S5-nano-GGUF

S5-nano-GGUF

Runs Maritaca AI's S5-nano, a 4B model tuned for Brazilian Portuguese with function calling and a think mode, from GGUF files of 2.8 to 8.7 GB for llama.cpp and compatible apps.

Made by maritaca-ai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: S5-nano-GGUF holds GGUF copies of S5-nano, a 4B-parameter model from Maritaca AI. Maritaca built it on Qwen3.5-4B with more pretraining and post-training, and the card says it improves a lot on Brazilian benchmarks. The repository offers three sizes for llama.cpp and compatible apps: BF16 at 8.7 GB, Q8_0 at 4.6 GB and Q4_K_M at 2.8 GB.

Who it suits: People who want a small model that works well in Brazilian Portuguese and handles function calling, tool use and structured output on their own machine.

Industry | Who uses it | How they use it | Example
Software development | Developers in Brazil | Add a local Portuguese-language model to an app | A chat feature that answers in Portuguese
Agent building | Agent developers | Use function calling and structured output | An agent that calls a tool with typed fields
Education | Teachers and students | Practice with a model that reads Brazilian text | Questions answered in Brazilian Portuguese
Research | Language-model researchers | Compare small quantized models | Q4_K_M against Q8_0

If you want the GPU-ready BF16 or FP8 weights for vLLM or SGLang, you can skip this repository, because they live in the original repositories.

Where to use it:

• llama.cpp and apps that load GGUF files
• Windows, macOS or Linux, after you fetch the files with Python
• vLLM and SGLang, but through the original weights rather than this repository

Use case | Who | What they get back
Local chat | Individuals | Answers in Portuguese or English from a small model
Tool use | Agent builders | Function calls and structured outputs
Long documents | Analysts | A model with a context window of up to 262,144 tokens

How to set it up:

• Python and pip
• Disk space for the file you pick, from 2.8 GB to 8.7 GB
• llama.cpp or another app that loads GGUF files

1. Install the Hugging Face hub tools.
2. Download the model files.
3. Open one GGUF file in llama.cpp or your app, following the model card.

pip install -U huggingface_hub
hf download maritaca-ai/S5-nano-GGUF

How to use it: Load a file in llama.cpp and chat. The model has two modes. Think reasons before it answers and is the default. Instruct answers directly with no reasoning step.

Example: You ask a question in Brazilian Portuguese with the model in Think mode. It reasons first and then answers. In Instruct mode, the same question gets a direct reply with no reasoning step.

For newcomers and experts: If you are new to GGUF, start with Q4_K_M, the smallest file, and ask a short question in Portuguese. Switch to Q8_0 if you want closer quality and have the memory.

The weights include a multi-token prediction layer that can be used for speculative decoding. The context window goes up to 262,144 tokens. The card reports evaluation results for the original BF16 checkpoint, not for these GGUF files, so quantized quality may differ.

Pick a GGUF size for your memory
Load it in llama.cpp
Choose Think or Instruct mode
Ask in Portuguese or English

Good to know:

• Evaluation: the card's results are for the original BF16 weights, not these GGUF files
• Other formats: vLLM and SGLang users need the original BF16 or FP8 repositories

Title: Size of each S5-nano version in GB
GGUF Q4_K_M: 2.8 GB
GGUF Q8_0: 4.6 GB
FP8: 5.8 GB
GGUF BF16: 8.7 GB
Original BF16: 9.3 GB

• License: apache-2.0
• Cost to run: no API key or paid service is named, only your own computer
• Data it reaches: downloading fetches model files, and chats run locally in your app
• Not for: vLLM or SGLang serving from this repository, since the card points those users to the original weights

On Hugging Face, 13 November 2026

• Library: gguf
• Parameters: 4.3B
• Weights: about 2.2 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 3,421
• Likes: 14
• License: apache-2.0
• Languages: pt, en
• Based on: maritaca-ai/S5-nano
• Tags: gguf, llama.cpp, conversational

Links

• Model: https://huggingface.co/maritaca-ai/S5-nano-GGUF
• Files: https://huggingface.co/maritaca-ai/S5-nano-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download maritaca-ai/S5-nano-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.