это позволяет модели сфокусироваться на ранее ненаблюдаемых областях и продвижении состояния сцены, не тратя ресурсы на запоминание предыдущих результатов или вывод структуры изображения из положения камеры.'}, 'en': {'title': 'GEN3C: Mastering Video Generation with 3D Precision and Camera Control', 'desc': 'GEN3C is a generative video model that enhances video generation by incorporating precise camera control and maintaining temporal 3D consistency. Unlike previous models that often lack 3D information, GEN3C utilizes a 3D cache of point clouds derived from depth predictions, allowing for more coherent object presence in videos. The model is conditioned on 2D renderings from this cache, enabling it to generate new frames without needing to remember past outputs or infer scene structure from camera angles. This approach results in superior camera control and state-of-the-art performance in generating novel views, particularly in complex scenarios like driving scenes.'}, 'zh': {'title': 'GEN3C：精确相机控制与时间一致性的视频生成模型', 'desc': '我们提出了GEN3C，这是一种具有精确相机控制和时间一致性的生成视频模型。以往的视频模型虽然能够生成逼真的视频，但往往缺乏3D信息，导致物体出现和消失的不一致性。GEN3C通过3D缓存来指导生成过程，利用从种子图像或先前生成帧中预测的像素深度获得的点云。这样，GEN3C能够在用户提供的新相机轨迹下，专注于生成未观察到的区域，并有效推进场景状态到下一个帧。'}}}, {'id': 'https://huggingface.co/papers/2503.02951', 'title': 'KodCode: A Diverse, Challenging, and Verifiable Synthetic Dataset for Coding', 'url': 'https://huggingface.co/papers/2503.02951', 'abstract': 'We introduce KodCode, a synthetic dataset that addresses the persistent challenge of acquiring high-quality, verifiable training data across diverse difficulties and domains for training Large Language Models for coding. Existing code-focused resources typically fail to ensure either the breadth of coverage (e.g., spanning simple coding tasks to advanced algorithmic problems) or verifiable correctness (e.g., unit tests). In contrast, KodCode comprises question-solution-test triplets that are systematically validated via a self-verification procedure. Our pipeline begins by synthesizing a broad range of coding questions, then generates solutions and test cases with additional attempts allocated to challenging problems. Finally, post-training data synthesis is done by rewriting questions into diverse formats and generating responses under a test-based reject sampling procedure from a reasoning model (DeepSeek R1). This pipeline yields a large-scale, robust and diverse coding dataset. KodCode is suitable for supervised fine-tuning and the paired unit tests also provide great potential for RL tuning. Fine-tuning experiments on coding benchmarks (HumanEval(+), MBPP(+), BigCodeBench, and LiveCodeBench) demonstrate that KodCode-tuned models achieve state-of-the-art performance, surpassing models like Qwen2.5-Coder-32B-Instruct and DeepSeek-R1-Distill-Llama-70B.', 'score': 6, 'issue_id': 2555, 'pub_date': '2025-03-04', 'pub_date_card': {'ru': '4 марта', 'en': 'March 4', 'zh': '3月4日'}, 'hash': '6c344ba0bf71ac84', 'authors': ['Zhangchen Xu', 'Yang Liu', 'Yueqin Yin', 'Mingyuan Zhou', 'Radha Poovendran'], 'affiliations': ['Microsoft', 'The University of Texas at Austin', 'University of Washington'], 'pdf_title_img': 'assets/pdf/title_img/2503.02951.jpg', 'data': {'categories': ['#dataset', '#rl', '#optimization', '#synthetic', '#training'], 'emoji': '🧑\u200d💻', 'ru': {'title': 'KodCode: Синтетические данные для обучения ИИ программированию', 'desc': 'KodCode - это синтетический набор данных для обучения больших языковых моделей программированию. Он состоит из триплетов вопрос-решение-тест, которые проходят процедуру самопроверки. Процесс создания KodCode включает синтез вопросов по программированию, генерацию решений и тестовых случаев, а также постобработку данных. Эксперименты показывают, что модели, обученные на KodCode, достигают наилучших результатов на различных бенчмарках по программированию.'}, 'en': {'title': 'KodCode: Elevating Coding Models with Verified Data', 'desc': 'KodCode is a synthetic dataset designed to improve the training of Large Language Models (LLMs) for coding tasks by providing high-quality, verifiable data. It includes question-solution-test triplets that are validated through a self-verification process, ensuring both correctness and a wide range of coding difficulties. The dataset is generated using a systematic pipeline that synthesizes coding questions, creates solutions, and develops test cases, particularly focusing on challenging problems. Fine-tuning experiments show that models trained on KodCode outperform existing models on various coding benchmarks, demonstrating its effectiveness in enhancing LLM performance.'}, 'zh': {'title': 'KodCode：高质量编码数据集的解决方案', 'desc': '我们介绍了KodCode，这是一个合成数据集，旨在解决获取高质量、可验证的训练数据的挑战，以训练大型语言模型进行编码。现有的代码资源通常无法确保覆盖范围广泛或正确性可验证。KodCode由问题-解决方案-测试三元组组成，通过自我验证程序系统地验证。我们的流程包括合成各种编码问题，生成解决方案和测试用例，并在后期通过重写问题和生成响应来进行数据合成，最终生成一个大规模、强大且多样化的编码数据集。'