onebee-gf-dpo-v1-4epoch

DPO on the v0 preference dataset run for 4 epochs — a deliberate overfitting experiment, not a recommended checkpoint.

Model Overview

A deliberate overfitting experiment: the same 200-pair v0 preference dataset as dpo-v0, but trained for 4 epochs instead of 1, to study how DPO degrades when over-trained on a small preference set. Published for reproducibility of that specific experiment, not as a general-purpose checkpoint.

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

Property	Details
Model	`onebee-gf-dpo-v1-4epoch`
Parameters	~2B effective (base) + LoRA rank 16 adapter
Architecture	Gemma4 (multimodal, text + vision)
Base Model	`google/gemma-4-E2B-it`
Language	English
Context Length	131,072 tokens (inherited from base model)
Training Method	LoRA DPO, 4 epochs (vs. 1 epoch for dpo-v0), same 200-pair v0 dataset
License	Apache-2.0 (inherited from base model)

Intended Use

Studying DPO overfitting behavior on small preference datasets. Not recommended for deployment or as a training base.

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

Same base capabilities as dpo-v0, with observed overfitting artifacts from extended training

Quick Start

Installation

pip install transformers torch

Usage

from transformers import AutoModelForCausalLM, AutoProcessor

model = AutoModelForCausalLM.from_pretrained("arjhinety/onebee-gf-dpo-v1-4epoch")
processor = AutoProcessor.from_pretrained("arjhinety/onebee-gf-dpo-v1-4epoch")

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.

See docs/dpo_results.md for the overfitting-experiment comparison against dpo-v0.

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

Limitations

Explicitly overfit by design — do not use this checkpoint as a general-purpose companion model.

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

Repo	Description
onebee-gf-sft-v0	Day 4 v0 SFT (202 examples)
onebee-gf-sft-v1	Proper-scale SFT (2232 examples)
onebee-gf-dpo-v0	Week 2 DPO v0 (200 pairs)
onebee-gf-dpo-v1-4epoch	DPO overfitting experiment
onebee-gf-dpo-v1-scale	Proper-scale DPO, pre-distillation
onebee-gf-distill-v1	SFT+DPO+distillation — current best overall
onebee-gf-dpo-v1-scale-gguf	GGUF quantizations

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).

Loading…