Spaces:
Running
Running
new format
Browse files
app.py
CHANGED
@@ -55,12 +55,10 @@ def get_data_cross_xquad_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
55 |
df_list = []
|
56 |
|
57 |
for model in MODEL_LIST:
|
58 |
-
|
59 |
-
|
60 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_xquad'][res] for res in ALL_RESULTS[model][eval_mode]['cross_xquad']]
|
61 |
-
|
62 |
|
63 |
try:
|
|
|
|
|
64 |
overall_acc = [results['overall_acc'] for results in results_list]
|
65 |
overall_acc = median(overall_acc)
|
66 |
|
@@ -70,20 +68,18 @@ def get_data_cross_xquad_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
70 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
71 |
AC3_3 = median(AC3_3)
|
72 |
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
|
|
|
|
|
|
77 |
|
78 |
-
|
79 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
80 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
81 |
-
"Accuracy": overall_acc,
|
82 |
-
"Cross-Lingual Consistency": consistency_score_3,
|
83 |
-
"AC3": AC3_3,
|
84 |
-
}
|
85 |
|
86 |
-
|
|
|
87 |
|
88 |
|
89 |
df = pd.DataFrame(df_list)
|
@@ -104,7 +100,6 @@ def get_data_cross_xquad_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
104 |
|
105 |
return df
|
106 |
|
107 |
-
|
108 |
CROSS_XQUAD_ZERO_SHOT_OVERALL = get_data_cross_xquad_overall(eval_mode="zero_shot")
|
109 |
CROSS_XQUAD_FIVE_SHOT_OVERALL = get_data_cross_xquad_overall(eval_mode="five_shot")
|
110 |
|
@@ -114,12 +109,10 @@ def get_data_cross_xquad_language(eval_mode='zero_shot', fillna=True, rank=True)
|
|
114 |
df_list = []
|
115 |
|
116 |
for model in MODEL_LIST:
|
117 |
-
|
118 |
-
|
119 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_xquad'][res] for res in ALL_RESULTS[model][eval_mode]['cross_xquad']]
|
120 |
-
|
121 |
-
|
122 |
try:
|
|
|
|
|
123 |
English = [results['language_acc']['English'] for results in results_list]
|
124 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
125 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
@@ -130,23 +123,19 @@ def get_data_cross_xquad_language(eval_mode='zero_shot', fillna=True, rank=True)
|
|
130 |
Chinese = median(Chinese)
|
131 |
Spanish = median(Spanish)
|
132 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
133 |
|
134 |
-
|
135 |
-
English = -1
|
136 |
-
Vietnamese = -1
|
137 |
-
Chinese = -1
|
138 |
-
Spanish = -1
|
139 |
-
|
140 |
-
res = {
|
141 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
142 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
143 |
-
"English": English,
|
144 |
-
"Vietnamese": Vietnamese,
|
145 |
-
"Chinese": Chinese,
|
146 |
-
"Spanish": Spanish,
|
147 |
-
}
|
148 |
|
149 |
-
|
|
|
150 |
|
151 |
|
152 |
df = pd.DataFrame(df_list)
|
@@ -167,7 +156,6 @@ def get_data_cross_xquad_language(eval_mode='zero_shot', fillna=True, rank=True)
|
|
167 |
|
168 |
return df
|
169 |
|
170 |
-
|
171 |
CROSS_XQUAD_ZERO_SHOT_LANGUAGE = get_data_cross_xquad_language(eval_mode="zero_shot")
|
172 |
CROSS_XQUAD_FIVE_SHOT_LANGUAGE = get_data_cross_xquad_language(eval_mode="five_shot")
|
173 |
|
@@ -186,12 +174,11 @@ def get_data_cross_mmlu_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
186 |
df_list = []
|
187 |
|
188 |
for model in MODEL_LIST:
|
189 |
-
|
190 |
-
|
191 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cross_mmlu']]
|
192 |
-
|
193 |
|
194 |
try:
|
|
|
|
|
|
|
195 |
overall_acc = [results['overall_acc'] for results in results_list]
|
196 |
overall_acc = median(overall_acc)
|
197 |
|
@@ -201,20 +188,17 @@ def get_data_cross_mmlu_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
201 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
202 |
AC3_3 = median(AC3_3)
|
203 |
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
"Accuracy": overall_acc,
|
213 |
-
"Cross-Lingual Consistency": consistency_score_3,
|
214 |
-
"AC3": AC3_3,
|
215 |
-
}
|
216 |
|
217 |
-
|
|
|
218 |
|
219 |
|
220 |
df = pd.DataFrame(df_list)
|
@@ -235,7 +219,6 @@ def get_data_cross_mmlu_overall(eval_mode='zero_shot', fillna=True, rank=True):
|
|
235 |
|
236 |
return df
|
237 |
|
238 |
-
|
239 |
CROSS_MMLU_ZERO_SHOT_OVERALL = get_data_cross_mmlu_overall(eval_mode="zero_shot")
|
240 |
CROSS_MMLU_FIVE_SHOT_OVERALL = get_data_cross_mmlu_overall(eval_mode="five_shot")
|
241 |
|
@@ -245,12 +228,11 @@ def get_data_cross_mmlu_language(eval_mode='zero_shot', fillna=True, rank=True):
|
|
245 |
df_list = []
|
246 |
|
247 |
for model in MODEL_LIST:
|
|
|
|
|
248 |
|
|
|
249 |
|
250 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cross_mmlu']]
|
251 |
-
|
252 |
-
|
253 |
-
try:
|
254 |
English = [results['language_acc']['English'] for results in results_list]
|
255 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
256 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
@@ -267,30 +249,22 @@ def get_data_cross_mmlu_language(eval_mode='zero_shot', fillna=True, rank=True):
|
|
267 |
Spanish = median(Spanish)
|
268 |
Malay = median(Malay)
|
269 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
270 |
|
271 |
-
|
272 |
-
English = -1
|
273 |
-
Vietnamese = -1
|
274 |
-
Chinese = -1
|
275 |
-
Indonesian = -1
|
276 |
-
Filipino = -1
|
277 |
-
Spanish = -1
|
278 |
-
Malay = -1
|
279 |
-
|
280 |
-
res = {
|
281 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
282 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
283 |
-
"English": English,
|
284 |
-
"Vietnamese": Vietnamese,
|
285 |
-
"Chinese": Chinese,
|
286 |
-
"Indonesian": Indonesian,
|
287 |
-
"Filipino": Filipino,
|
288 |
-
"Spanish": Spanish,
|
289 |
-
"Malay": Malay,
|
290 |
-
}
|
291 |
-
|
292 |
-
df_list.append(res)
|
293 |
|
|
|
|
|
294 |
|
295 |
df = pd.DataFrame(df_list)
|
296 |
# If there are any models that are the same, merge them
|
@@ -310,7 +284,6 @@ def get_data_cross_mmlu_language(eval_mode='zero_shot', fillna=True, rank=True):
|
|
310 |
|
311 |
return df
|
312 |
|
313 |
-
|
314 |
CROSS_MMLU_ZERO_SHOT_LANGUAGE = get_data_cross_mmlu_language(eval_mode="zero_shot")
|
315 |
CROSS_MMLU_FIVE_SHOT_LANGUAGE = get_data_cross_mmlu_language(eval_mode="five_shot")
|
316 |
|
@@ -325,12 +298,11 @@ def get_data_cross_logiqa_overall(eval_mode='zero_shot', fillna=True, rank=True)
|
|
325 |
df_list = []
|
326 |
|
327 |
for model in MODEL_LIST:
|
|
|
|
|
328 |
|
|
|
329 |
|
330 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_logiqa'][res] for res in ALL_RESULTS[model][eval_mode]['cross_logiqa']]
|
331 |
-
|
332 |
-
|
333 |
-
try:
|
334 |
overall_acc = [results['overall_acc'] for results in results_list]
|
335 |
overall_acc = median(overall_acc)
|
336 |
|
@@ -340,20 +312,18 @@ def get_data_cross_logiqa_overall(eval_mode='zero_shot', fillna=True, rank=True)
|
|
340 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
341 |
AC3_3 = median(AC3_3)
|
342 |
|
343 |
-
|
344 |
-
|
345 |
-
|
346 |
-
|
|
|
|
|
|
|
347 |
|
348 |
-
|
349 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
350 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
351 |
-
"Accuracy": overall_acc,
|
352 |
-
"Cross-Lingual Consistency": consistency_score_3,
|
353 |
-
"AC3": AC3_3,
|
354 |
-
}
|
355 |
|
356 |
-
|
|
|
357 |
|
358 |
|
359 |
df = pd.DataFrame(df_list)
|
@@ -384,12 +354,11 @@ def get_data_cross_logiqa_language(eval_mode='zero_shot', fillna=True, rank=True
|
|
384 |
df_list = []
|
385 |
|
386 |
for model in MODEL_LIST:
|
|
|
|
|
387 |
|
|
|
388 |
|
389 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cross_logiqa'][res] for res in ALL_RESULTS[model][eval_mode]['cross_logiqa']]
|
390 |
-
|
391 |
-
|
392 |
-
try:
|
393 |
English = [results['language_acc']['English'] for results in results_list]
|
394 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
395 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
@@ -406,30 +375,24 @@ def get_data_cross_logiqa_language(eval_mode='zero_shot', fillna=True, rank=True
|
|
406 |
Spanish = median(Spanish)
|
407 |
Malay = median(Malay)
|
408 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
409 |
|
410 |
-
|
411 |
-
English = -1
|
412 |
-
Vietnamese = -1
|
413 |
-
Chinese = -1
|
414 |
-
Indonesian = -1
|
415 |
-
Filipino = -1
|
416 |
-
Spanish = -1
|
417 |
-
Malay = -1
|
418 |
-
|
419 |
-
res = {
|
420 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
421 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
422 |
-
"English": English,
|
423 |
-
"Vietnamese": Vietnamese,
|
424 |
-
"Chinese": Chinese,
|
425 |
-
"Indonesian": Indonesian,
|
426 |
-
"Filipino": Filipino,
|
427 |
-
"Spanish": Spanish,
|
428 |
-
"Malay": Malay,
|
429 |
-
}
|
430 |
|
431 |
-
|
|
|
432 |
|
|
|
433 |
|
434 |
df = pd.DataFrame(df_list)
|
435 |
# If there are any models that are the same, merge them
|
@@ -462,24 +425,23 @@ def get_data_sg_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
462 |
df_list = []
|
463 |
|
464 |
for model in MODEL_LIST:
|
465 |
-
|
466 |
-
|
467 |
-
results_list = [ALL_RESULTS[model][eval_mode]['sg_eval'][res] for res in ALL_RESULTS[model][eval_mode]['sg_eval']]
|
468 |
-
|
469 |
|
470 |
try:
|
|
|
|
|
471 |
accuracy = median([results['accuracy'] for results in results_list])
|
472 |
|
473 |
-
|
474 |
-
|
|
|
|
|
|
|
475 |
|
476 |
-
|
477 |
-
|
478 |
-
|
479 |
-
|
480 |
-
}
|
481 |
|
482 |
-
df_list.append(res)
|
483 |
|
484 |
|
485 |
df = pd.DataFrame(df_list)
|