}}}, {'id': 'https://huggingface.co/papers/2503.03278', 'title': 'Enhancing Abnormality Grounding for Vision Language Models with Knowledge Descriptions', 'url': 'https://huggingface.co/papers/2503.03278', 'abstract': 'Visual Language Models (VLMs) have demonstrated impressive capabilities in visual grounding tasks. However, their effectiveness in the medical domain, particularly for abnormality detection and localization within medical images, remains underexplored. A major challenge is the complex and abstract nature of medical terminology, which makes it difficult to directly associate pathological anomaly terms with their corresponding visual features. In this work, we introduce a novel approach to enhance VLM performance in medical abnormality detection and localization by leveraging decomposed medical knowledge. Instead of directly prompting models to recognize specific abnormalities, we focus on breaking down medical concepts into fundamental attributes and common visual patterns. This strategy promotes a stronger alignment between textual descriptions and visual features, improving both the recognition and localization of abnormalities in medical images.We evaluate our method on the 0.23B Florence-2 base model and demonstrate that it achieves comparable performance in abnormality grounding to significantly larger 7B LLaVA-based medical VLMs, despite being trained on only 1.5% of the data used for such models. Experimental results also demonstrate the effectiveness of our approach in both known and previously unseen abnormalities, suggesting its strong generalization capabilities.', 'score': 3, 'issue_id': 2560, 'pub_date': '2025-03-05', 'pub_date_card': {'ru': '5 марта', 'en': 'March 5', 'zh': '3月5日'}, 'hash': '6103dbe5d60b5f3f', 'authors': ['Jun Li', 'Che Liu', 'Wenjia Bai', 'Rossella Arcucci', 'Cosmin I. Bercea', 'Julia A. Schnabel'], 'affiliations': ['Helmholtz AI and Helmholtz Munich, Germany', 'Imperial College London, UK', 'Kings College London, UK', 'Munich Center for Machine Learning, Germany', 'Technical University of Munich, Germany'], 'pdf_title_img': 'assets/pdf/title_img/2503.03278.jpg', 'data': {'categories': ['#cv', '#multimodal', '#healthcare', '#alignment', '#transfer_learning'], 'emoji': '🔬', 'ru': {'title': 'Декомпозиция медицинских знаний для повышения эффективности VLM в анализе медицинских изображений', 'desc': 'Эта статья представляет новый подход к улучшению работы визуальных языковых моделей (VLM) в обнаружении и локализации аномалий на медицинских изображениях. Вместо прямого распознавания конкретных патологий, метод фокусируется на разложении медицинских концепций на базовые атрибуты и общие визуальные паттерны. Это улучшает связь между текстовыми описаниями и визуальными характеристиками, повышая точность распознавания и локализации аномалий. Метод был протестирован на модели Florence-2 и показал результаты, сравнимые с гораздо более крупными медицинскими VLM, несмотря на использование значительно меньшего объема данных для обучения.'}, 'en': {'title': 'Enhancing Medical VLMs through Decomposed Knowledge', 'desc': 'This paper presents a new method to improve Visual Language Models (VLMs) for detecting and locating abnormalities in medical images. The authors address the challenge of complex medical terminology by breaking down medical concepts into simpler attributes and common visual patterns. This approach enhances the alignment between text descriptions and visual features, leading to better performance in recognizing and localizing abnormalities. The proposed method shows competitive results with larger models while using significantly less training data, indicating its efficiency and strong generalization capabilities.'}, 'zh': {'title': '分解医学知识，提升视觉语言模型的异常检测能力', 'desc': '视觉语言模型（VLMs）在视觉定位任务中表现出色，但在医学领域，尤其是医学图像中的异常检测和定位方面，仍然缺乏研究。医学术语的复杂性使得将病理异常术语与相应的视觉特征直接关联变得困难。我们提出了一种新方法，通过分解医学知识来增强VLM在医学异常检测和定位中的性能。该方法通过将医学概念分解为基本属性和常见视觉模式，促进了文本描述与视觉特征之间的更强对齐，从而提高了医学图像中异常的识别和定位能力。'}}}, {'id': 'https://huggingface.co/papers/2503.01836', 'title': 'CrowdSelect: Synthetic Instruction Data Selection with Multi-LLM Wisdom', 'url': 'https://huggingface.co/papers/2503.01836', 'abstract': "Distilling advanced Large Language Models' instruction-following capabilities into smaller models using a selected subset has become a mainstream approach in model training. While existing synthetic instruction data selection strategies rely mainly on single-dimensional signals (i.e., reward scores, model perplexity), they fail to capture the complexity of instruction-following across diverse fields. Therefore, we investigate more diverse signals to capture comprehensive instruction-response pair characteristics and propose three foundational metrics that leverage Multi-LLM wisdom, informed by (1) diverse LLM responses and (2) reward model assessment. Building upon base metrics, we propose CrowdSelect, an integrated metric incorporating a clustering-based approach to maintain response diversity. Our comprehensive experiments demonstrate that our foundation metrics consistently improve performance across 4 base models on MT-bench and Arena-Hard. CrowdSelect, efficiently incorporating all metrics, achieves state-of-the-art performance in both Full and LoRA fine-tuning, showing improvements of 4.81% on Arena-Hard and 11.1% on MT-bench with Llama-3.2-3b-instruct. We hope our findings will bring valuable insights for future research in this direction. Code are available at https://github.com/listentm/crowdselect.", 'score': 3, 'issue_id': 2560, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': 'd59d65fb3b60c043', 'authors': ['Yisen Li', 'Lingfeng Yang', 'Wenxuan Shen', 'Pan Zhou', 'Yao Wan', 'Weiwei Lin', 'Dongping Chen'], 'affiliations': ['Huazhong University of Science and Technology', 'South China University of Technology'], 'pdf_title_img': 'assets/pdf/title_img/2503.01836.jpg', 'data': {'categories': ['#small_models', '#training', '#synthetic', '#optimization'], 'emoji': '🧠', 'ru': {'title': 'CrowdSelect: умный отбор инструкций для обучения языковых моделей', 'desc': 'Статья описывает новый метод отбора инструкций для обучения языковых моделей, названный CrowdSelect. Он использует три основных метрики, основанные на оценках различных большиx языковых моделей и моделей вознаграждения. CrowdSelect также включает кластеризацию для сохранения разнообразия ответов. Эксперименты показали, что этот метод превосходит существующие подходы на бенчмарках MT-bench и Arena-Hard. Авторы надеются, что их исследование внесет вклад в развитие этого направления.'