Nekomata-14B
Japanese-continual-pretrained 14B model built on Qwen, trained on Trainium.
About
Nekomata-14B is a Japanese language model released by rinna on December 21, 2023. It continues pre-training of Qwen-14B on 66 billion tokens of mixed Japanese and English data, including Japanese CC-100, C4, OSCAR, the Pile, Wikipedia, and a rinna-curated Japanese dataset. Training ran on AWS Trainium accelerators across 16 EC2 trn1.32xlarge nodes in about 7 days.
The name comes from the nekomata, a cat-like creature from Japanese folklore. The model inherits Qwen's 150k+ vocabulary for efficient Japanese tokenization and supports 8192-token sequences. Note the custom license: it inherits the Tongyi Qianwen model license terms rather than a standard open-source license, so check commercial-use terms before production deployment.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 128; Trending score 74/100 from researcher assessment.