@@ -515,24 +477,20 @@ def get_data_us_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
515 |
|
516 |
for model in MODEL_LIST:
|
517 |
|
518 |
-
|
519 |
-
results_list = [ALL_RESULTS[model][eval_mode]['us_eval'][res] for res in ALL_RESULTS[model][eval_mode]['us_eval']]
|
520 |
-
|
521 |
-
|
522 |
try:
|
|
|
523 |
accuracy = median([results['accuracy'] for results in results_list])
|
524 |
|
525 |
-
|
526 |
-
|
527 |
-
|
|
|
|
|
528 |
|
529 |
-
|
530 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
531 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
532 |
-
"Accuracy": accuracy,
|
533 |
-
}
|
534 |
|
535 |
-
|
|
|
536 |
|
537 |
|
538 |
df = pd.DataFrame(df_list)
|
@@ -567,26 +525,21 @@ def get_data_cn_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
567 |
df_list = []
|
568 |
|
569 |
for model in MODEL_LIST:
|
570 |
-
|
571 |
-
|
572 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cn_eval'][res] for res in ALL_RESULTS[model][eval_mode]['cn_eval']]
|
573 |
-
|
574 |
|
575 |
try:
|
|
|
576 |
accuracy = median([results['accuracy'] for results in results_list])
|
577 |
|
578 |
-
|
579 |
-
|
580 |
-
|
|
|
|
|
581 |
|
582 |
-
|
583 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
584 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
585 |
-
"Accuracy": accuracy,
|
586 |
-
}
|
587 |
-
|
588 |
-
df_list.append(res)
|
589 |
|
|
|
|
|
590 |
|
591 |
df = pd.DataFrame(df_list)
|
592 |
# If there are any models that are the same, merge them
|
@@ -606,7 +559,6 @@ def get_data_cn_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
606 |
|
607 |
return df
|
608 |
|
609 |
-
|
610 |
CN_EVAL_ZERO_SHOT = get_data_cn_eval(eval_mode="zero_shot")
|
611 |
CN_EVAL_FIVE_SHOT = get_data_cn_eval(eval_mode="five_shot")
|
612 |
|
@@ -614,7 +566,6 @@ CN_EVAL_FIVE_SHOT = get_data_cn_eval(eval_mode="five_shot")
|
|
614 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
615 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
616 |
|
617 |
-
|
618 |
def get_data_ph_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
619 |
|
620 |
df_list = []
|
@@ -622,23 +573,21 @@ def get_data_ph_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
622 |
for model in MODEL_LIST:
|
623 |
|
624 |
|
625 |
-
results_list = [ALL_RESULTS[model][eval_mode]['ph_eval'][res] for res in ALL_RESULTS[model][eval_mode]['ph_eval']]
|
626 |
|
627 |
|
628 |
try:
|
|
|
629 |
accuracy = median([results['accuracy'] for results in results_list])
|
|
|
|
|
|
|
|
|
|
|
630 |
|
631 |
-
|
632 |
-
accuracy = -1
|
633 |
-
|
634 |
|
635 |
-
|
636 |
-
|
637 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
638 |
-
"Accuracy": accuracy,
|
639 |
-
}
|
640 |
-
|
641 |
-
df_list.append(res)
|
642 |
|
643 |
|
644 |
df = pd.DataFrame(df_list)
|
@@ -673,25 +622,21 @@ def get_data_sing2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
673 |
df_list = []
|
674 |
|
675 |
for model in MODEL_LIST:
|
676 |
-
|
677 |
-
|
678 |
-
results_list = [ALL_RESULTS[model][eval_mode]['sing2eng'][res] for res in ALL_RESULTS[model][eval_mode]['sing2eng']]
|
679 |
-
|
680 |
|
681 |
try:
|
|
|
682 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
683 |
|
684 |
-
|
685 |
-
|
686 |
-
|
|
|
|
|
687 |
|
688 |
-
|
689 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
690 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
691 |
-
"BLEU": bleu_score,
|
692 |
-
}
|
693 |
|
694 |
-
|
|
|
695 |
|
696 |
|
697 |
df = pd.DataFrame(df_list)
|
@@ -725,25 +670,21 @@ def get_data_flores_ind2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
725 |
df_list = []
|
726 |
|
727 |
for model in MODEL_LIST:
|
728 |
-
|
729 |
-
|
730 |
-
results_list = [ALL_RESULTS[model][eval_mode]['flores_ind2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_ind2eng']]
|
731 |
-
|
732 |
|
733 |
try:
|
|
|
734 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
735 |
|
736 |
-
|
737 |
-
|
738 |
-
|
|
|
|
|
739 |
|
740 |
-
|
741 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
742 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
743 |
-
"BLEU": bleu_score,
|
744 |
-
}
|
745 |
|
746 |
-
|
|
|
747 |
|
748 |
|
749 |
df = pd.DataFrame(df_list)
|
@@ -779,26 +720,21 @@ def get_data_flores_vie2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
779 |
df_list = []
|
780 |
|
781 |
for model in MODEL_LIST:
|
782 |
-
|
783 |
-
|
784 |
-
results_list = [ALL_RESULTS[model][eval_mode]['flores_vie2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_vie2eng']]
|
785 |
-
|
786 |
|
787 |
try:
|
|
|
788 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
789 |
|
790 |
-
|
791 |
-
|
792 |
-
|
793 |
-
|
794 |
-
|
795 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
796 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
797 |
-
"BLEU": bleu_score,
|
798 |
-
}
|
799 |
|
800 |
-
|
801 |
|
|
|
|
|
802 |
|
803 |
df = pd.DataFrame(df_list)
|
804 |
# If there are any models that are the same, merge them
|
@@ -831,26 +767,21 @@ def get_data_flores_zho2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
831 |
df_list = []
|
832 |
|
833 |
for model in MODEL_LIST:
|
834 |
-
|
835 |
-
|
836 |
-
results_list = [ALL_RESULTS[model][eval_mode]['flores_zho2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_zho2eng']]
|
837 |
-
|
838 |
|
839 |
try:
|
|
|
840 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
841 |
|
842 |
-
|
843 |
-
|
844 |
-
|
845 |
-
|
846 |
-
|
847 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
848 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
849 |
-
"BLEU": bleu_score,
|
850 |
-
}
|
851 |
|
852 |
-
|
853 |
|
|
|
|
|
854 |
|
855 |
df = pd.DataFrame(df_list)
|
856 |
# If there are any models that are the same, merge them
|
@@ -870,7 +801,6 @@ def get_data_flores_zho2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
870 |
|
871 |
return df
|
872 |
|
873 |
-
|
874 |
FLORES_ZHO2ENG_ZERO_SHOT = get_data_flores_zho2eng(eval_mode="zero_shot")
|
875 |
FLORES_ZHO2ENG_FIVE_SHOT = get_data_flores_zho2eng(eval_mode="five_shot")
|
876 |
|
@@ -884,26 +814,20 @@ def get_data_flores_zsm2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
884 |
df_list = []
|
885 |
|
886 |
for model in MODEL_LIST:
|
887 |
-
|
888 |
-
|
889 |
-
results_list = [ALL_RESULTS[model][eval_mode]['flores_zsm2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_zsm2eng']]
|
890 |
-
|
891 |
-
|
892 |
try:
|
|
|
893 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
894 |
|
895 |
-
|
896 |
-
|
897 |
-
|
898 |
-
|
899 |
-
|
900 |
-
|
901 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
902 |
-
"BLEU": bleu_score,
|
903 |
-
}
|
904 |
-
|
905 |
-
df_list.append(res)
|
906 |
|
|
|
|
|
907 |
|
908 |
df = pd.DataFrame(df_list)
|
909 |
# If there are any models that are the same, merge them
|
@@ -923,7 +847,6 @@ def get_data_flores_zsm2eng(eval_mode='zero_shot', fillna=True, rank=True):
|
|
923 |
|
924 |
return df
|
925 |
|
926 |
-
|
927 |
FLORES_ZSM2ENG_ZERO_SHOT = get_data_flores_zho2eng(eval_mode="zero_shot")
|
928 |
FLORES_ZSM2ENG_FIVE_SHOT = get_data_flores_zho2eng(eval_mode="five_shot")
|
929 |
|
@@ -937,27 +860,21 @@ def get_data_mmlu(eval_mode='zero_shot', fillna=True, rank=True):
|
|
937 |
df_list = []
|
938 |
|
939 |
for model in MODEL_LIST:
|
940 |
-
|
941 |
-
|
942 |
-
results_list = [ALL_RESULTS[model][eval_mode]['mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['mmlu']]
|
943 |
-
|
944 |
-
|
945 |
try:
|
|
|
946 |
accuracy = median([results['accuracy'] for results in results_list])
|
947 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
948 |
except:
|
949 |
accuracy = -1
|
950 |
|
951 |
-
|
952 |
-
res = {
|
953 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
954 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
955 |
-
"Accuracy": accuracy,
|
956 |
-
}
|
957 |
-
|
958 |
-
df_list.append(res)
|
959 |
-
|
960 |
-
|
961 |
df = pd.DataFrame(df_list)
|
962 |
# If there are any models that are the same, merge them
|
963 |
# E.g. if out["Model"] has the same value in two places, merge & take whichever one is not NaN else just take the first one
|
@@ -984,32 +901,26 @@ MMLU_FIVE_SHOT = get_data_mmlu(eval_mode="five_shot")
|
|
984 |
|
985 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
986 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
987 |
-
|
988 |
-
|
989 |
def get_data_mmlu_full(eval_mode='zero_shot', fillna=True, rank=True):
|
990 |
|
991 |
df_list = []
|
992 |
|
993 |
for model in MODEL_LIST:
|
994 |
-
|
995 |
-
|
996 |
-
results_list = [ALL_RESULTS[model][eval_mode]['mmlu_full'][res] for res in ALL_RESULTS[model][eval_mode]['mmlu_full']]
|
997 |
-
|
998 |
-
|
999 |
try:
|
|
|
1000 |
accuracy = median([results['accuracy'] for results in results_list])
|
1001 |
|
1002 |
-
|
1003 |
-
|
1004 |
-
|
|
|
|
|
1005 |
|
1006 |
-
|
1007 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1008 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1009 |
-
"Accuracy": accuracy,
|
1010 |
-
}
|
1011 |
|
1012 |
-
|
|
|
1013 |
|
1014 |
|
1015 |
df = pd.DataFrame(df_list)
|
@@ -1030,40 +941,31 @@ def get_data_mmlu_full(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1030 |
|
1031 |
return df
|
1032 |
|
1033 |
-
|
1034 |
MMLU_FULL_ZERO_SHOT = get_data_mmlu_full(eval_mode="zero_shot")
|
1035 |
MMLU_FULL_FIVE_SHOT = get_data_mmlu_full(eval_mode="five_shot")
|
1036 |
|
1037 |
|