}, 'en': {'title': 'Enhancing Model Training with Diverse Instruction Metrics', 'desc': 'This paper focuses on improving the training of smaller models by distilling the instruction-following abilities of larger language models (LLMs). It critiques existing methods that use simple metrics for selecting synthetic instruction data, which do not adequately reflect the complexity of instruction-following tasks. The authors propose new metrics that utilize diverse responses from multiple LLMs and a reward model to better assess instruction-response pairs. Their method, CrowdSelect, combines these metrics with a clustering approach to enhance response diversity, leading to significant performance improvements in various model evaluations.'}, 'zh': {'title': '提升小模型的指令跟随能力', 'desc': '本论文探讨了如何将大型语言模型的指令跟随能力提炼到更小的模型中。现有的合成指令数据选择策略主要依赖单一维度的信号，未能全面捕捉指令跟随的复杂性。我们提出了三种基础指标，利用多种大型语言模型的智慧，结合多样的响应和奖励模型评估。通过综合实验，我们的CrowdSelect指标在多个基模型上表现出色，显著提升了性能，展示了未来研究的潜力。'}}}, {'id': 'https://huggingface.co/papers/2503.03044', 'title': 'QE4PE: Word-level Quality Estimation for Human Post-Editing', 'url': 'https://huggingface.co/papers/2503.03044', 'abstract': "Word-level quality estimation (QE) detects erroneous spans in machine translations, which can direct and facilitate human post-editing. While the accuracy of word-level QE systems has been assessed extensively, their usability and downstream influence on the speed, quality and editing choices of human post-editing remain understudied. Our QE4PE study investigates the impact of word-level QE on machine translation (MT) post-editing in a realistic setting involving 42 professional post-editors across two translation directions. We compare four error-span highlight modalities, including supervised and uncertainty-based word-level QE methods, for identifying potential errors in the outputs of a state-of-the-art neural MT model. Post-editing effort and productivity are estimated by behavioral logs, while quality improvements are assessed by word- and segment-level human annotation. We find that domain, language and editors' speed are critical factors in determining highlights' effectiveness, with modest differences between human-made and automated QE highlights underlining a gap between accuracy and usability in professional workflows.", 'score': 1, 'issue_id': 2560, 'pub_date': '2025-03-04', 'pub_date_card': {'ru': '4 марта', 'en': 'March 4', 'zh': '3月4日'}, 'hash': 'e4d3d7db506b6e1c', 'authors': ['Gabriele Sarti', 'Vilém Zouhar', 'Grzegorz Chrupała', 'Ana Guerberof-Arenas', 'Malvina Nissim', 'Arianna Bisazza'], 'affiliations': ['CLCG, University of Groningen', 'CSAI, Tilburg University', 'ETH Zürich'], 'pdf_title_img': 'assets/pdf/title_img/2503.03044.jpg', 'data': {'categories': ['#translation', '#data', '#multilingual', '#healthcare'], 'emoji': '🔍', 'ru': {'title': 'Оценка качества перевода: мост между точностью и практичностью', 'desc': 'Статья исследует влияние оценки качества перевода на уровне слов (word-level QE) на процесс постредактирования машинного перевода. В исследовании участвовали 42 профессиональных редактора, работавших с двумя направлениями перевода. Сравнивались четыре модальности подсветки ошибок, включая методы на основе обучения с учителем и неопределенности. Результаты показывают, что эффективность подсветки зависит от домена, языка и скорости работы редакторов, при этом разница между ручной и автоматической QE оказалась незначительной.'}, 'en': {'title': 'Enhancing Post-Editing Efficiency with Word-Level Quality Estimation', 'desc': 'This paper explores how word-level quality estimation (QE) can help improve the efficiency of human post-editing in machine translation (MT). It examines the effectiveness of different methods for highlighting potential translation errors, comparing supervised and uncertainty-based approaches. The study involves 42 professional post-editors and assesses their editing speed and quality improvements through detailed behavioral logs and human annotations. The findings reveal that factors like domain, language, and editor speed significantly influence the effectiveness of error highlights, indicating a need to bridge the gap between the accuracy of QE systems and their practical usability in real-world editing tasks.'}, 'zh': {'title': '提升机器翻译后编辑效率的关键', 'desc': '本文研究了词级质量评估（QE）在机器翻译后编辑中的影响。我们分析了42名专业后编辑在两种翻译方向下的表现，比较了四种错误范围高亮方式，包括监督和基于不确定性的词级QE方法。研究发现，领域、语言和编辑速度是影响高亮效果的关键因素。结果表明，人工和自动QE高亮之间存在适度差异，突显了专业工作流程中准确性与可用性之间的差距。'