Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• AMALIA is a fully open source large language model specifically designed for European Portuguese.technical 100%
• AMALIA prioritizes European Portuguese by using more high-quality pt-PT data during mid- and post-training stages.method 100%
• A suite of European Portuguese benchmarks, including translated standard tasks and four new datasets, was released to evaluate pt-PT more faithfully.method 100%
• AMALIA matches strong baselines on translated benchmarks while substantially improving performance on pt-PT-specific evaluations.result 100%
• The primary source for AMALIA's European Portuguese data collection was Arquivo.pt, the Portuguese Web Archive.method 100%
• The creation of AMALIA involved pretraining on a large-scale European Portuguese text corpus and post-training on specific datasets for instruction following, conversational reasoning, and problem solving.method 90%
• A strategy of sampling 32 candidate responses, scoring them with ArmoRM, and selecting the highest- and lowest-reward responses for prompts led to overall performance improvements but also declines in mathematical reasoning and instruction following.method 90%
• AMALIA is a fully open source large language model specifically designed for European Portuguese.technical 100%
• AMALIA prioritizes European Portuguese by using more high-quality pt-PT data during mid- and post-training stages.method 100%
• A suite of European Portuguese benchmarks, including translated standard tasks and four new datasets, has been released to evaluate pt-PT more faithfully.method 100%
• AMALIA matches strong baselines on translated benchmarks.result 100%
• AMALIA substantially improves performance on European Portuguese-specific evaluations.result 100%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card