Resources
Explore the data and resources used in Latam-GPT
Trueque Benchmark
Trueque is a human-reviewed collaborative evaluation benchmark for measuring LLM performance on questions about Latin American knowledge and culture.
Explore 500 curated questions on history, culture, geography, and gastronomy from 20 Latin American countries.
Dataset available on Hugging Face
CHOCLO
CHOCLO is a benchmark specialized in Latin American cultural knowledge to evaluate how well language models understand and represent the culture of the region.
Over 100,000 rows with questions on geography, fauna, flora, traditions, gastronomy, and public figures from 18 countries, with three difficulty levels.
Dataset available on Hugging Face
Copuchat - Contribute Data
Copuchat is an experimental application built on GPT 4.1, by OpenAI, that simulates real conversations with users from Latin America and the Caribbean to improve the alignment of future versions of Latam-GPT.
Help improve Latam-GPT and participate in anonymous conversations that will be useful for training the model.
Participate in conversations to contribute to Latam-GPT training