pulse-familiar-fenn-lora

Pulse Familiar Fenn LoRA

A small PEFT/LoRA adapter for nvidia/Nemotron-Mini-4B-Instruct, trained to make Fenn — the Pulse Familiar ASCII creature — speak in short, warm, non-clinical heartbeat-reactive lines.

This adapter is used by the Build Small Hackathon Space: build-small-hackathon/pulse-familiar.

Intended use

The adapter is not a medical model. It is a character voice layer. The app feeds it only coarse qualitative cues such as “pulse is fast and tight” or “heartbeat is calm and even”; it does not need raw biometric records.

Training data

The SFT set was generated from synthetic mood/cue prompts and curated creature-style responses. No raw health export, dates, or personal biometric records are included in this adapter.

Base model

• Base: nvidia/Nemotron-Mini-4B-Instruct
• Adapter type: LoRA / PEFT
• Target app: Pulse Familiar

Loading

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = "nvidia/Nemotron-Mini-4B-Instruct"
adapter = "build-small-hackathon/pulse-familiar-fenn-lora"

tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, adapter)

Limitations

• Style adapter only; it does not diagnose, infer illness, or provide medical advice.
• Best outputs come from short, qualitative physiological cues and the Fenn system prompt.
• The base model license is NVIDIA's open/community model license; comply with the base model terms.