Smith-3/simon-fcyt-umss-v2
Smith-3/simon-fcyt-umss-v2 is a 0.35 billion parameter language model fine-tuned by Smith-3 from LiquidAI/LFM2.5-350M. Optimized for academic support within the TecnoTime application for UMSS students, it generates brief, structured JSON responses for class reminders, motivational check-ins, and academic progress tracking. The model excels at producing consistent, parseable outputs for integration with Android systems, focusing on reinforcing student habits and well-being.
Loading preview...
Overview
Smith-3/simon-fcyt-umss-v2 is a 0.35 billion parameter language model developed by Smith-3, specifically fine-tuned for the Faculty of Sciences and Technology (FCyT) at the Universidad Mayor de San Simón (UMSS). It is designed to power the TecnoTime application, providing structured, brief, and consistent responses to help students manage academic activities.
Key Capabilities & Purpose
This model specializes in generating specific JSON outputs, rather than free-form text, making it highly suitable for automated system integration. Its primary academic purposes include:
- Class Reminders: Prompting students about upcoming classes and schedules.
- Motivational Messages: Delivering positive reinforcement to encourage academic continuity.
- Daily Check-ins: Facilitating short interactions to monitor daily progress and well-being.
- Positive Reinforcement: Offering encouragement at critical points in the semester.
Technical Details & Training
The model is a fine-tuning of LiquidAI/LFM2.5-350M using SFT (Supervised Fine-Tuning) and LoRA (Low-Rank Adaptation) with Unsloth. It is trained to produce outputs compatible with the SimonResponse JSON structure required by the TecnoTime Android system. The training involved 3 epochs with a learning rate of 2e-4 and a cosine scheduler, utilizing a dataset formatted with LiquidAI messages.
Structured Output
A core differentiator is its strict adherence to a predefined JSON output format, including fields like why, opening_tag, icon_tag, and conditional fields such as question and choices for interactive notifications. This ensures seamless integration and predictable behavior within the TecnoTime application.
Intended Use
This model is intended for:
- Integration into the TecnoTime application.
- Generating structured JSON responses for academic notifications.
- Providing brief motivational support to students.
- Local or low-cost inference using GGUF quantizations.
It is explicitly not intended for providing official academic information, psychological/medical/legal counseling, or automated administrative decisions.