1038 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1039 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1040 |
-
|
1041 |
-
|
1042 |
def get_data_c_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
1043 |
|
1044 |
df_list = []
|
1045 |
|
1046 |
-
for model in MODEL_LIST:
|
1047 |
-
|
1048 |
-
|
1049 |
-
results_list = [ALL_RESULTS[model][eval_mode]['c_eval'][res] for res in ALL_RESULTS[model][eval_mode]['c_eval']]
|
1050 |
-
|
1051 |
-
|
1052 |
try:
|
|
|
1053 |
accuracy = median([results['accuracy'] for results in results_list])
|
1054 |
|
1055 |
-
|
1056 |
-
|
1057 |
-
|
1058 |
-
|
1059 |
-
|
1060 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1061 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1062 |
-
"Accuracy": accuracy,
|
1063 |
-
}
|
1064 |
|
1065 |
-
|
1066 |
|
|
|
|
|
1067 |
|
1068 |
df = pd.DataFrame(df_list)
|
1069 |
# If there are any models that are the same, merge them
|
@@ -1083,7 +985,6 @@ def get_data_c_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1083 |
|
1084 |
return df
|
1085 |
|
1086 |
-
|
1087 |
C_EVAL_ZERO_SHOT = get_data_c_eval(eval_mode="zero_shot")
|
1088 |
C_EVAL_FIVE_SHOT = get_data_c_eval(eval_mode="five_shot")
|
1089 |
|
@@ -1097,25 +998,23 @@ def get_data_c_eval_full(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1097 |
df_list = []
|
1098 |
|
1099 |
for model in MODEL_LIST:
|
1100 |
-
|
1101 |
-
|
1102 |
-
results_list = [ALL_RESULTS[model][eval_mode]['c_eval_full'][res] for res in ALL_RESULTS[model][eval_mode]['c_eval_full']]
|
1103 |
-
|
1104 |
-
|
1105 |
try:
|
|
|
1106 |
accuracy = median([results['accuracy'] for results in results_list])
|
1107 |
|
1108 |
-
|
1109 |
-
|
|
|
|
|
|
|
1110 |
|
|
|
|
|
|
|
|
|
1111 |
|
1112 |
-
res = {
|
1113 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1114 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1115 |
-
"Accuracy": accuracy,
|
1116 |
-
}
|
1117 |
|
1118 |
-
df_list.append(res)
|
1119 |
|
1120 |
|
1121 |
df = pd.DataFrame(df_list)
|
@@ -1152,25 +1051,24 @@ def get_data_cmmlu(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1152 |
df_list = []
|
1153 |
|
1154 |
for model in MODEL_LIST:
|
1155 |
-
|
1156 |
-
|
1157 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cmmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cmmlu']]
|
1158 |
-
|
1159 |
-
|
1160 |
try:
|
|
|
1161 |
accuracy = median([results['accuracy'] for results in results_list])
|
1162 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1163 |
except:
|
1164 |
-
|
1165 |
|
1166 |
|
1167 |
-
res = {
|
1168 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1169 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1170 |
-
"Accuracy": accuracy,
|
1171 |
-
}
|
1172 |
|
1173 |
-
df_list.append(res)
|
1174 |
|
1175 |
|
1176 |
df = pd.DataFrame(df_list)
|
@@ -1197,9 +1095,6 @@ CMMLU_FIVE_SHOT = get_data_cmmlu(eval_mode="five_shot")
|
|
1197 |
|
1198 |
|
1199 |
|
1200 |
-
|
1201 |
-
|
1202 |
-
|
1203 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1204 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1205 |
|
@@ -1209,25 +1104,24 @@ def get_data_cmmlu_full(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1209 |
df_list = []
|
1210 |
|
1211 |
for model in MODEL_LIST:
|
1212 |
-
|
1213 |
-
|
1214 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cmmlu_full'][res] for res in ALL_RESULTS[model][eval_mode]['cmmlu_full']]
|
1215 |
-
|
1216 |
|
1217 |
try:
|
|
|
1218 |
accuracy = median([results['accuracy'] for results in results_list])
|
1219 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1220 |
except:
|
1221 |
-
|
1222 |
|
1223 |
|
1224 |
-
res = {
|
1225 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1226 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1227 |
-
"Accuracy": accuracy,
|
1228 |
-
}
|
1229 |
|
1230 |
-
df_list.append(res)
|
1231 |
|
1232 |
|
1233 |
df = pd.DataFrame(df_list)
|
@@ -1263,25 +1157,20 @@ def get_data_zbench(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1263 |
df_list = []
|
1264 |
|
1265 |
for model in MODEL_LIST:
|
1266 |
-
|
1267 |
-
|
1268 |
-
results_list = [ALL_RESULTS[model][eval_mode]['zbench'][res] for res in ALL_RESULTS[model][eval_mode]['zbench']]
|
1269 |
-
|
1270 |
-
|
1271 |
try:
|
|
|
1272 |
accuracy = median([results['accuracy'] for results in results_list])
|
1273 |
|
1274 |
-
|
1275 |
-
|
1276 |
-
|
|
|
|
|
1277 |
|
1278 |
-
|
1279 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1280 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1281 |
-
"Accuracy": accuracy,
|
1282 |
-
}
|
1283 |
|
1284 |
-
|
|
|
1285 |
|
1286 |
|
1287 |
df = pd.DataFrame(df_list)
|
@@ -1316,21 +1205,23 @@ def get_data_indommlu(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1316 |
|
1317 |
for model in MODEL_LIST:
|
1318 |
|
1319 |
-
results_list = [ALL_RESULTS[model][eval_mode]['indommlu'][res] for res in ALL_RESULTS[model][eval_mode]['indommlu']]
|
1320 |
|
1321 |
try:
|
|
|
1322 |
accuracy = median([results['accuracy'] for results in results_list])
|
1323 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1324 |
except:
|
1325 |
-
|
1326 |
|
1327 |
-
res = {
|
1328 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1329 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1330 |
-
"Accuracy": accuracy,
|
1331 |
-
}
|
1332 |
|
1333 |
-
df_list.append(res)
|
1334 |
|
1335 |
|
1336 |
df = pd.DataFrame(df_list)
|
@@ -1358,33 +1249,25 @@ INDOMMLU_FIVE_SHOT = get_data_indommlu(eval_mode="five_shot")
|
|
1358 |
|
1359 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1360 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1361 |
-
|
1362 |
-
|
1363 |
def get_data_ind_emotion(eval_mode='zero_shot', fillna=True, rank=True):
|
1364 |
|
1365 |
df_list = []
|
1366 |
|
1367 |
for model in MODEL_LIST:
|
1368 |
-
|
1369 |
-
|
1370 |
-
results_list = [ALL_RESULTS[model][eval_mode]['ind_emotion'][res] for res in ALL_RESULTS[model][eval_mode]['ind_emotion']]
|
1371 |
-
|
1372 |
-
|
1373 |
try:
|
|
|
1374 |
accuracy = median([results['accuracy'] for results in results_list])
|
1375 |
|
1376 |
-
|
1377 |
-
|
1378 |
-
|
1379 |
-
|
1380 |
-
|
1381 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1382 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1383 |
-
"Accuracy": accuracy,
|
1384 |
-
}
|
1385 |
|
1386 |
-
|
1387 |
|
|
|
|
|
1388 |
|
1389 |
df = pd.DataFrame(df_list)
|
1390 |
# If there are any models that are the same, merge them
|
@@ -1404,7 +1287,6 @@ def get_data_ind_emotion(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1404 |
|
1405 |
return df
|
1406 |
|
1407 |
-
|
1408 |
IND_EMOTION_ZERO_SHOT = get_data_ind_emotion(eval_mode="zero_shot")
|
1409 |
IND_EMOTION_FIVE_SHOT = get_data_ind_emotion(eval_mode="five_shot")
|
1410 |
|
@@ -1420,25 +1302,21 @@ def get_data_ocnli(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1420 |
df_list = []
|
1421 |
|
1422 |
for model in MODEL_LIST:
|
1423 |
-
|
1424 |
-
|
1425 |
-
results_list = [ALL_RESULTS[model][eval_mode]['ocnli'][res] for res in ALL_RESULTS[model][eval_mode]['ocnli']]
|
1426 |
-
|
1427 |
|
1428 |
try:
|
|
|
1429 |
accuracy = median([results['accuracy'] for results in results_list])
|
1430 |
|
1431 |
-
|
1432 |
-
|
1433 |
-
|
|
|
|
|
1434 |
|
1435 |
-
|
1436 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1437 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1438 |
-
"Accuracy": accuracy,
|
1439 |
-
}
|
1440 |
|
1441 |
-
|
|
|
1442 |
|
1443 |
|
1444 |
df = pd.DataFrame(df_list)
|
@@ -1474,26 +1352,21 @@ def get_data_c3(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1474 |
df_list = []
|
1475 |
|
1476 |
for model in MODEL_LIST:
|
1477 |
-
|
1478 |
-
|
1479 |
-
results_list = [ALL_RESULTS[model][eval_mode]['c3'][res] for res in ALL_RESULTS[model][eval_mode]['c3']]
|
1480 |
-
|
1481 |
|
1482 |
try:
|
|
|
1483 |
accuracy = median([results['accuracy'] for results in results_list])
|
1484 |
|
1485 |
-
|
1486 |
-
|
1487 |
-
|
|
|
|
|
1488 |
|
1489 |
-
|
1490 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1491 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1492 |
-
"Accuracy": accuracy,
|
1493 |
-
}
|
1494 |
-
|
1495 |
-
df_list.append(res)
|
1496 |
|
|
|
|
|
1497 |
|
1498 |
df = pd.DataFrame(df_list)
|
1499 |
# If there are any models that are the same, merge them
|
@@ -1528,25 +1401,21 @@ def get_data_dream(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1528 |
df_list = []
|
1529 |
|
1530 |
for model in MODEL_LIST:
|
1531 |
-
|
1532 |
-
|
1533 |
-
results_list = [ALL_RESULTS[model][eval_mode]['dream'][res] for res in ALL_RESULTS[model][eval_mode]['dream']]
|
1534 |
-
|
1535 |
|
1536 |
try:
|
|
|
1537 |
accuracy = median([results['accuracy'] for results in results_list])
|
1538 |
|
1539 |
-
|
1540 |
-
|
1541 |
-
|
|
|
|
|
1542 |
|
1543 |
-
|
1544 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1545 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1546 |
-
"Accuracy": accuracy,
|
1547 |
-
}
|
1548 |
|
1549 |
-
|
|
|
1550 |
|
1551 |
|
1552 |
df = pd.DataFrame(df_list)
|
@@ -1567,47 +1436,36 @@ def get_data_dream(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1567 |
|
1568 |
return df
|
1569 |
|
1570 |
-
|
1571 |
DREAM_ZERO_SHOT = get_data_dream(eval_mode="zero_shot")
|
1572 |
DREAM_FIVE_SHOT = get_data_dream(eval_mode="five_shot")
|
1573 |
|
1574 |
