Instructions to use tardellirs/aac-board-generator-770m-ptbr-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="tardellirs/aac-board-generator-770m-ptbr-GGUF",
	filename="aac-board-generator-770m-ptbr-Q4_K_M.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Use Docker

docker model run hf.co/tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tardellirs/aac-board-generator-770m-ptbr-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tardellirs/aac-board-generator-770m-ptbr-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Ollama
How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with Ollama:
```
ollama run hf.co/tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
```

Unsloth Studio

How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for tardellirs/aac-board-generator-770m-ptbr-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for tardellirs/aac-board-generator-770m-ptbr-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for tardellirs/aac-board-generator-770m-ptbr-GGUF to start chatting

Atomic Chat new
Docker Model Runner
How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with Docker Model Runner:
```
docker model run hf.co/tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M
```

Lemonade

How to use tardellirs/aac-board-generator-770m-ptbr-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull tardellirs/aac-board-generator-770m-ptbr-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.aac-board-generator-770m-ptbr-GGUF-Q4_K_M

List all available models

lemonade list

aac-board-generator-770m-ptbr-GGUF / README.md

tardellirs

Upload folder using huggingface_hub

ae08f1f verified 4 days ago

preview code

Raw

History Blame Contribute Delete

3.68 kB

metadata

language:
  - pt
license: gemma
library_name: gguf
pipeline_tag: text-generation
base_model: tardellirs/aac-board-generator-770m-ptbr
tags:
  - gguf
  - llama-cpp
  - text-generation
  - gemma3
  - aac
  - augmentative-alternative-communication
  - arasaac
  - board-generation
  - brazilian-portuguese
  - cpu

AAC Board Generator 770M (pt-BR) — GGUF

llama.cpp GGUF builds of aac-board-generator-770m-ptbr — a 771.6M Brazilian-Portuguese multi-function AAC model that drafts pictogram-board word lists and follows in-assistant board-editing instructions (add / remove / generate N), on CPU. See the model card for what it does, training, and evaluation.

Files

File	Quant	Size	Notes
`aac-board-generator-770m-ptbr-Q4_K_M.gguf`	Q4_K_M	561 MB	smallest; slight quality drop
`aac-board-generator-770m-ptbr-Q5_K_M.gguf`	Q5_K_M	606 MB	good size/quality balance
`aac-board-generator-770m-ptbr-Q6_K.gguf`	Q6_K	766 MB	near-lossless
`aac-board-generator-770m-ptbr-Q8_0.gguf`	Q8_0	824 MB	recommended — safe near-lossless
`aac-board-generator-770m-ptbr-f16.gguf`	f16	1.55 GB	full precision reference

At Q8_0 the model needs on the order of ~0.9–1.1 GB RAM to serve (weights + a small context) and produces a ~12-item board in about 4 s on 4 CPU threads.

Usage

llama-server -m aac-board-generator-770m-ptbr-Q8_0.gguf -t 4 -c 2048 --host 127.0.0.1 --port 8080

Gemma has no system role — put the instruction in the user turn and call /completion:

import requests
INSTR = ("Você monta pranchas de CAA (pictogramas, pt-BR). Para o PEDIDO, liste ~12 itens concretos e relevantes, "
         "um por linha, no formato palavra|tipo|sinônimos (tipo: v/s/a/e/l/p). Só a lista.")
pedido = "monta uma prancha de brincar no parquinho"
prompt = f"<start_of_turn>user\n{INSTR}\n\nPEDIDO: {pedido}<end_of_turn>\n<start_of_turn>model\n"
print(requests.post("http://127.0.0.1:8080/completion",
                    json={"prompt": prompt, "temperature": 0, "n_predict": 320}).json()["content"])

Generation is greedy / deterministic (temperature 0). Output is one item per line, word|type|synonyms.

Notes on the build (why these GGUFs are faithful)

This model is a vocabulary-trimmed Gemma 3 (262k → 64k tokens). A reconstructed SentencePiece tokenizer is not bit-exact for a trimmed vocabulary and would silently degrade greedy generation. These GGUFs instead use the BPE tokenizer.json path, which tokenizes identically to the original — so the quantized model reproduces the untrimmed parent's outputs. Verified: identical token counts and identical boards vs. the fp16 reference.

License & attribution

Released under the Gemma license (inherited from Gemma 3). Built for the ARASAAC ecosystem (CC BY-NC-SA content). See the model card for full terms.

Developed for and used in Papuguinho — www.papuguinho.com.