REVIEW 7 cited by
SeaLLMs -- Large Language Models for Southeast Asia
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Despite the remarkable achievements of large language models (LLMs) in various tasks, there remains a linguistic bias that favors high-resource languages, such as English, often at the expense of low-resource and regional languages. To address this imbalance, we introduce SeaLLMs, an innovative series of language models that specifically focuses on Southeast Asian (SEA) languages. SeaLLMs are built upon the Llama-2 model and further advanced through continued pre-training with an extended vocabulary, specialized instruction and alignment tuning to better capture the intricacies of regional languages. This allows them to respect and reflect local cultural norms, customs, stylistic preferences, and legal considerations. Our comprehensive evaluation demonstrates that SeaLLM-13b models exhibit superior performance across a wide spectrum of linguistic tasks and assistant-style instruction-following capabilities relative to comparable open-source models. Moreover, they outperform ChatGPT-3.5 in non-Latin languages, such as Thai, Khmer, Lao, and Burmese, by large margins while remaining lightweight and cost-effective to operate.
Forward citations
Cited by 7 Pith papers
-
MyCulture: Exploring Malaysia's Diverse Culture under Low-Resource Language Constraints
MyCulture, a new Malay-language cultural benchmark, shows LLM accuracy drops by at least 17% when multiple-choice questions are converted to an open-ended format.
-
Adapting Language-Specific LLMs to a Reasoning Model in One Day via Model Merging -- An Open Recipe
A Thai 70B model trained with an SFT-plus-DARE-merge recipe matches DeepSeek R1 on reasoning benchmarks while retaining most Thai language quality.
-
SEALGuard: Safeguarding the Multilingual Conversations in Southeast Asian Languages for LLM Software Systems
A LoRA-adapted SeaLLM model detects unsafe and jailbreak prompts in nine Southeast Asian languages with 97% recall and 98% F1 on the authors' new SEALSBench benchmark, far above zero-shot LlamaGuard.
-
Do Large Language Models Know Folktales? A Case Study of Yokai in Japanese Folktales
A benchmark of 809 yokai questions shows Japanese-centric LLMs, particularly Llama-3-based continual pretraining models, outperform English-centric models on Japanese folktale knowledge.
-
Improving Multilingual Social Media Insights: Aspect-based Comment Analysis
A new multilingual dataset and an SFT+DPO LLM pipeline for generating comment aspect terms yields small clustering improvements, but the benchmark construction and missing artifacts limit the strength of the evidence.
-
Less, but Better: Efficient Multilingual Expansion for LLMs via Layer-wise Mixture-of-Experts
A layer-wise expert allocation algorithm based on hidden-state similarity, plus a routing classifier, improves parameter efficiency and reduces forgetting when expanding LLMs to new languages.
-
FiLLM -- A Filipino-optimized Large Language Model based on Southeast Asia Large Language Model (SEALLM)
A LoRA-tuned Filipino LLM is compared to CalamanCy and found weaker, but the reported numbers and statistical test are internally contradictory.
Discussion (0). Continue with ORCID to comment.