}}}, {'id': 'https://huggingface.co/papers/2503.01729', 'title': 'FLAME: A Federated Learning Benchmark for Robotic Manipulation', 'url': 'https://huggingface.co/papers/2503.01729', 'abstract': 'Recent progress in robotic manipulation has been fueled by large-scale datasets collected across diverse environments. Training robotic manipulation policies on these datasets is traditionally performed in a centralized manner, raising concerns regarding scalability, adaptability, and data privacy. While federated learning enables decentralized, privacy-preserving training, its application to robotic manipulation remains largely unexplored. We introduce FLAME (Federated Learning Across Manipulation Environments), the first benchmark designed for federated learning in robotic manipulation. FLAME consists of: (i) a set of large-scale datasets of over 160,000 expert demonstrations of multiple manipulation tasks, collected across a wide range of simulated environments; (ii) a training and evaluation framework for robotic policy learning in a federated setting. We evaluate standard federated learning algorithms in FLAME, showing their potential for distributed policy learning and highlighting key challenges. Our benchmark establishes a foundation for scalable, adaptive, and privacy-aware robotic learning.', 'score': 1, 'issue_id': 2558, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': '893358a382c79250', 'authors': ['Santiago Bou Betran', 'Alberta Longhini', 'Miguel Vasco', 'Yuchong Zhang', 'Danica Kragic'], 'affiliations': ['KTH Royal Institute of Technology, Stockholm, Sweden'], 'pdf_title_img': 'assets/pdf/title_img/2503.01729.jpg', 'data': {'categories': ['#robotics', '#benchmark', '#dataset'], 'emoji': '🤖', 'ru': {'title': 'Федеративное обучение для масштабируемой и конфиденциальной робототехники', 'desc': 'Статья представляет FLAME - первый бенчмарк для федеративного обучения в робототехнической манипуляции. FLAME включает в себя большой набор данных с более чем 160 000 экспертных демонстраций различных задач манипуляции, собранных в симулированных средах. Бенчмарк также предоставляет фреймворк для обучения и оценки робототехнических политик в федеративной среде. Авторы оценивают стандартные алгоритмы федеративного обучения на FLAME, демонстрируя их потенциал для распределенного обучения политик и выявляя ключевые проблемы.'}, 'en': {'title': 'Empowering Robots with Federated Learning for Privacy and Scalability', 'desc': 'This paper presents FLAME, a benchmark for applying federated learning to robotic manipulation tasks. It addresses the limitations of centralized training by allowing robots to learn from diverse datasets while preserving data privacy. FLAME includes over 160,000 expert demonstrations from various simulated environments, facilitating decentralized training. The study evaluates existing federated learning algorithms, demonstrating their effectiveness and identifying challenges in distributed policy learning for robotics.'}, 'zh': {'title': '联邦学习助力机器人操控的未来', 'desc': '这篇论文介绍了FLAME（跨操控环境的联邦学习），这是一个为机器人操控设计的基准测试。FLAME包含超过160,000个专家演示的大规模数据集，涵盖多种操控任务，收集自多种模拟环境。通过在FLAME中评估标准的联邦学习算法，论文展示了分布式策略学习的潜力，并指出了关键挑战。该基准为可扩展、适应性强且注重隐私的机器人学习奠定了基础。'}}}, {'id': 'https://huggingface.co/papers/2503.01378', 'title': 'CognitiveDrone: A VLA Model and Evaluation Benchmark for Real-Time Cognitive Task Solving and Reasoning in UAVs', 'url': 'https://huggingface.co/papers/2503.01378', 'abstract': 'This paper introduces CognitiveDrone, a novel Vision-Language-Action (VLA) model tailored for complex Unmanned Aerial Vehicles (UAVs) tasks that demand advanced cognitive abilities. Trained on a dataset comprising over 8,000 simulated flight trajectories across three key categories-Human Recognition, Symbol Understanding, and Reasoning-the model generates real-time 4D action commands based on first-person visual inputs and textual instructions. To further enhance performance in intricate scenarios, we propose CognitiveDrone-R1, which integrates an additional Vision-Language Model (VLM) reasoning module to simplify task directives prior to high-frequency control. Experimental evaluations using our open-source benchmark, CognitiveDroneBench, reveal that while a racing-oriented model (RaceVLA) achieves an overall success rate of 31.3%, the base CognitiveDrone model reaches 59.6%, and CognitiveDrone-R1 attains a success rate of 77.2%. These results demonstrate improvements of up to 30% in critical cognitive tasks, underscoring the effectiveness of incorporating advanced reasoning capabilities into UAV control systems. Our contributions include the development of a state-of-the-art VLA model for UAV control and the introduction of the first dedicated benchmark for assessing cognitive tasks in drone operations. The complete repository is available at cognitivedrone.github.io', 'score': 1, 'issue_id': 2558, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': '8a4aab69ce92453d', 'authors': ['Artem Lykov', 'Valerii Serpiva', 'Muhammad Haris Khan', 'Oleg Sautenkov', 'Artyom Myshlyaev', 'Grik Tadevosyan', 'Yasheerah Yaqoot', 'Dzmitry Tsetserukou'], 'affiliations': ['Intelligent Space Robotics Laboratory, Science Center for Digital Engineering, Technology. Skolkovo Institute'], 'pdf_title_img': 'assets/pdf/title_img/2503.01378.jpg', 'data': {'categories': ['#open_source', '#reasoning', '#benchmark', '#dataset', '#multimodal', '#cv'], 'emoji': '🚁', 'ru': {'title': 'Умные дроны: когнитивное управление БПЛА с помощью ИИ', 'desc': 'В статье представлена модель CognitiveDrone - новая модель Зрение-Язык-Действие (VLA) для сложных задач беспилотных летательных аппаратов (БПЛА). Модель обучена на наборе данных из более чем 8000 симулированных полетов и генерирует команды действий в реальном времени на основе визуальных входных данных и текстовых инструкций. Усовершенствованная версия CognitiveDrone-R1 включает дополнительный модуль рассуждений на основе Модели Зрения-Языка (VLM) для упрощения сложных задач. Экспериментальная оценка показывает, что CognitiveDrone-R1 достигает успешности выполнения задач в 77.2%, что на 30% лучше базовых моделей в критических когнитивных задачах.'