-
|
1575 |
-
|
1576 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1577 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1578 |
-
|
1579 |
-
|
1580 |
def get_data_samsum(eval_mode='zero_shot', fillna=True, rank=True):
|
1581 |
|
1582 |
df_list = []
|
1583 |
|
1584 |
for model in MODEL_LIST:
|
1585 |
-
|
1586 |
-
|
1587 |
-
results_list = [ALL_RESULTS[model][eval_mode]['samsum'][res] for res in ALL_RESULTS[model][eval_mode]['samsum']]
|
1588 |
-
|
1589 |
|
1590 |
try:
|
|
|
|
|
1591 |
rouge1 = median([results['rouge1'] for results in results_list])
|
1592 |
rouge2 = median([results['rouge2'] for results in results_list])
|
1593 |
rougeL = median([results['rougeL'] for results in results_list])
|
1594 |
|
1595 |
-
|
1596 |
-
|
1597 |
-
|
1598 |
-
|
1599 |
-
|
|
|
|
|
1600 |
|
1601 |
-
|
1602 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1603 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1604 |
-
"ROUGE-1": rouge1,
|
1605 |
-
"ROUGE-2": rouge2,
|
1606 |
-
"ROUGE-L": rougeL,
|
1607 |
-
}
|
1608 |
-
|
1609 |
-
df_list.append(res)
|
1610 |
|
|
|
|
|
1611 |
|
1612 |
df = pd.DataFrame(df_list)
|
1613 |
# If there are any models that are the same, merge them
|
@@ -1641,31 +1499,29 @@ def get_data_dialogsum(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1641 |
df_list = []
|
1642 |
|
1643 |
for model in MODEL_LIST:
|
1644 |
-
|
1645 |
-
|
1646 |
-
results_list = [ALL_RESULTS[model][eval_mode]['dialogsum'][res] for res in ALL_RESULTS[model][eval_mode]['dialogsum']]
|
1647 |
-
|
1648 |
-
|
1649 |
try:
|
|
|
|
|
1650 |
rouge1 = median([results['rouge1'] for results in results_list])
|
1651 |
rouge2 = median([results['rouge2'] for results in results_list])
|
1652 |
rougeL = median([results['rougeL'] for results in results_list])
|
1653 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1654 |
except:
|
1655 |
-
|
1656 |
-
rouge2 = -1
|
1657 |
-
rougeL = -1
|
1658 |
|
1659 |
|
1660 |
-
res = {
|
1661 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1662 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1663 |
-
"ROUGE-1": rouge1,
|
1664 |
-
"ROUGE-2": rouge2,
|
1665 |
-
"ROUGE-L": rougeL,
|
1666 |
-
}
|
1667 |
|
1668 |
-
df_list.append(res)
|
1669 |
|
1670 |
|
1671 |
df = pd.DataFrame(df_list)
|
@@ -1703,24 +1559,23 @@ def get_data_sst2(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1703 |
|
1704 |
for model in MODEL_LIST:
|
1705 |
|
1706 |
-
|
1707 |
-
results_list = [ALL_RESULTS[model][eval_mode]['sst2'][res] for res in ALL_RESULTS[model][eval_mode]['sst2']]
|
1708 |
-
|
1709 |
-
|
1710 |
try:
|
|
|
1711 |
accuracy = median([results['accuracy'] for results in results_list])
|
1712 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1713 |
except:
|
1714 |
-
|
1715 |
|
1716 |
|
1717 |
-
res = {
|
1718 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1719 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1720 |
-
"Accuracy": accuracy,
|
1721 |
-
}
|
1722 |
|
1723 |
-
df_list.append(res)
|
1724 |
|
1725 |
|
1726 |
df = pd.DataFrame(df_list)
|
@@ -1757,26 +1612,21 @@ def get_data_cola(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1757 |
df_list = []
|
1758 |
|
1759 |
for model in MODEL_LIST:
|
1760 |
-
|
1761 |
-
|
1762 |
-
results_list = [ALL_RESULTS[model][eval_mode]['cola'][res] for res in ALL_RESULTS[model][eval_mode]['cola']]
|
1763 |
-
|
1764 |
|
1765 |
try:
|
|
|
1766 |
accuracy = median([results['accuracy'] for results in results_list])
|
1767 |
|
1768 |
-
|
1769 |
-
|
1770 |
-
|
1771 |
-
|
1772 |
-
|
1773 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1774 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1775 |
-
"Accuracy": accuracy,
|
1776 |
-
}
|
1777 |
|
1778 |
-
|
1779 |
|
|
|
|
|
1780 |
|
1781 |
df = pd.DataFrame(df_list)
|
1782 |
# If there are any models that are the same, merge them
|
@@ -1814,24 +1664,20 @@ def get_data_qqp(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1814 |
|
1815 |
for model in MODEL_LIST:
|
1816 |
|
1817 |
-
|
1818 |
-
results_list = [ALL_RESULTS[model][eval_mode]['qqp'][res] for res in ALL_RESULTS[model][eval_mode]['qqp']]
|
1819 |
-
|
1820 |
-
|
1821 |
try:
|
|
|
1822 |
accuracy = median([results['accuracy'] for results in results_list])
|
1823 |
|
1824 |
-
|
1825 |
-
|
1826 |
-
|
|
|
|
|
1827 |
|
1828 |
-
|
1829 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1830 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1831 |
-
"Accuracy": accuracy,
|
1832 |
-
}
|
1833 |
|
1834 |
-
|
|
|
1835 |
|
1836 |
|
1837 |
df = pd.DataFrame(df_list)
|
@@ -1869,25 +1715,21 @@ def get_data_mnli(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1869 |
df_list = []
|
1870 |
|
1871 |
for model in MODEL_LIST:
|
1872 |
-
|
1873 |
-
|
1874 |
-
results_list = [ALL_RESULTS[model][eval_mode]['mnli'][res] for res in ALL_RESULTS[model][eval_mode]['mnli']]
|
1875 |
-
|
1876 |
-
|
1877 |
try:
|
|
|
1878 |
accuracy = median([results['accuracy'] for results in results_list])
|
1879 |
|
1880 |
-
|
1881 |
-
|
1882 |
-
|
|
|
|
|
1883 |
|
1884 |
-
|
1885 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1886 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1887 |
-
"Accuracy": accuracy,
|
1888 |
-
}
|
1889 |
|
1890 |
-
|
|
|
1891 |
|
1892 |
|
1893 |
df = pd.DataFrame(df_list)
|
@@ -1925,26 +1767,21 @@ def get_data_qnli(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1925 |
df_list = []
|
1926 |
|
1927 |
for model in MODEL_LIST:
|
1928 |
-
|
1929 |
-
|
1930 |
-
results_list = [ALL_RESULTS[model][eval_mode]['qnli'][res] for res in ALL_RESULTS[model][eval_mode]['qnli']]
|
1931 |
-
|
1932 |
|
1933 |
try:
|
|
|
1934 |
accuracy = median([results['accuracy'] for results in results_list])
|
1935 |
|
1936 |
-
|
1937 |
-
|
1938 |
-
|
1939 |
-
|
1940 |
-
|
1941 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1942 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1943 |
-
"Accuracy": accuracy,
|
1944 |
-
}
|
1945 |
|
1946 |
-
|
1947 |
|
|
|
|
|
1948 |
|
1949 |
df = pd.DataFrame(df_list)
|
1950 |
# If there are any models that are the same, merge them
|
@@ -1981,26 +1818,21 @@ def get_data_wnli(eval_mode='zero_shot', fillna=True, rank=True):
|
|
1981 |
df_list = []
|
1982 |
|
1983 |
for model in MODEL_LIST:
|
1984 |
-
|
1985 |
-
|
1986 |
-
results_list = [ALL_RESULTS[model][eval_mode]['wnli'][res] for res in ALL_RESULTS[model][eval_mode]['wnli']]
|
1987 |
-
|
1988 |
|
1989 |
try:
|
|
|
1990 |
accuracy = median([results['accuracy'] for results in results_list])
|
1991 |
|
1992 |
-
|
1993 |
-
|
1994 |
-
|
1995 |
-
|
1996 |
-
|
1997 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1998 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1999 |
-
"Accuracy": accuracy,
|
2000 |
-
}
|
2001 |
|
2002 |
-
|
2003 |
|
|
|
|
|
2004 |
|
2005 |
df = pd.DataFrame(df_list)
|
2006 |
# If there are any models that are the same, merge them
|
@@ -2020,14 +1852,10 @@ def get_data_wnli(eval_mode='zero_shot', fillna=True, rank=True):
|
|
2020 |
|
2021 |
return df
|
2022 |
|
2023 |
-
|
2024 |
WNLI_ZERO_SHOT = get_data_wnli(eval_mode="zero_shot")
|
2025 |
WNLI_FIVE_SHOT = get_data_wnli(eval_mode="five_shot")
|
2026 |
|
2027 |
|
2028 |
-
|
2029 |
-
|
2030 |
-
|
2031 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
2032 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
2033 |
|
@@ -2037,26 +1865,20 @@ def get_data_rte(eval_mode='zero_shot', fillna=True, rank=True):
|
|
2037 |
df_list = []
|
2038 |
|
2039 |
for model in MODEL_LIST:
|
2040 |
-
|
2041 |
-
|
2042 |
-
results_list = [ALL_RESULTS[model][eval_mode]['rte'][res] for res in ALL_RESULTS[model][eval_mode]['rte']]
|
2043 |
-
|
2044 |
-
|
2045 |
try:
|
|
|
2046 |
accuracy = median([results['accuracy'] for results in results_list])
|
2047 |
|
2048 |
-
|
2049 |
-
|
2050 |
-
|
2051 |
-
|
2052 |
-
|
2053 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
2054 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
2055 |
-
"Accuracy": accuracy,
|
2056 |
-
}
|
2057 |
|
2058 |
-
|
2059 |
|
|
|
|
|
2060 |
|
2061 |
df = pd.DataFrame(df_list)
|
2062 |
# If there are any models that are the same, merge them
|
@@ -2081,39 +1903,28 @@ RTE_ZERO_SHOT = get_data_rte(eval_mode="zero_shot")
|
|
2081 |
RTE_FIVE_SHOT = get_data_rte(eval_mode="five_shot")
|
2082 |
|
2083 |
|
2084 |
-
|
2085 |
-
|
2086 |
-
|
2087 |
-
|
2088 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
2089 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
2090 |
-
|
2091 |
-
|
2092 |
def get_data_mrpc(eval_mode='zero_shot', fillna=True, rank=True):
|
2093 |
|
2094 |
df_list = []
|
2095 |
|
2096 |
for model in MODEL_LIST:
|
2097 |
-
|
2098 |
-
|
2099 |
-
results_list = [ALL_RESULTS[model][eval_mode]['mrpc'][res] for res in ALL_RESULTS[model][eval_mode]['mrpc']]
|
2100 |
-
|
2101 |
-
|
2102 |
try:
|
|
|
2103 |
accuracy = median([results['accuracy'] for results in results_list])
|
2104 |
|
2105 |
-
|
2106 |
-
|
2107 |
-
|
|
|
|
|
2108 |
|
2109 |
-
|
2110 |
-
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
2111 |
-
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
2112 |
-
"Accuracy": accuracy,
|
2113 |
-
}
|
2114 |
-
|
2115 |
-
df_list.append(res)
|
2116 |
|
|
|
|
|
2117 |
|
2118 |
df = pd.DataFrame(df_list)
|
2119 |
# If there are any models that are the same, merge them
|
@@ -2210,8 +2021,8 @@ with block:
