Crossing Linguistic Horizons

Finetuning and Comprehensive Evaluation of Vietnamese Large Language Models

Bias-Toxicity Question Answering Leaderboard

Models	XQuAD					MLQA
Models	DRR→\|	DRG→\|	SAR→\|	SAG→\|	Tox↓	DRR→\|	DRG→\|	SAR→\|	SAG→\|	Tox↓
URA-LLaMa 70B	-	0.39 ± 0.01	-	0.41 ± 0.00	0.02 ± 0.00	-	0.14 ± 0.02	-	0.42 ± 0.03	0.02 ± 0.00
URA-LLaMa 13B	-	0.39 ± 0.01	-	0.45 ± 0.01	0.02 ± 0.00	-	0.17 ± 0.1	-	0.38 ± 0.00	0.02 ± 0.00
URA-LLaMa 7B	-	0.43 ± 0.01	-	0.48 ± 0.00	0.03 ± 0.00	-	0.18 ± 0.01	-	0.37 ± 0.01	0.02 ± 0.00
LLaMa-2 13B	-	0.35 ± 0.03	-	0.46 ± 0.00	0.01 ± 0.00	-	0.27 ± 0.01	-	0.43 ± 0.00	0.01 ± 0.00
LLaMa-2 7B	-	0.46 ± 0.01	-	0.42 ± 0.00	0.01 ± 0.00	-	0.21 ± 0.06	-	0.45 ± 0.00	0.01 ± 0.00
Vietcuna 7B	-	0.50 ± 0.00	-	-	0.04 ± 0.00	-	0.23 ± 0.09	-	0.49 ± 0.01	0.04 ± 0.00
GPT-3.5	-	0.43 ± 0.01	-	0.48 ± 0.00	0.02 ± 0.00	-	0.18 ± 0.01	-	0.40 ± 0.00	0.02 ± 0.00
GPT-4	-	0.40 ± 0.01	-	0.45 ± 0.00	0.02 ± 0.00	-	0.16 ± 0.01	-	0.41 ± 0.01	0.02 ± 0.00