whereisanzi/ maracatu-80m:latest

10 1 month ago

Open-weight Brazilian Portuguese LLM trained from scratch on 1.6B tokens. 87.8M params, Llama-style with GQA. Validation perplexity 21.34. Apache 2.0. Base model.

... /
params
e64d9a60fab9 · 79B
{
"num_ctx": 1024,
"repeat_penalty": 1.1,
"temperature": 0.8,
"top_k": 50,
"top_p": 0.9
}