}, 'en': {'title': 'CognitiveDrone: Elevating UAV Intelligence with Vision-Language-Action!', 'desc': 'This paper presents CognitiveDrone, a new Vision-Language-Action (VLA) model designed for complex tasks performed by Unmanned Aerial Vehicles (UAVs). It is trained on a dataset of over 8,000 simulated flight paths focusing on Human Recognition, Symbol Understanding, and Reasoning. The model can generate real-time 4D action commands from visual inputs and text instructions, with an enhanced version, CognitiveDrone-R1, that includes a Vision-Language Model (VLM) reasoning module for better task management. Experimental results show significant performance improvements, with CognitiveDrone-R1 achieving a 77.2% success rate, highlighting the importance of advanced reasoning in UAV operations.'}, 'zh': {'title': '智能无人机的认知飞行新纪元', 'desc': '本文介绍了一种名为CognitiveDrone的新型视觉-语言-行动（VLA）模型，专为复杂的无人机任务设计，具备高级认知能力。该模型在超过8000条模拟飞行轨迹的数据集上进行训练，涵盖人类识别、符号理解和推理三个关键类别。CognitiveDrone-R1通过集成额外的视觉-语言模型（VLM）推理模块，进一步提升在复杂场景中的表现。实验结果显示，CognitiveDrone模型的成功率达到59.6%，而CognitiveDrone-R1的成功率更是高达77.2%，证明了将高级推理能力融入无人机控制系统的有效性。'}}}, {'id': 'https://huggingface.co/papers/2503.00502', 'title': 'Interact, Instruct to Improve: A LLM-Driven Parallel Actor-Reasoner Framework for Enhancing Autonomous Vehicle Interactions', 'url': 'https://huggingface.co/papers/2503.00502', 'abstract': "Autonomous Vehicles (AVs) have entered the commercialization stage, but their limited ability to interact and express intentions still poses challenges in interactions with Human-driven Vehicles (HVs). Recent advances in large language models (LLMs) enable bidirectional human-machine communication, but the conflict between slow inference speed and the need for real-time decision-making challenges practical deployment. To address these issues, this paper introduces a parallel Actor-Reasoner framework designed to enable explicit bidirectional AV-HV interactions across multiple scenarios. First, by facilitating interactions between the LLM-driven Reasoner and heterogeneous simulated HVs during training, an interaction memory database, referred to as the Actor, is established. Then, by introducing the memory partition module and the two-layer memory retrieval module, the Actor's ability to handle heterogeneous HVs is significantly enhanced. Ablation studies and comparisons with other decision-making methods demonstrate that the proposed Actor-Reasoner framework significantly improves safety and efficiency. Finally, with the combination of the external Human-Machine Interface (eHMI) information derived from Reasoner's reasoning and the feasible action solutions retrieved from the Actor, the effectiveness of the proposed Actor-Reasoner is confirmed in multi-scenario field interactions. Our code is available at https://github.com/FanGShiYuu/Actor-Reasoner.", 'score': 1, 'issue_id': 2555, 'pub_date': '2025-03-01', 'pub_date_card': {'ru': '1 марта', 'en': 'March 1', 'zh': '3月1日'}, 'hash': 'd184a5cae68093d5', 'authors': ['Shiyu Fang', 'Jiaqi Liu', 'Chengkai Xu', 'Chen Lv', 'Peng Hang', 'Jian Sun'], 'affiliations': ['College of Transportation, Tongji University, Shanghai 201804, China', 'Nanyang Technological University, 639798, Singapore', 'State Key Lab of Intelligent Transportation System, Beijing 100088, China'], 'pdf_title_img': 'assets/pdf/title_img/2503.00502.jpg', 'data': {'categories': ['#rl', '#robotics', '#inference', '#optimization', '#agents', '#reasoning'], 'emoji': '🚗', 'ru': {'title': 'Интеллектуальное взаимодействие автономных и обычных автомобилей с помощью больших языковых моделей', 'desc': 'Эта статья представляет новую архитектуру Actor-Reasoner для улучшения взаимодействия между автономными и управляемыми человеком транспортными средствами. Авторы используют большие языковые модели для создания базы данных взаимодействий и двухуровневую систему извлечения памяти для работы с разнородными транспортными средствами. Предложенный подход значительно повышает безопасность и эффективность принятия решений по сравнению с другими методами. Эксперименты в реальных условиях подтверждают эффективность предложенной архитектуры Actor-Reasoner в различных сценариях.'}, 'en': {'title': 'Enhancing AV-HV Interactions with the Actor-Reasoner Framework', 'desc': "This paper presents a new framework called the Actor-Reasoner to improve interactions between Autonomous Vehicles (AVs) and Human-driven Vehicles (HVs). It leverages large language models (LLMs) to facilitate real-time communication and decision-making, addressing the challenge of slow inference speeds. The framework includes an interaction memory database, which enhances the AV's ability to understand and respond to various HV behaviors. Experimental results show that this approach significantly boosts both safety and efficiency in multi-scenario driving situations."}, 'zh': {'title': '提升自动驾驶与人类驾驶互动的智能框架', 'desc': '这篇论文介绍了一种新的并行演员-推理器框架，旨在改善自动驾驶汽车（AV）与人类驾驶汽车（HV）之间的互动。通过在训练过程中促进大语言模型（LLM）驱动的推理器与不同类型的模拟HV之间的互动，建立了一个互动记忆数据库。引入记忆分区模块和双层记忆检索模块后，演员的处理能力得到了显著提升。实验结果表明，该框架在多场景交互中显著提高了安全性和效率。'