|
|
2210 |
- **Mode of Evaluation**: Zero-Shot, Five-Shot
|
2211 |
|
2212 |
### The following table shows the performance of the models on the SeaEval benchmark.
|
2213 |
-
- For **Zero-
|
2214 |
-
-
|
2215 |
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
2216 |
|
2217 |
""")
|
@@ -2348,7 +2159,7 @@ with block:
|
|
2348 |
|
2349 |
|
2350 |
|
2351 |
-
with gr.TabItem("Cultural Reasoning
|
2352 |
|
2353 |
# dataset 3: SG_EVAL
|
2354 |
with gr.TabItem("SG_EVAL"):
|
@@ -2697,7 +2508,7 @@ with block:
|
|
2697 |
""")
|
2698 |
|
2699 |
|
2700 |
-
with gr.TabItem("FLORES
|
2701 |
|
2702 |
|
2703 |
# dataset 8:
|
@@ -2805,7 +2616,7 @@ with block:
|
|
2805 |
""")
|
2806 |
|
2807 |
|
2808 |
-
with gr.TabItem("Emotion
|
2809 |
|
2810 |
# dataset 18:
|
2811 |
with gr.TabItem("ind_emotion"):
|
@@ -2941,7 +2752,7 @@ with block:
|
|
2941 |
|
2942 |
|
2943 |
|
2944 |
-
with gr.TabItem("Fundamental NLP"):
|
2945 |
|
2946 |
|
2947 |
# dataset
|
|
|
55 |
df_list = []
|
56 |
|
57 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
58 |
|
59 |
try:
|
60 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_xquad'][res] for res in ALL_RESULTS[model][eval_mode]['cross_xquad']]
|
61 |
+
|
62 |
overall_acc = [results['overall_acc'] for results in results_list]
|
63 |
overall_acc = median(overall_acc)
|
64 |
|
|
|
68 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
69 |
AC3_3 = median(AC3_3)
|
70 |
|
71 |
+
res = {
|
72 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
73 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
74 |
+
"Accuracy": overall_acc,
|
75 |
+
"Cross-Lingual Consistency": consistency_score_3,
|
76 |
+
"AC3": AC3_3,
|
77 |
+
}
|
78 |
|
79 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
|
81 |
+
except:
|
82 |
+
print('Not found in model: {} for {}'.format(model, "cross_xquad_overall"))
|
83 |
|
84 |
|
85 |
df = pd.DataFrame(df_list)
|
|
|
100 |
|
101 |
return df
|
102 |
|
|
|
103 |
CROSS_XQUAD_ZERO_SHOT_OVERALL = get_data_cross_xquad_overall(eval_mode="zero_shot")
|
104 |
CROSS_XQUAD_FIVE_SHOT_OVERALL = get_data_cross_xquad_overall(eval_mode="five_shot")
|
105 |
|
|
|
109 |
df_list = []
|
110 |
|
111 |
for model in MODEL_LIST:
|
112 |
+
|
|
|
|
|
|
|
|
|
113 |
try:
|
114 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_xquad'][res] for res in ALL_RESULTS[model][eval_mode]['cross_xquad']]
|
115 |
+
|
116 |
English = [results['language_acc']['English'] for results in results_list]
|
117 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
118 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
|
|
123 |
Chinese = median(Chinese)
|
124 |
Spanish = median(Spanish)
|
125 |
|
126 |
+
res = {
|
127 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
128 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
129 |
+
"English": English,
|
130 |
+
"Vietnamese": Vietnamese,
|
131 |
+
"Chinese": Chinese,
|
132 |
+
"Spanish": Spanish,
|
133 |
+
}
|
134 |
|
135 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
+
except:
|
138 |
+
print('Not found in model: {} for {}'.format(model, "cross_xquad_lang"))
|
139 |
|
140 |
|
141 |
df = pd.DataFrame(df_list)
|
|
|
156 |
|
157 |
return df
|
158 |
|
|
|
159 |
CROSS_XQUAD_ZERO_SHOT_LANGUAGE = get_data_cross_xquad_language(eval_mode="zero_shot")
|
160 |
CROSS_XQUAD_FIVE_SHOT_LANGUAGE = get_data_cross_xquad_language(eval_mode="five_shot")
|
161 |
|
|
|
174 |
df_list = []
|
175 |
|
176 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
177 |
|
178 |
try:
|
179 |
+
|
180 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cross_mmlu']]
|
181 |
+
|
182 |
overall_acc = [results['overall_acc'] for results in results_list]
|
183 |
overall_acc = median(overall_acc)
|
184 |
|
|
|
188 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
189 |
AC3_3 = median(AC3_3)
|
190 |
|
191 |
+
res = {
|
192 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
193 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
194 |
+
"Accuracy": overall_acc,
|
195 |
+
"Cross-Lingual Consistency": consistency_score_3,
|
196 |
+
"AC3": AC3_3,
|
197 |
+
}
|
198 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
199 |
|
200 |
+
except:
|
201 |
+
print('Not found in model: {} for {}'.format(model, "cross_mmlu_overall"))
|
202 |
|
203 |
|
204 |
df = pd.DataFrame(df_list)
|
|
|
219 |
|
220 |
return df
|
221 |
|
|
|
222 |
CROSS_MMLU_ZERO_SHOT_OVERALL = get_data_cross_mmlu_overall(eval_mode="zero_shot")
|
223 |
CROSS_MMLU_FIVE_SHOT_OVERALL = get_data_cross_mmlu_overall(eval_mode="five_shot")
|
224 |
|
|
|
228 |
df_list = []
|
229 |
|
230 |
for model in MODEL_LIST:
|
231 |
+
|
232 |
+
try:
|
233 |
|
234 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cross_mmlu']]
|
235 |
|
|
|
|
|
|
|
|
|
236 |
English = [results['language_acc']['English'] for results in results_list]
|
237 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
238 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
|
|
249 |
Spanish = median(Spanish)
|
250 |
Malay = median(Malay)
|
251 |
|
252 |
+
res = {
|
253 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
254 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
255 |
+
"English": English,
|
256 |
+
"Vietnamese": Vietnamese,
|
257 |
+
"Chinese": Chinese,
|
258 |
+
"Indonesian": Indonesian,
|
259 |
+
"Filipino": Filipino,
|
260 |
+
"Spanish": Spanish,
|
261 |
+
"Malay": Malay,
|
262 |
+
}
|
263 |
|
264 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
265 |
|
266 |
+
except:
|
267 |
+
print('Not found in model: {} for {}'.format(model, "cross_mmlu_lang"))
|
268 |
|
269 |
df = pd.DataFrame(df_list)
|
270 |
# If there are any models that are the same, merge them
|
|
|
284 |
|
285 |
return df
|
286 |
|
|
|
287 |
CROSS_MMLU_ZERO_SHOT_LANGUAGE = get_data_cross_mmlu_language(eval_mode="zero_shot")
|
288 |
CROSS_MMLU_FIVE_SHOT_LANGUAGE = get_data_cross_mmlu_language(eval_mode="five_shot")
|
289 |
|
|
|
298 |
df_list = []
|
299 |
|
300 |
for model in MODEL_LIST:
|
301 |
+
|
302 |
+
try:
|
303 |
|
304 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_logiqa'][res] for res in ALL_RESULTS[model][eval_mode]['cross_logiqa']]
|
305 |
|
|
|
|
|
|
|
|
|
306 |
overall_acc = [results['overall_acc'] for results in results_list]
|
307 |
overall_acc = median(overall_acc)
|
308 |
|
|
|
312 |
AC3_3 = [results['AC3_3'] for results in results_list]
|
313 |
AC3_3 = median(AC3_3)
|
314 |
|
315 |
+
res = {
|
316 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
317 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
318 |
+
"Accuracy": overall_acc,
|
319 |
+
"Cross-Lingual Consistency": consistency_score_3,
|
320 |
+
"AC3": AC3_3,
|
321 |
+
}
|
322 |
|
323 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
324 |
|
325 |
+
except:
|
326 |
+
print('Not found in model: {} for {}'.format(model, "cross_logiqa_overall"))
|
327 |
|
328 |
|
329 |
df = pd.DataFrame(df_list)
|
|
|
354 |
df_list = []
|
355 |
|
356 |
for model in MODEL_LIST:
|
357 |
+
|
358 |
+
try:
|
359 |
|
360 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cross_logiqa'][res] for res in ALL_RESULTS[model][eval_mode]['cross_logiqa']]
|
361 |
|
|
|
|
|
|
|
|
|
362 |
English = [results['language_acc']['English'] for results in results_list]
|
363 |
Vietnamese = [results['language_acc']['Vietnamese'] for results in results_list]
|
364 |
Chinese = [results['language_acc']['Chinese'] for results in results_list]
|
|
|
375 |
Spanish = median(Spanish)
|
376 |
Malay = median(Malay)
|
377 |
|
378 |
+
res = {
|
379 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
380 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
381 |
+
"English": English,
|
382 |
+
"Vietnamese": Vietnamese,
|
383 |
+
"Chinese": Chinese,
|
384 |
+
"Indonesian": Indonesian,
|
385 |
+
"Filipino": Filipino,
|
386 |
+
"Spanish": Spanish,
|
387 |
+
"Malay": Malay,
|
388 |
+
}
|
389 |
|
390 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
391 |
|
392 |
+
except:
|
393 |
+
print('Not found in model: {} for {}'.format(model, "cross_logiqa_language"))
|
394 |
|
395 |
+
|
396 |
|
397 |
df = pd.DataFrame(df_list)
|
398 |
# If there are any models that are the same, merge them
|
|
|
425 |
df_list = []
|
426 |
|
427 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
428 |
|
429 |
try:
|
430 |
+
|
431 |
+
results_list = [ALL_RESULTS[model][eval_mode]['sg_eval'][res] for res in ALL_RESULTS[model][eval_mode]['sg_eval']]
|
432 |
accuracy = median([results['accuracy'] for results in results_list])
|
433 |
|
434 |
+
res = {
|
435 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
436 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
437 |
+
"Accuracy": accuracy,
|
438 |
+
}
|
439 |
|
440 |
+
df_list.append(res)
|
441 |
+
|
442 |
+
except:
|
443 |
+
print('Not found in model: {} for {}'.format(model, "sg_eval"))
|
|
|
444 |
|
|
|
445 |
|
446 |
|
447 |
df = pd.DataFrame(df_list)
|
|
|
477 |
|
478 |
for model in MODEL_LIST:
|
479 |
|
|
|
|
|
|
|
|
|
480 |
try:
|
481 |
+
results_list = [ALL_RESULTS[model][eval_mode]['us_eval'][res] for res in ALL_RESULTS[model][eval_mode]['us_eval']]
|
482 |
