Nishef/SmolLM2-360M-Full_DPO_20251225_043457

SmolLM2-360M - Dpo

Model Description

This model is a LoRA Adapter fine-tuned from HuggingFaceTB/SmolLM2-360M using the Dpo training method.

Direct Preference Optimization - Paired preference optimization without reference model

This model was developed as part of thesis research on LLM Alignment using Preference Optimization Methods.

Model Details

Property	Value
Base Model	HuggingFaceTB/SmolLM2-360M
Training Method	Dpo
Model Type	LoRA Adapter
Training Date	December 2025
Framework	PyTorch + Transformers + PEFT

Benchmark Results

Benchmark	Score
HellaSwag (10-shot)	0.550
TruthfulQA (0-shot MC2)	0.361
MMLU-Mini (5-shot)	0.264

Comparative Analysis

The following chart compares this method against other training approaches on the same base model:

Training Configuration

Parameter	Value
Epochs	1
Batch Size	2
Gradient Accumulation	8
Effective Batch Size	16
Learning Rate	2e-4
Max Sequence Length	512
LoRA Rank	16
LoRA Alpha	32
Dataset	Combined Preference Dataset (HH-RLHF + SHP + OpenAssistant)

Combined Preference Dataset (kto_combined)

Training uses a Combined Preference Dataset built via Round-Robin Sampling from three sources:

Source	Total Samples	Interactions
Anthropic HH-RLHF	321,600	61,568
Stanford Human Preferences (SHP)	697,436	38,984
OpenAssistant Conversations v1	16,810	8,904
Total	1,035,846	109,456

Actual Training Statistics (subset split train_prefs[:32090]):

• Training samples: 13,300 (paired examples)
• Validation samples: 700 (5%)
• Round-Robin distribution: 1,130 interactions per source
• Seed: 42 (for reproducibility)

Usage

Loading as LoRA Adapter

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM2-360M")
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM2-360M")

# Load adapter
model = PeftModel.from_pretrained(base_model, "Nishef/SmolLM2-360M-Full_DPO_20251225_043457")

# Generate text
inputs = tokenizer("Hello, how are you?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0]))

Training Methodology

Dpo

Direct Preference Optimization - Paired preference optimization without reference model

Key Features:

• Paired preference optimization
• Direct policy optimization without reward model
• Efficient single-stage training
• Bradley-Terry preference modeling

Citation

If you use this model in your research, please cite:

@misc{smollm2_360m_dpo_2025,
  title = {SmolLM2-360M Fine-tuned with Dpo},
  author = {Thesis Research},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/Nishef/SmolLM2-360M-Full_DPO_20251225_043457}
}

Repository Structure

.
├── adapter_config.json      # LoRA configuration
├── adapter_model.safetensors # Model weights
├── tokenizer files          # Tokenizer configuration
├── eval_summary.csv         # Evaluation results
├── thesis_plots/            # Visualization assets
│   ├── benchmark_results.png
│   └── training_loss.png
└── README.md               # This file

Acknowledgments

• Base Model: HuggingFaceTB/SmolLM2-360M
• Training Framework: Hugging Face Transformers
• Fine-tuning Library: PEFT

License

This model is released under the Apache 2.0 license.

---

This model was created as part of thesis research on LLM alignment using preference optimization methods.