---
language:
- pt
license: apache-2.0
base_model: marinarosa/minicpm5-1b-vivamais-v1
library_name: gguf
tags:
- gguf
- llama-cpp
- minicpm5
- vivamais
- brazilian-portuguese
- grounded-qa
---

# minicpm5-1b-vivamais-v1-GGUF

GGUF export of `marinarosa/minicpm5-1b-vivamais-v1` for llama.cpp-compatible local inference.

- Quantization: `Q4_K_M`
- Model file: `minicpm5-1b-vivamais-v1-Q4_K_M.gguf`
- Target use: Viva Mais dashboard Q&A in Brazilian Portuguese

This repo contains only the redacted fine-tuned model export. Raw WhatsApp
exports, full transcriptions, and client identifiers are not published.

## llama.cpp

Use a recent llama.cpp build with MiniCPM5 support:

```bash
cat > prompt.txt <<'EOF'
<|im_start|>system
Voce responde perguntas do painel Viva Mais usando apenas o contexto fornecido.
<|im_end|>
<|im_start|>user
Pergunta aqui
<|im_end|>
<|im_start|>assistant
EOF

llama-cli -m minicpm5-1b-vivamais-v1-Q4_K_M.gguf -c 4096 -ngl 99 -f prompt.txt
```

For app parity, render chat prompts with the MiniCPM5 chat template and
`enable_thinking=False`.