}}}, {'id': 'https://huggingface.co/papers/2503.01763', 'title': "Retrieval Models Aren't Tool-Savvy: Benchmarking Tool Retrieval for Large Language Models", 'url': 'https://huggingface.co/papers/2503.01763', 'abstract': 'Tool learning aims to augment large language models (LLMs) with diverse tools, enabling them to act as agents for solving practical tasks. Due to the limited context length of tool-using LLMs, adopting information retrieval (IR) models to select useful tools from large toolsets is a critical initial step. However, the performance of IR models in tool retrieval tasks remains underexplored and unclear. Most tool-use benchmarks simplify this step by manually pre-annotating a small set of relevant tools for each task, which is far from the real-world scenarios. In this paper, we propose ToolRet, a heterogeneous tool retrieval benchmark comprising 7.6k diverse retrieval tasks, and a corpus of 43k tools, collected from existing datasets. We benchmark six types of models on ToolRet. Surprisingly, even the models with strong performance in conventional IR benchmarks, exhibit poor performance on ToolRet. This low retrieval quality degrades the task pass rate of tool-use LLMs. As a further step, we contribute a large-scale training dataset with over 200k instances, which substantially optimizes the tool retrieval ability of IR models.', 'score': 0, 'issue_id': 2558, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': 'e6a23582f741dc5b', 'authors': ['Zhengliang Shi', 'Yuhan Wang', 'Lingyong Yan', 'Pengjie Ren', 'Shuaiqiang Wang', 'Dawei Yin', 'Zhaochun Ren'], 'affiliations': ['Baidu Inc., Beijing, China', 'Leiden University, Leiden, The Netherlands', 'Shandong University, Qingdao, China'], 'pdf_title_img': 'assets/pdf/title_img/2503.01763.jpg', 'data': {'categories': ['#training', '#optimization', '#benchmark', '#dataset', '#data'], 'emoji': '🔍', 'ru': {'title': 'ToolRet: Новый вызов для моделей поиска инструментов ИИ', 'desc': 'ToolRet - это новый эталонный тест для оценки поиска инструментов в контексте обучения инструментам для больших языковых моделей (LLM). Он включает 7,6 тысяч разнообразных задач поиска и корпус из 43 тысяч инструментов. Исследование показало, что даже модели с высокой производительностью в традиционных тестах информационного поиска плохо справляются с ToolRet. Авторы также предоставили обучающий набор данных из более чем 200 тысяч примеров для улучшения способностей моделей к поиску инструментов.'}, 'en': {'title': 'Enhancing Tool Retrieval for Language Models with ToolRet', 'desc': 'This paper introduces ToolRet, a benchmark designed to evaluate the effectiveness of information retrieval (IR) models in selecting tools for large language models (LLMs) in practical tasks. The authors highlight that existing benchmarks often rely on a limited set of pre-annotated tools, which does not reflect real-world complexities. Their findings reveal that even high-performing IR models struggle with tool retrieval in this new context, leading to lower task success rates for LLMs. To address this issue, they provide a large-scale training dataset that significantly enhances the tool retrieval capabilities of IR models.'}, 'zh': {'title': '工具检索：提升LLMs的实用能力', 'desc': '本文探讨了工具学习如何增强大型语言模型（LLMs）的能力，使其能够作为代理解决实际任务。由于工具使用的LLMs具有有限的上下文长度，因此采用信息检索（IR）模型从大量工具集中选择有用工具是关键的初步步骤。我们提出了ToolRet，一个包含7.6k多样化检索任务和43k工具的异构工具检索基准，旨在评估IR模型在工具检索任务中的表现。研究发现，即使在传统IR基准上表现良好的模型，在ToolRet上的表现却很差，这降低了工具使用LLMs的任务通过率。'}}}, {'id': 'https://huggingface.co/papers/2503.01449', 'title': 'Benchmarking Large Language Models for Multi-Language Software Vulnerability Detection', 'url': 'https://huggingface.co/papers/2503.01449', 'abstract': 'Recent advancements in generative AI have led to the widespread adoption of large language models (LLMs) in software engineering, addressing numerous long-standing challenges. However, a comprehensive study examining the capabilities of LLMs in software vulnerability detection (SVD), a crucial aspect of software security, is currently lacking. Existing research primarily focuses on evaluating LLMs using C/C++ datasets. It typically explores only one or two strategies among prompt engineering, instruction tuning, and sequence classification fine-tuning for open-source LLMs. Consequently, there is a significant knowledge gap regarding the effectiveness of diverse LLMs in detecting vulnerabilities across various programming languages. To address this knowledge gap, we present a comprehensive empirical study evaluating the performance of LLMs on the SVD task. We have compiled a comprehensive dataset comprising 8,260 vulnerable functions in Python, 7,505 in Java, and 28,983 in JavaScript. We assess five open-source LLMs using multiple approaches, including prompt engineering, instruction tuning, and sequence classification fine-tuning. These LLMs are benchmarked against five fine-tuned small language models and two open-source static application security testing tools. Furthermore, we explore two avenues to improve LLM performance on SVD: a) Data perspective: Retraining models using downsampled balanced datasets. b) Model perspective: Investigating ensemble learning methods that combine predictions from multiple LLMs. Our comprehensive experiments demonstrate that SVD remains a challenging task for LLMs. This study provides a thorough understanding of the role of LLMs in SVD and offers practical insights for future advancements in leveraging generative AI to enhance software security practices.', 'score': 0, 'issue_id': 2558, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': '1b4593bb9d78ec53', 