Gallito v3 (gallito-v3-llama-1b) — offline vocational-coaching SLM (RESEARCH PREVIEW)
Small language model for "Eligiendo mi Camino", an offline vocational-guidance coach ("Gallito") for
Peruvian 5th-year secondary students. Distilled (journey-consistent SFT) from a Claude Haiku teacher on
de-identified real student journeys. Base: Llama-3.2-1B-Instruct. Runs the 8-step coaching journey
(self-knowledge, interests, career research, family-conversation prep, actionable plan).
Intended use
Research / offline educational coaching under adult supervision. Spanish (Peru).
⚠️ Limitations & safety (read before any use with minors)
- This is a research artifact, NOT a substitute for professional vocational or psychological counseling.
- Small distilled models can fabricate (invent a career/figure/institution/URL) and vary in quality; a full experience evaluation (faithfulness, usefulness, safety-in-context, etc.) accompanies the paper.
- Do not deploy to minors without human/psychologist review and the app's crisis-detection guardrail (distress → refer to a trusted adult / Peru's Línea 113). It is not a crisis service.
- Evaluated as part of a study comparing the teacher (cloud) vs distilled variants; see the paper.
Try it
Load with 🤗 Transformers (AutoModelForCausalLM.from_pretrained("ezequielmolina/gallito-v3-llama-1b")). The offline app
(transformers.js / WebLLM) is at the project repo.
- Downloads last month
- 32
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support