Evaluation profile

IndoBias-Pairs — parity-aware culturally grounded bias

1sub-evals

0.62%total index weight

1components

Within-component eval weight: Fairness 6.2%.

Model score (higher is better)Predicted score

About this eval

Culturally grounded prototypical-association parity across Indonesian languages and social-bias domains.

Included in the behavior ranking.

Measure	Component	Direction	Total index weight	Within-component weight
parity_scoreindobias/indobias.csv:parity_scoreMeasures whether culturally grounded paired associations are balanced across five bias domains and four Indonesian languages.	fairness_nondiscrimination:1.000indobias-pairs	Higher is better	0.62%	Fairness 6.2%

Measures whether culturally grounded paired associations are balanced across five bias domains and four Indonesian languages.

Rank	Model	Value	Provenance
1	olmo-3-7b	91.96	official
2	olmo-3-7b-instruct	91.81	official
3	multilingual-e5-base	89.76	official
4	llama-2-7b-chat	89.73	official
5	bert-base-multilingual-cased	89.61	official
6	llama-2-7b	88.58	official
7	xlm-roberta-base	88.11	official
8	gemma-3-4b	87.05	official
9	merak-7b-v4	86.96	official
10	qwen2-7b-instruct	86.94	official
11	qwen-2.5-7b-instruct	86.3	official
12	gemma-2-9b-it	85.94	official
13	qwen3-8b	85.73	official
14	qwen2-7b	85.7	official
15	qwen2.5-7b	85.28	official
16	llama-3.1-8b-instruct	85.01	official
17	gemma-sea-lion-v3-9b	84.95	official
18	komodo-7b-base	84.78	official
19	qwen3-8b-base	84.48	official
20	seallms-v3-7b-chat	84.38	official
21	seallms-v3-7b	84.11	official
22	gemma-sea-lion-v3-9b-it	83.91	official
23	gemma-3-4b-pt	82.53	official
24	llama-3.1-8b-base	81.67	official
25	sahabatai-v1-instruct	81.25	official
26	sahabatai-v1-base	80.92	official