'authors': ['Ting Zhang', 'Chengran Yang', 'Yindu Su', 'Martin Weyssow', 'Hung Nguyen', 'Tan Bui', 'Hong Jin Kang', 'Yikun Li', 'Eng Lieh Ouh', 'Lwin Khin Shar', 'David Lo'], 'affiliations': ['School of Computer Science, University of Sydney, Australia', 'School of Computing and Information Systems, Singapore Management University, Singapore'], 'pdf_title_img': 'assets/pdf/title_img/2503.01449.jpg', 'data': {'categories': ['#open_source', '#plp', '#training', '#security', '#benchmark', '#dataset', '#data'], 'emoji': '🛡️', 'ru': {'title': 'LLM на страже безопасности кода: новые горизонты в обнаружении уязвимостей', 'desc': 'Статья представляет комплексное исследование возможностей больших языковых моделей (LLM) в обнаружении уязвимостей программного обеспечения (SVD). Авторы оценивают производительность пяти открытых LLM на наборах данных, включающих уязвимые функции на Python, Java и JavaScript, используя различные подходы, такие как инженерия промптов, настройка инструкций и тонкая настройка классификации последовательностей. Исследование также изучает способы улучшения производительности LLM в SVD, включая переобучение на сбалансированных наборах данных и использование ансамблевых методов обучения. Результаты показывают, что SVD остается сложной задачей для LLM, предоставляя ценные insights для будущих разработок в области применения генеративного ИИ для повышения безопасности программного обеспечения.'}, 'en': {'title': 'Unlocking LLMs for Software Vulnerability Detection', 'desc': "This paper investigates the effectiveness of large language models (LLMs) in detecting software vulnerabilities, an important area for software security. It highlights the lack of comprehensive studies on LLMs' capabilities across various programming languages, as most existing research focuses on C/C++ datasets. The authors present an empirical study using a dataset of over 44,000 vulnerable functions from Python, Java, and JavaScript, evaluating five open-source LLMs with different strategies like prompt engineering and instruction tuning. The findings reveal that while LLMs show promise, software vulnerability detection remains a challenging task, providing valuable insights for future improvements in this field."}, 'zh': {'title': '提升软件安全：大型语言模型在漏洞检测中的应用', 'desc': '最近生成性人工智能的进展使得大型语言模型（LLMs）在软件工程中得到了广泛应用，解决了许多长期存在的挑战。然而，目前缺乏对LLMs在软件漏洞检测（SVD）能力的全面研究，这对软件安全至关重要。现有研究主要集中在使用C/C++数据集评估LLMs，通常只探讨了提示工程、指令调优和序列分类微调中的一两种策略。因此，我们进行了一项全面的实证研究，评估LLMs在不同编程语言中检测漏洞的有效性。'}}}, {'id': 'https://huggingface.co/papers/2503.01372', 'title': 'SwiLTra-Bench: The Swiss Legal Translation Benchmark', 'url': 'https://huggingface.co/papers/2503.01372', 'abstract': "In Switzerland legal translation is uniquely important due to the country's four official languages and requirements for multilingual legal documentation. However, this process traditionally relies on professionals who must be both legal experts and skilled translators -- creating bottlenecks and impacting effective access to justice. To address this challenge, we introduce SwiLTra-Bench, a comprehensive multilingual benchmark of over 180K aligned Swiss legal translation pairs comprising laws, headnotes, and press releases across all Swiss languages along with English, designed to evaluate LLM-based translation systems. Our systematic evaluation reveals that frontier models achieve superior translation performance across all document types, while specialized translation systems excel specifically in laws but under-perform in headnotes. Through rigorous testing and human expert validation, we demonstrate that while fine-tuning open SLMs significantly improves their translation quality, they still lag behind the best zero-shot prompted frontier models such as Claude-3.5-Sonnet. Additionally, we present SwiLTra-Judge, a specialized LLM evaluation system that aligns best with human expert assessments.", 'score': 0, 'issue_id': 2558, 'pub_date': '2025-03-03', 'pub_date_card': {'ru': '3 марта', 'en': 'March 3', 'zh': '3月3日'}, 'hash': '3de5be81537fa0fd', 'authors': ['Joel Niklaus', 'Jakob Merane', 'Luka Nenadic', 'Sina Ahmadi', 'Yingqiang Gao', 'Cyrill A. H. Chevalley', 'Claude Humbel', 'Christophe Gösken', 'Lorenzo Tanzi', 'Thomas Lüthi', 'Stefan Palombo', 'Spencer Poff', 'Boling Yang', 'Nan Wu', 'Matthew Guillod', 'Robin Mamié', 'Daniel Brunner', 'Julio Pereyra', 'Niko Grupen'], 'affiliations': ['Canton of Solothurn', 'ETH Zurich', 'Max Planck Institute for Research on Collective Goods', 'Swiss Federal Supreme Court', 'University of Basel', 'University of Geneva', 'University of Lausanne', 'University of Zurich'], 'pdf_title_img': 'assets/pdf/title_img/2503.01372.jpg', 'data': {'categories': ['#multilingual', '#open_source', '#benchmark', '#dataset', '#machine_translation'], 'emoji': '⚖️', 'ru': {'title': 'Революция в юридическом переводе: ИИ покоряет многоязычную Швейцарию', 'desc': 'Статья представляет SwiLTra-Bench - многоязычный набор данных для оценки систем машинного перевода юридических текстов в Швейцарии. Авторы провели систематическую оценку различных моделей, включая крупные языковые модели и специализированные системы перевода. Результаты показывают, что передовые модели достигают лучших результатов во всех типах документов, а дообучение открытых моделей значительно улучшает качество перевода. Также представлена система SwiLTra-Judge для оценки качества перевода, которая хорошо коррелирует с оценками экспертов.'