onebee-gf-sft-v1

onebee-gf-sft-v1

Proper-scale LoRA SFT checkpoint on gemma-4-E2B-it (40 personas, 2232 examples) — current best SFT-only checkpoint.

Model Overview

Proper-scale LoRA SFT checkpoint — 10x the data of sft-v0 (2232 train examples, 40 personas, memory-aware conversational format), 2 epochs. Best SFT-only checkpoint in this project; the DPO and distillation checkpoints chain off this one.

GGUF quantizations available: this project's current-best checkpoint is also published as [quantizations (F16 reference plus 12 quant levels)](https://huggingface.co/arjhinety/onebee-gf-dpo-v1-scale-gguf) for llama.cpp-based on-device inference (quantized from dpo-v1-scale, not this checkpoint).

Model Details

Intended Use

Intended Use

As a base for further post-training (DPO/distillation), or for studying the isolated effect of SFT before preference optimization is applied.

Out-of-Scope Use

Not evaluated or intended for: safety-critical decisions, medical/legal/financial advice, or any deployment where a wrong or overconfident answer causes real harm. This is a research artifact from an open-source project studying post-training and memory architecture on small models — see the project README for the full research framing before using it in any production context.

Capabilities

• Companion-persona conversational responses conditioned on retrieved memories
• Improved abstention calibration over sft-v0 after a documented bug-fix cycle

Quick Start

Installation

pip install transformers torch

Usage

from transformers import AutoModelForCausalLM, AutoProcessor

model = AutoModelForCausalLM.from_pretrained("arjhinety/onebee-gf-sft-v1")
processor = AutoProcessor.from_pretrained("arjhinety/onebee-gf-sft-v1")

messages = [
    {"role": "system", "content": "You are a warm AI companion who remembers this user."},
    {"role": "user", "content": "What conference did I say I was attending?"},
]
inputs = processor.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=128)
print(processor.decode(output[0], skip_special_tokens=True))

Evaluation

Scored against PMB (Personalized Memory Benchmark), 688 adversarial probes across 8 categories, with an LLM judge under dual-order (position-bias-controlled) scoring plus a rule-based abstention detector.

Full methodology, all numbers, and honest limitations: docs/proper_scale_results.md.

Limitations

SFT alone, no preference optimization — use onebee-gf-dpo-v1-scale or onebee-gf-distill-v1 for the strongest results. Single seed/run.

This project reports negative/inconclusive results as honestly as positive ones — read the linked docs before assuming any number here is a clean win.

Other Checkpoints From This Project

Citation

@software{small_mind_companion,
  title  = {small-mind-companion: Post-training and cognitive architecture for a small multimodal companion LLM},
  author = {arjhinety},
  year   = {2026},
  url    = {https://github.com/arjhinety/small-mind-companion}
}

License

Apache-2.0, inherited from the base model (google/gemma-4-E2B-it).