Cabrita
Pioneering Brazilian Portuguese instruction-tuned LLM based on OpenLLaMA 3B.
About
Cabrita is one of the first instruction-tuning efforts for Brazilian Portuguese, developed by the 22-hours team and released in 2023. It is based on the OpenLLaMA 3B architecture with Low-Rank Adaptation (LoRA), fine-tuned on the Portuguese portion of mC4 with a custom Portuguese tokenizer.
The follow-up Open Cabrita 3B effort documented a full experimental timeline: the 3B configuration trained stably on a single TPU v3-8 pod over about two weeks, while larger LLaMA-2 and Gemma variants failed to converge, leading the team to conclude that data quality alignment matters more than raw scale for Portuguese training. The release date below is era-level (2023) since only year granularity is verifiable.
Cabrita is cited in the Brazilian LLM literature as a pioneering community effort alongside Sabiá and Bode. Weights are openly available on Hugging Face under the 22h organization.
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 — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 111; Trending score 55/100 from researcher assessment.