accuracy = median([results['accuracy'] for results in results_list])
|
483 |
|
484 |
+
res = {
|
485 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
486 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
487 |
+
"Accuracy": accuracy,
|
488 |
+
}
|
489 |
|
490 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
491 |
|
492 |
+
except:
|
493 |
+
print('Not found in model: {} for {}'.format(model, "us_eval"))
|
494 |
|
495 |
|
496 |
df = pd.DataFrame(df_list)
|
|
|
525 |
df_list = []
|
526 |
|
527 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
528 |
|
529 |
try:
|
530 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cn_eval'][res] for res in ALL_RESULTS[model][eval_mode]['cn_eval']]
|
531 |
accuracy = median([results['accuracy'] for results in results_list])
|
532 |
|
533 |
+
res = {
|
534 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
535 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
536 |
+
"Accuracy": accuracy,
|
537 |
+
}
|
538 |
|
539 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
540 |
|
541 |
+
except:
|
542 |
+
print('Not found in model: {} for {}'.format(model, "cn_eval"))
|
543 |
|
544 |
df = pd.DataFrame(df_list)
|
545 |
# If there are any models that are the same, merge them
|
|
|
559 |
|
560 |
return df
|
561 |
|
|
|
562 |
CN_EVAL_ZERO_SHOT = get_data_cn_eval(eval_mode="zero_shot")
|
563 |
CN_EVAL_FIVE_SHOT = get_data_cn_eval(eval_mode="five_shot")
|
564 |
|
|
|
566 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
567 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
568 |
|
|
|
569 |
def get_data_ph_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
570 |
|
571 |
df_list = []
|
|
|
573 |
for model in MODEL_LIST:
|
574 |
|
575 |
|
|
|
576 |
|
577 |
|
578 |
try:
|
579 |
+
results_list = [ALL_RESULTS[model][eval_mode]['ph_eval'][res] for res in ALL_RESULTS[model][eval_mode]['ph_eval']]
|
580 |
accuracy = median([results['accuracy'] for results in results_list])
|
581 |
+
res = {
|
582 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
583 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
584 |
+
"Accuracy": accuracy,
|
585 |
+
}
|
586 |
|
587 |
+
df_list.append(res)
|
|
|
|
|
588 |
|
589 |
+
except:
|
590 |
+
print('Not found in model: {} for {}'.format(model, "ph_eval"))
|
|
|
|
|
|
|
|
|
|
|
591 |
|
592 |
|
593 |
df = pd.DataFrame(df_list)
|
|
|
622 |
df_list = []
|
623 |
|
624 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
625 |
|
626 |
try:
|
627 |
+
results_list = [ALL_RESULTS[model][eval_mode]['sing2eng'][res] for res in ALL_RESULTS[model][eval_mode]['sing2eng']]
|
628 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
629 |
|
630 |
+
res = {
|
631 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
632 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
633 |
+
"BLEU": bleu_score,
|
634 |
+
}
|
635 |
|
636 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
637 |
|
638 |
+
except:
|
639 |
+
print('Not found in model: {} for {}'.format(model, "sing2eng"))
|
640 |
|
641 |
|
642 |
df = pd.DataFrame(df_list)
|
|
|
670 |
df_list = []
|
671 |
|
672 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
673 |
|
674 |
try:
|
675 |
+
results_list = [ALL_RESULTS[model][eval_mode]['flores_ind2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_ind2eng']]
|
676 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
677 |
|
678 |
+
res = {
|
679 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
680 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
681 |
+
"BLEU": bleu_score,
|
682 |
+
}
|
683 |
|
684 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
685 |
|
686 |
+
except:
|
687 |
+
print('Not found in model: {} for {}'.format(model, "flores_ind2eng"))
|
688 |
|
689 |
|
690 |
df = pd.DataFrame(df_list)
|
|
|
720 |
df_list = []
|
721 |
|
722 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
723 |
|
724 |
try:
|
725 |
+
results_list = [ALL_RESULTS[model][eval_mode]['flores_vie2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_vie2eng']]
|
726 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
727 |
|
728 |
+
res = {
|
729 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
730 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
731 |
+
"BLEU": bleu_score,
|
732 |
+
}
|
|
|
|
|
|
|
|
|
733 |
|
734 |
+
df_list.append(res)
|
735 |
|
736 |
+
except:
|
737 |
+
print('Not found in model: {} for {}'.format(model, "flores_vie2eng"))
|
738 |
|
739 |
df = pd.DataFrame(df_list)
|
740 |
# If there are any models that are the same, merge them
|
|
|
767 |
df_list = []
|
768 |
|
769 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
770 |
|
771 |
try:
|
772 |
+
results_list = [ALL_RESULTS[model][eval_mode]['flores_zho2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_zho2eng']]
|
773 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
774 |
|
775 |
+
res = {
|
776 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
777 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
778 |
+
"BLEU": bleu_score,
|
779 |
+
}
|
|
|
|
|
|
|
|
|
780 |
|
781 |
+
df_list.append(res)
|
782 |
|
783 |
+
except:
|
784 |
+
print('Not found in model: {} for {}'.format(model, "flores_zho2eng"))
|
785 |
|
786 |
df = pd.DataFrame(df_list)
|
787 |
# If there are any models that are the same, merge them
|
|
|
801 |
|
802 |
return df
|
803 |
|
|
|
804 |
FLORES_ZHO2ENG_ZERO_SHOT = get_data_flores_zho2eng(eval_mode="zero_shot")
|
805 |
FLORES_ZHO2ENG_FIVE_SHOT = get_data_flores_zho2eng(eval_mode="five_shot")
|
806 |
|
|
|
814 |
df_list = []
|
815 |
|
816 |
for model in MODEL_LIST:
|
817 |
+
|
|
|
|
|
|
|
|
|
818 |
try:
|
819 |
+
results_list = [ALL_RESULTS[model][eval_mode]['flores_zsm2eng'][res] for res in ALL_RESULTS[model][eval_mode]['flores_zsm2eng']]
|
820 |
bleu_score = median([results['bleu_score'] for results in results_list])
|
821 |
|
822 |
+
res = {
|
823 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
824 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
825 |
+
"BLEU": bleu_score,
|
826 |
+
}
|
827 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
828 |
|
829 |
+
except:
|
830 |
+
print('Not found in model: {} for {}'.format(model, "flores_zsm2eng"))
|
831 |
|
832 |
df = pd.DataFrame(df_list)
|
833 |
# If there are any models that are the same, merge them
|
|
|
847 |
|
848 |
return df
|
849 |
|
|
|
850 |
FLORES_ZSM2ENG_ZERO_SHOT = get_data_flores_zho2eng(eval_mode="zero_shot")
|
851 |
FLORES_ZSM2ENG_FIVE_SHOT = get_data_flores_zho2eng(eval_mode="five_shot")
|
852 |
|
|
|
860 |
df_list = []
|
861 |
|
862 |
for model in MODEL_LIST:
|
863 |
+
|
|
|
|
|
|
|
|
|
864 |
try:
|
865 |
+
results_list = [ALL_RESULTS[model][eval_mode]['mmlu'][res] for res in ALL_RESULTS[model][eval_mode]['mmlu']]
|
866 |
accuracy = median([results['accuracy'] for results in results_list])
|
867 |
|
868 |
+
res = {
|
869 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
870 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
871 |
+
"Accuracy": accuracy,
|
872 |
+
}
|
873 |
+
df_list.append(res)
|
874 |
+
|
875 |
except:
|
876 |
accuracy = -1
|
877 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
878 |
df = pd.DataFrame(df_list)
|
879 |
# If there are any models that are the same, merge them
|
880 |
# E.g. if out["Model"] has the same value in two places, merge & take whichever one is not NaN else just take the first one
|
|
|
901 |
|
902 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
903 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
|
|
|
|
904 |
def get_data_mmlu_full(eval_mode='zero_shot', fillna=True, rank=True):
|
905 |
|
906 |
df_list = []
|
907 |
|
908 |
for model in MODEL_LIST:
|
909 |
+
|
|
|
|
|
|
|
|
|
910 |
try:
|
911 |
+
results_list = [ALL_RESULTS[model][eval_mode]['mmlu_full'][res] for res in ALL_RESULTS[model][eval_mode]['mmlu_full']]
|
912 |
accuracy = median([results['accuracy'] for results in results_list])
|
913 |
|
914 |
+
res = {
|
915 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
916 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
917 |
+
"Accuracy": accuracy,
|
918 |
+
}
|
919 |
|
920 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
921 |
|
922 |
+
except:
|
923 |
+
print('Not found in model: {} for {}'.format(model, "mmlu_full"))
|
924 |
|
925 |
|
926 |
df = pd.DataFrame(df_list)
|
|
|
941 |
|
942 |
return df
|
943 |
|
|
|
944 |
MMLU_FULL_ZERO_SHOT = get_data_mmlu_full(eval_mode="zero_shot")
|
945 |
MMLU_FULL_FIVE_SHOT = get_data_mmlu_full(eval_mode="five_shot")
|
946 |
|
947 |
|
948 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
949 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
|
|
|
|
950 |
def get_data_c_eval(eval_mode='zero_shot', fillna=True, rank=True):
|
951 |
|
952 |
df_list = []
|
953 |
|
954 |
+
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
|
|
955 |
try:
|
956 |
+
results_list = [ALL_RESULTS[model][eval_mode]['c_eval'][res] for res in ALL_RESULTS[model][eval_mode]['c_eval']]
|
957 |
accuracy = median([results['accuracy'] for results in results_list])
|
958 |
|
959 |
+
res = {
|
