AAC Pictogram Retriever (pt-BR) — GGUF

GGUF quantizations of Papuguinho/aac-pictogram-retriever-ptbr for use with llama.cpp.

A compact Brazilian-Portuguese embedding model that finds the right ARASAAC pictogram for a short search term — built for Augmentative and Alternative Communication (AAC). It was designed for pictogram search in Papuguinho, a tool for building communication boards (integration in progress).

• ~157M parameters, runs comfortably on CPU (no GPU).
• One model, two jobs: retrieval (find the best pictogram among ~14k) and reranking (order a shortlist).
• ~190 MB RAM at Q8_0 — fits easily in any modern machine.
• Competitive with much larger cloud embedding APIs on this domain, while being open and self-hostable.

Model family & resources

---

What it's for

In AAC, a child or therapist searches for a word (e.g. "escovar os dentes", "quero água") and the app must surface the pictogram that best represents it, so the child can point to it to communicate. This is a short-query, closed-domain, Brazilian-Portuguese retrieval problem where general-purpose embedders and leaderboard rankings do not transfer well. This model is tuned specifically for it.

Intended use: semantic search and reranking of ARASAAC pictograms from pt-BR words/phrases, in AAC applications.

Not intended for: general-purpose sentence embeddings, languages other than Brazilian Portuguese, or domains outside pictogram/AAC search.

---

Quick start with llama.cpp

The GGUF serves the full pipeline (pooling + dense projections). Load in ~190 MB at Q8:

# Download Q8_0 (recommended — near-lossless vs. fp16)
# Keep tokenizer.model next to the .gguf file
llama-server -m aac-Q8_0.gguf --embedding --pooling mean --host 127.0.0.1 --port 8080

Then POST to /v1/embeddings:

curl http://127.0.0.1:8080/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"input": "task: search result | query: escovar os dentes"}'

Always use the asymmetric prompts:

• Query: task: search result | query: <text>
• Document: title: none | text: <pictogram text>

Then normalize embeddings and take cosine similarity.

Quantization quality

Cosine similarity vs. the fp16 reference:

---

Evaluation

Evaluated on the aac-pictogram-retrieval-ptbr benchmark — a held-out set of genuinely unseen, non-canonical pt-BR queries against the full ~14k-pictogram catalog.

Best R@1 of all 12 systems (including cloud APIs). Only gemini-embedding-001 edges it by ~1 pt of R@5. Fully self-hostable, open source, no GPU required.

---

Limitations & responsible use

• Domain-specific. Tuned for ARASAAC pictogram search in Brazilian Portuguese. Not a general-purpose embedder.
• LLM-distilled supervision and evaluation. Training labels and benchmark use LLM-judged relevance, not human gold annotations. Validate against human judgments for clinical or high-stakes deployments.
• AAC is assistive technology used by vulnerable users (often non-verbal children). Keep a human in the loop: present pictograms for a person to confirm, and prefer omitting an item over showing a wrong one.

License & attribution

• Released under the Gemma license, inherited from the EmbeddingGemma base.
• ARASAAC pictograms are authored by the Government of Aragón under CC BY-NC-SA. If distributing pictogram content, follow ARASAAC's attribution and non-commercial terms.

Citation

Developed for Papuguinho (integration in progress) — www.papuguinho.com.

@software{aac_pictogram_retriever_ptbr,
  title     = {AAC Pictogram Retriever (pt-BR)},
  author    = {Stekel, Tardelli R. C. and Antonelli, Manuela and Baroni, Gabriel P. and {Carvalho Neto}, José A. and Souza, Marcela J. V.},
  year      = {2026},
  url       = {https://huggingface.co/Papuguinho/aac-pictogram-retriever-ptbr-GGUF},
  doi       = {10.57967/hf/9658},
  publisher = {Hugging Face},
  note      = {GGUF quantizations of the CPU-friendly Brazilian-Portuguese embedding model for ARASAAC pictogram retrieval and reranking, built for Papuguinho (www.papuguinho.com)}
}