}, 'en': {'title': 'Enhancing Legal Translation with SwiLTra-Bench and LLMs', 'desc': 'This paper addresses the challenges of legal translation in Switzerland, where multiple languages complicate the process. It introduces SwiLTra-Bench, a benchmark dataset with over 180,000 aligned legal translation pairs to evaluate large language model (LLM) translation systems. The findings show that while advanced models perform well across various document types, specialized systems are better for translating laws but struggle with headnotes. The study also highlights the effectiveness of fine-tuning open-source language models, although they still do not match the performance of top zero-shot models like Claude-3.5-Sonnet.'}, 'zh': {'title': '瑞士法律翻译的智能解决方案', 'desc': '在瑞士，由于有四种官方语言，法律翻译显得尤为重要。传统上，这一过程依赖于既是法律专家又是翻译高手的专业人士，导致了瓶颈，影响了公正的有效获取。为了解决这个问题，我们推出了SwiLTra-Bench，这是一个包含超过18万对瑞士法律翻译的多语言基准数据集，旨在评估基于大语言模型的翻译系统。我们的评估显示，前沿模型在所有文档类型的翻译表现上优于其他系统，而专门的翻译系统在法律文本中表现出色，但在头注方面表现不佳。'}}}];
const articlesContainer = document.getElementById('articles-container');
const sortDropdown = document.getElementById('sort-dropdown');
const categoryFiltersContainer = document.getElementById('category-filters');
const categoryFiltersLogicOptions = document.getElementById('category-options');
const categoryToggle = document.getElementById('category-toggle');
const clearCategoriesButton = document.getElementById('clear-categories');
let selectedCategories = [];
let selectedArticles = [];
let sortBy = 'issue_id';
let showLimitHint = false;
let filterLogicIsAnd = false;
function getUrlParameters() {
const urlParams = new URLSearchParams(window.location.search);
const categoriesParam = urlParams.get('cat');
let categories = categoriesParam ? categoriesParam.split(',') : [];
categories = categories.map(element => `#${element}`);
return categories
}
function updateUrlWithCategories() {
let cleanedCategories = selectedCategories.map(element => element.replace(/^#/, ''));
const newUrl = cleanedCategories.length > 0
? `${window.location.pathname}?cat=${cleanedCategories.join(',')}`
: window.location.pathname;
console.log("cleanedCategories", cleanedCategories)
window.history.pushState({}, '', newUrl);
}
function loadSettings() {
const themeToggle = document.getElementById('theme-toggle');
const sortDropdown = document.getElementById('sort-dropdown');
const isDarkMode = localStorage.getItem('darkMode') === 'true';
let settingSortBy = localStorage.getItem('sort_by');
filterLogicIsAnd = localStorage.getItem('filter_logic_is_and') === 'true';
if (isDarkMode) {
document.body.classList.remove('light-theme');
document.body.classList.add('dark-theme');
themeToggle.checked = true;
const title = document.getElementById('doomgrad');
title.innerHTML = "hf nightly";
const titleSign = document.getElementById('doomgrad-icon');
titleSign.classList.add('rotate');
}
if ((!settingSortBy) || (settingSortBy === 'null')) {
settingSortBy = 'issue_id';
}
if (filterLogicIsAnd) {
document.getElementById('filter-logic-and').checked = true;
} else {
document.getElementById('filter-logic-or').checked = true;
}
sortDropdown.value = settingSortBy;
sortBy = settingSortBy;
}
document.getElementById('theme-toggle').addEventListener('change', toggleTheme);
document.getElementById('filter-logic-and').addEventListener('change', () => {
filterLogicIsAnd = true;
localStorage.setItem('filter_logic_is_and', 'true');
filterAndRenderArticles();
updateSelectedArticlesTitle();
});
document.getElementById('filter-logic-or').addEventListener('change', () => {
filterLogicIsAnd = false;
localStorage.setItem('filter_logic_is_and', 'false');
filterAndRenderArticles();
updateSelectedArticlesTitle();
});
function getUniqueCategories(articles) {
const categories = new Set();
articles.forEach(article => {
if (article.data && article.data.categories) {
article.data.categories.forEach(cat => categories.add(cat));
}
});
let res = Array.from(categories);
res.sort();
return res;
}
function createCategoryButtons() {
//const categories = getUniqueCategories(articlesData);
const categories = ['#3d (1)', '#agents (1)', '#agi', '#alignment (1)', '#architecture (1)', '#audio', '#benchmark (6)', '#cv (2)', '#data (3)', '#dataset (7)', '#diffusion', '#ethics', '#games', '#graphs', '#hallucinations', '#healthcare (2)', '#inference (1)', '#interpretability', '#leakage', '#long_context', '#low_resource (1)', '#machine_translation (1)', '#math', '#multilingual (3)', '#multimodal (3)', '#open_source (5)', '#optimization (4)', '#plp (1)', '#rag', '#reasoning (2)', '#rl (2)', '#rlhf', '#robotics (2)', '#science', '#security (1)', '#small_models (1)', '#story_generation', '#survey', '#synthetic (2)', '#training (5)', '#transfer_learning (1)', '#video (1)'];
categories.forEach(category => {
let catNameSplitted = category.split(/(\s+)/);
let catName = catNameSplitted[0];
const button = document.createElement('span');
button.textContent = catName;
button.className = 'category-button';
if (catNameSplitted.length < 2) {
button.classList.add('inactive');
};
button.onclick = () => toggleCategory(catName, button);
categoryFiltersContainer.appendChild(button);
});
}
function toggleCategory(category, button) {
const index = selectedCategories.indexOf(category);
if (index === -1) {
selectedCategories.push(category);
button.classList.add('active');
} else {
selectedCategories.splice(index, 1);
button.classList.remove('active');
}
filterAndRenderArticles();
saveCategorySelection();