960 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
961 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
962 |
+
"Accuracy": accuracy,
|
963 |
+
}
|
|
|
|
|
|
|
|
|
964 |
|
965 |
+
df_list.append(res)
|
966 |
|
967 |
+
except:
|
968 |
+
print('Not found in model: {} for {}'.format(model, "c_eval"))
|
969 |
|
970 |
df = pd.DataFrame(df_list)
|
971 |
# If there are any models that are the same, merge them
|
|
|
985 |
|
986 |
return df
|
987 |
|
|
|
988 |
C_EVAL_ZERO_SHOT = get_data_c_eval(eval_mode="zero_shot")
|
989 |
C_EVAL_FIVE_SHOT = get_data_c_eval(eval_mode="five_shot")
|
990 |
|
|
|
998 |
df_list = []
|
999 |
|
1000 |
for model in MODEL_LIST:
|
1001 |
+
|
|
|
|
|
|
|
|
|
1002 |
try:
|
1003 |
+
results_list = [ALL_RESULTS[model][eval_mode]['c_eval_full'][res] for res in ALL_RESULTS[model][eval_mode]['c_eval_full']]
|
1004 |
accuracy = median([results['accuracy'] for results in results_list])
|
1005 |
|
1006 |
+
res = {
|
1007 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1008 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1009 |
+
"Accuracy": accuracy,
|
1010 |
+
}
|
1011 |
|
1012 |
+
df_list.append(res)
|
1013 |
+
|
1014 |
+
except:
|
1015 |
+
print('Not found in model: {} for {}'.format(model, "c_eval_full"))
|
1016 |
|
|
|
|
|
|
|
|
|
|
|
1017 |
|
|
|
1018 |
|
1019 |
|
1020 |
df = pd.DataFrame(df_list)
|
|
|
1051 |
df_list = []
|
1052 |
|
1053 |
for model in MODEL_LIST:
|
1054 |
+
|
|
|
|
|
|
|
|
|
1055 |
try:
|
1056 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cmmlu'][res] for res in ALL_RESULTS[model][eval_mode]['cmmlu']]
|
1057 |
accuracy = median([results['accuracy'] for results in results_list])
|
1058 |
|
1059 |
+
res = {
|
1060 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1061 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1062 |
+
"Accuracy": accuracy,
|
1063 |
+
}
|
1064 |
+
|
1065 |
+
df_list.append(res)
|
1066 |
+
|
1067 |
except:
|
1068 |
+
print('Not found in model: {} for {}'.format(model, "cmmlu"))
|
1069 |
|
1070 |
|
|
|
|
|
|
|
|
|
|
|
1071 |
|
|
|
1072 |
|
1073 |
|
1074 |
df = pd.DataFrame(df_list)
|
|
|
1095 |
|
1096 |
|
1097 |
|
|
|
|
|
|
|
1098 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1099 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1100 |
|
|
|
1104 |
df_list = []
|
1105 |
|
1106 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1107 |
|
1108 |
try:
|
1109 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cmmlu_full'][res] for res in ALL_RESULTS[model][eval_mode]['cmmlu_full']]
|
1110 |
accuracy = median([results['accuracy'] for results in results_list])
|
1111 |
|
1112 |
+
res = {
|
1113 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1114 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1115 |
+
"Accuracy": accuracy,
|
1116 |
+
}
|
1117 |
+
|
1118 |
+
df_list.append(res)
|
1119 |
+
|
1120 |
except:
|
1121 |
+
print('Not found in model: {} for {}'.format(model, "cmmlu_full"))
|
1122 |
|
1123 |
|
|
|
|
|
|
|
|
|
|
|
1124 |
|
|
|
1125 |
|
1126 |
|
1127 |
df = pd.DataFrame(df_list)
|
|
|
1157 |
df_list = []
|
1158 |
|
1159 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
|
|
1160 |
try:
|
1161 |
+
results_list = [ALL_RESULTS[model][eval_mode]['zbench'][res] for res in ALL_RESULTS[model][eval_mode]['zbench']]
|
1162 |
accuracy = median([results['accuracy'] for results in results_list])
|
1163 |
|
1164 |
+
res = {
|
1165 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1166 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1167 |
+
"Accuracy": accuracy,
|
1168 |
+
}
|
1169 |
|
1170 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
1171 |
|
1172 |
+
except:
|
1173 |
+
print('Not found in model: {} for {}'.format(model, "zbench"))
|
1174 |
|
1175 |
|
1176 |
df = pd.DataFrame(df_list)
|
|
|
1205 |
|
1206 |
for model in MODEL_LIST:
|
1207 |
|
|
|
1208 |
|
1209 |
try:
|
1210 |
+
results_list = [ALL_RESULTS[model][eval_mode]['indommlu'][res] for res in ALL_RESULTS[model][eval_mode]['indommlu']]
|
1211 |
accuracy = median([results['accuracy'] for results in results_list])
|
1212 |
|
1213 |
+
res = {
|
1214 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1215 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1216 |
+
"Accuracy": accuracy,
|
1217 |
+
}
|
1218 |
+
|
1219 |
+
df_list.append(res)
|
1220 |
+
|
1221 |
except:
|
1222 |
+
print('Not found in model: {} for {}'.format(model, "indommlu"))
|
1223 |
|
|
|
|
|
|
|
|
|
|
|
1224 |
|
|
|
1225 |
|
1226 |
|
1227 |
df = pd.DataFrame(df_list)
|
|
|
1249 |
|
1250 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1251 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
|
|
|
|
1252 |
def get_data_ind_emotion(eval_mode='zero_shot', fillna=True, rank=True):
|
1253 |
|
1254 |
df_list = []
|
1255 |
|
1256 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
|
|
1257 |
try:
|
1258 |
+
results_list = [ALL_RESULTS[model][eval_mode]['ind_emotion'][res] for res in ALL_RESULTS[model][eval_mode]['ind_emotion']]
|
1259 |
accuracy = median([results['accuracy'] for results in results_list])
|
1260 |
|
1261 |
+
res = {
|
1262 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1263 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1264 |
+
"Accuracy": accuracy,
|
1265 |
+
}
|
|
|
|
|
|
|
|
|
1266 |
|
1267 |
+
df_list.append(res)
|
1268 |
|
1269 |
+
except:
|
1270 |
+
print('Not found in model: {} for {}'.format(model, "ind_emotion"))
|
1271 |
|
1272 |
df = pd.DataFrame(df_list)
|
1273 |
# If there are any models that are the same, merge them
|
|
|
1287 |
|
1288 |
return df
|
1289 |
|
|
|
1290 |
IND_EMOTION_ZERO_SHOT = get_data_ind_emotion(eval_mode="zero_shot")
|
1291 |
IND_EMOTION_FIVE_SHOT = get_data_ind_emotion(eval_mode="five_shot")
|
1292 |
|
|
|
1302 |
df_list = []
|
1303 |
|
1304 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1305 |
|
1306 |
try:
|
1307 |
+
results_list = [ALL_RESULTS[model][eval_mode]['ocnli'][res] for res in ALL_RESULTS[model][eval_mode]['ocnli']]
|
1308 |
accuracy = median([results['accuracy'] for results in results_list])
|
1309 |
|
1310 |
+
res = {
|
1311 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1312 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1313 |
+
"Accuracy": accuracy,
|
1314 |
+
}
|
1315 |
|
1316 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
1317 |
|
1318 |
+
except:
|
1319 |
+
print('Not found in model: {} for {}'.format(model, "ocnli"))
|
1320 |
|
1321 |
|
1322 |
df = pd.DataFrame(df_list)
|
|
|
1352 |
df_list = []
|
1353 |
|
1354 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1355 |
|
1356 |
try:
|
1357 |
+
results_list = [ALL_RESULTS[model][eval_mode]['c3'][res] for res in ALL_RESULTS[model][eval_mode]['c3']]
|
1358 |
accuracy = median([results['accuracy'] for results in results_list])
|
1359 |
|
1360 |
+
res = {
|
1361 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1362 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1363 |
+
"Accuracy": accuracy,
|
1364 |
+
}
|
1365 |
|
1366 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
1367 |
|
1368 |
+
except:
|
1369 |
+
print('Not found in model: {} for {}'.format(model, "c3"))
|
1370 |
|
1371 |
df = pd.DataFrame(df_list)
|
1372 |
# If there are any models that are the same, merge them
|
|
|
1401 |
df_list = []
|
1402 |
|
1403 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1404 |
|
1405 |
try:
|
1406 |
+
results_list = [ALL_RESULTS[model][eval_mode]['dream'][res] for res in ALL_RESULTS[model][eval_mode]['dream']]
|
1407 |
accuracy = median([results['accuracy'] for results in results_list])
|
1408 |
|
1409 |
+
res = {
|
1410 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1411 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1412 |
+
"Accuracy": accuracy,
|
1413 |
+
}
|
1414 |
|
1415 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
1416 |
|
1417 |
+
except:
|
1418 |
+
print('Not found in model: {} for {}'.format(model, "dream"))
|
1419 |
|
1420 |
|
1421 |
df = pd.DataFrame(df_list)
|
|
|
1436 |
|
1437 |
return df
|
1438 |
|
|
|
1439 |
DREAM_ZERO_SHOT = get_data_dream(eval_mode="zero_shot")
|
1440 |
DREAM_FIVE_SHOT = get_data_dream(eval_mode="five_shot")
|
1441 |
|
|
|
|
|
1442 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1443 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
|
|
|
|
1444 |
def get_data_samsum(eval_mode='zero_shot', fillna=True, rank=True):
|
1445 |
|
1446 |
df_list = []
|
1447 |
|
1448 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1449 |
|
1450 |
try:
|
1451 |
+
results_list = [ALL_RESULTS[model][eval_mode]['samsum'][res] for res in ALL_RESULTS[model][eval_mode]['samsum']]
|
1452 |
+
|
1453 |
rouge1 = median([results['rouge1'] for results in results_list])
|
1454 |
rouge2 = median([results['rouge2'] for results in results_list])
|
1455 |
rougeL = median([results['rougeL'] for results in results_list])
|
1456 |
|
1457 |
+
res = {
|
1458 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1459 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1460 |
+
"ROUGE-1": rouge1,
|
1461 |
+
"ROUGE-2": rouge2,
|
1462 |
+
"ROUGE-L": rougeL,
|
1463 |
+
}
|
1464 |
|
1465 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1466 |
|
1467 |
+
except:
|
1468 |
+
print('Not found in model: {} for {}'.format(model, "samsum"))
|
1469 |
|
1470 |
df = pd.DataFrame(df_list)
|
1471 |
# If there are any models that are the same, merge them
|
|
|
1499 |
df_list = []
|
1500 |
|
1501 |
for model in MODEL_LIST:
|
1502 |
+
|
|
|
|
|
|
|
|
|
1503 |
try:
|
1504 |
+
results_list = [ALL_RESULTS[model][eval_mode]['dialogsum'][res] for res in ALL_RESULTS[model][eval_mode]['dialogsum']]
|
1505 |
+
|
1506 |
rouge1 = median([results['rouge1'] for results in results_list])
|
1507 |
rouge2 = median([results['rouge2'] for results in results_list])
|
1508 |
rougeL = median([results['rougeL'] for results in results_list])
|
1509 |
|
1510 |
+
res = {
|
1511 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1512 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1513 |
+
"ROUGE-1": rouge1,
|
1514 |
+
"ROUGE-2": rouge2,
|
1515 |
+
"ROUGE-L": rougeL,
|
1516 |
+
}
|
1517 |
+
|
1518 |
+
df_list.append(res)
|
1519 |
+
|
1520 |
except:
|
1521 |
+
print('Not found in model: {} for {}'.format(model, "dialogsum"))
|
|
|
|
|
1522 |
|
1523 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1524 |
|
|
|
1525 |
|
1526 |
|
1527 |
df = pd.DataFrame(df_list)
|
|
|
1559 |
|
1560 |
for model in MODEL_LIST:
|
1561 |
|
|
|
|
|
|
|
|
|
1562 |
try:
|
1563 |
+
results_list = [ALL_RESULTS[model][eval_mode]['sst2'][res] for res in ALL_RESULTS[model][eval_mode]['sst2']]
|
1564 |
accuracy = median([results['accuracy'] for results in results_list])
|
1565 |
|
1566 |
+
res = {
|
1567 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1568 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1569 |
+
"Accuracy": accuracy,
|
1570 |
+
}
|
1571 |
+
|
1572 |
+
df_list.append(res)
|
1573 |
+
|
1574 |
except:
|
1575 |
+
print('Not found in model: {} for {}'.format(model, "sst2"))
|
1576 |
|
1577 |
|
|
|
|
|
|
|
|
|
|
|
1578 |
|
|
|
1579 |
|
1580 |
|
1581 |
df = pd.DataFrame(df_list)
|
|
|
1612 |
df_list = []
|
1613 |
|
1614 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1615 |
|
1616 |
try:
|
1617 |
+
results_list = [ALL_RESULTS[model][eval_mode]['cola'][res] for res in ALL_RESULTS[model][eval_mode]['cola']]
|
1618 |
accuracy = median([results['accuracy'] for results in results_list])
|
1619 |
|
1620 |
+
res = {
|
1621 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1622 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1623 |
+
"Accuracy": accuracy,
|
1624 |
+
}
|
|
|
|
|
|
|
|
|
1625 |
|
1626 |
+
df_list.append(res)
|
1627 |
|
1628 |
+
except:
|
1629 |
+
print('Not found in model: {} for {}'.format(model, "cola"))
|
1630 |
|
1631 |
df = pd.DataFrame(df_list)
|
1632 |
# If there are any models that are the same, merge them
|
|
|
1664 |
|
1665 |
for model in MODEL_LIST:
|
1666 |
|
|
|
|
|
|
|
|
|
1667 |
try:
|
1668 |
+
results_list = [ALL_RESULTS[model][eval_mode]['qqp'][res] for res in ALL_RESULTS[model][eval_mode]['qqp']]
|
1669 |
accuracy = median([results['accuracy'] for results in results_list])
|
1670 |
|
1671 |
+
res = {
|
1672 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1673 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1674 |
+
"Accuracy": accuracy,
|
1675 |
+
}
|
1676 |
|
1677 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
1678 |
|
1679 |
+
except:
|
1680 |
+
print('Not found in model: {} for {}'.format(model, "qqp"))
|
1681 |
|
1682 |
|
1683 |
df = pd.DataFrame(df_list)
|
|
|
1715 |
df_list = []
|
1716 |
|
1717 |
for model in MODEL_LIST:
|
1718 |
+
|
|
|
|
|
|
|
|
|
1719 |
try:
|
1720 |
+
results_list = [ALL_RESULTS[model][eval_mode]['mnli'][res] for res in ALL_RESULTS[model][eval_mode]['mnli']]
|
1721 |
accuracy = median([results['accuracy'] for results in results_list])
|
1722 |
|
1723 |
+
res = {
|
1724 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1725 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1726 |
+
"Accuracy": accuracy,
|
1727 |
+
}
|
1728 |
|
1729 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
1730 |
|
1731 |
+
except:
|
1732 |
+
print('Not found in model: {} for {}'.format(model, "mnli"))
|
1733 |
|
1734 |
|
1735 |
df = pd.DataFrame(df_list)
|
|
|
1767 |
df_list = []
|
1768 |
|
1769 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1770 |
|
1771 |
try:
|
1772 |
+
results_list = [ALL_RESULTS[model][eval_mode]['qnli'][res] for res in ALL_RESULTS[model][eval_mode]['qnli']]
|
1773 |
accuracy = median([results['accuracy'] for results in results_list])
|
1774 |
|
1775 |
+
res = {
|
1776 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1777 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1778 |
+
"Accuracy": accuracy,
|
1779 |
+
}
|
|
|
|
|
|
|
|
|
1780 |
|
1781 |
+
df_list.append(res)
|
1782 |
|
1783 |
+
except:
|
1784 |
+
print('Not found in model: {} for {}'.format(model, "qnli"))
|
1785 |
|
1786 |
df = pd.DataFrame(df_list)
|
1787 |
# If there are any models that are the same, merge them
|
|
|
1818 |
df_list = []
|
1819 |
|
1820 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
1821 |
|
1822 |
try:
|
1823 |
+
results_list = [ALL_RESULTS[model][eval_mode]['wnli'][res] for res in ALL_RESULTS[model][eval_mode]['wnli']]
|
1824 |
accuracy = median([results['accuracy'] for results in results_list])
|
1825 |
|
1826 |
+
res = {
|
1827 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1828 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1829 |
+
"Accuracy": accuracy,
|
1830 |
+
}
|
|
|
|
|
|
|
|
|
1831 |
|
1832 |
+
df_list.append(res)
|
1833 |
|
1834 |
+
except:
|
1835 |
+
print('Not found in model: {} for {}'.format(model, "wnli"))
|
1836 |
|
1837 |
df = pd.DataFrame(df_list)
|
1838 |
# If there are any models that are the same, merge them
|
|
|
1852 |
|
1853 |
return df
|
1854 |
|
|
|
1855 |
WNLI_ZERO_SHOT = get_data_wnli(eval_mode="zero_shot")
|
1856 |
WNLI_FIVE_SHOT = get_data_wnli(eval_mode="five_shot")
|
1857 |
|
1858 |
|
|
|
|
|
|
|
1859 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1860 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1861 |
|
|
|
1865 |
df_list = []
|
1866 |
|
1867 |
for model in MODEL_LIST:
|
|
|
|
|
|
|
|
|
|
|
1868 |
try:
|
1869 |
+
results_list = [ALL_RESULTS[model][eval_mode]['rte'][res] for res in ALL_RESULTS[model][eval_mode]['rte']]
|
1870 |
accuracy = median([results['accuracy'] for results in results_list])
|
1871 |
|
1872 |
+
res = {
|
1873 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1874 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1875 |
+
"Accuracy": accuracy,
|
1876 |
+
}
|
|
|
|
|
|
|
|
|
1877 |
|
1878 |
+
df_list.append(res)
|
1879 |
|
1880 |
+
except:
|
1881 |
+
print('Not found in model: {} for {}'.format(model, "rte"))
|
1882 |
|
1883 |
df = pd.DataFrame(df_list)
|
1884 |
# If there are any models that are the same, merge them
|
|
|
1903 |
RTE_FIVE_SHOT = get_data_rte(eval_mode="five_shot")
|
1904 |
|
1905 |
|
|
|
|
|
|
|
|
|
1906 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
1907 |
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
|
|
|
|
|
1908 |
def get_data_mrpc(eval_mode='zero_shot', fillna=True, rank=True):
|
1909 |
|
1910 |
df_list = []
|
1911 |
|
1912 |
for model in MODEL_LIST:
|
1913 |
+
|
|
|
|
|
|
|
|
|
1914 |
try:
|
1915 |
+
results_list = [ALL_RESULTS[model][eval_mode]['mrpc'][res] for res in ALL_RESULTS[model][eval_mode]['mrpc']]
|
1916 |
accuracy = median([results['accuracy'] for results in results_list])
|
1917 |
|
1918 |
+
res = {
|
1919 |
+
"Model Size (Params)": MODEL_TO_SIZE.get(model, ""),
|
1920 |
+
"Model": make_clickable_model(model, link=ALL_RESULTS[model]["model_link"]),
|
1921 |
+
"Accuracy": accuracy,
|
1922 |
+
}
|
1923 |
|
1924 |
+
df_list.append(res)
|
|
|
|
|
|
|
|
|
|
|
|
|
1925 |
|
1926 |
+
except:
|
1927 |
+
print('Not found in model: {} for {}'.format(model, "mrpc"))
|
1928 |
|
1929 |
df = pd.DataFrame(df_list)
|
1930 |
# If there are any models that are the same, merge them
|
|
|
2021 |
- **Mode of Evaluation**: Zero-Shot, Five-Shot
|
2022 |
|
2023 |
### The following table shows the performance of the models on the SeaEval benchmark.
|
2024 |
+
- For **Zero-Shot** performance, it is the median value from 5 distinct prompts shown on the above leaderboard to mitigate the influence of random variations induced by prompts.
|
2025 |
+
- I am trying to evaluate the base models for five-shot performance and instruction-tuned models for zero-shot.
|
2026 |
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
2027 |
|
2028 |
""")
|
|
|
2159 |
|
2160 |
|
2161 |
|
2162 |
+
with gr.TabItem("Cultural Reasoning"):
|
2163 |
|
2164 |
# dataset 3: SG_EVAL
|
2165 |
with gr.TabItem("SG_EVAL"):
|
|
|
2508 |
""")
|
2509 |
|
2510 |
|
2511 |
+
with gr.TabItem("FLORES-Translation"):
|
2512 |
|
2513 |
|
2514 |
# dataset 8:
|
|
|
2616 |
""")
|
2617 |
|
2618 |
|
2619 |
+
with gr.TabItem("Emotion"):
|
2620 |
|
2621 |
# dataset 18:
|
2622 |
with gr.TabItem("ind_emotion"):
|
|
|
2752 |
|
2753 |
|
2754 |
|
2755 |
+
with gr.TabItem("Fundamental NLP Tasks"):
|
2756 |
|
2757 |
|
2758 |
# dataset
|