NilayR/llama32-iterative-dpo-iter1

Iterative DPO Fine-Tune of Llama-3.2-1B (Iteration 1)

This repository contains the LoRA adapters from the first iteration of a Direct Preference Optimization (DPO) fine-tuning process on the meta-llama/Llama-3.2-1B-Instruct model.

This work is part of a project exploring iterative DPO, where the model refines itself over multiple cycles of preference data generation and training, inspired by the "Self-Rewarding Language Models" paper.

• Repository for Iteration 2: NilayR/llama32-iterative-dpo-iter2

Model Details

Model Description

This model is a fine-tuned version of meta-llama/Llama-3.2-1B-Instruct. It was trained using DPO on a preference dataset that the base model generated itself. An LLM Judge, powered by GPT-3.5-Turbo, evaluated pairs of model-generated responses to create the chosen/rejected pairs for training.

The goal of this iteration was to establish the first step in a self-improvement loop, aligning the model more closely with human-like preferences for accuracy, helpfulness, and clarity.

• Developed by: NilayR
• Model type: Causal Language Model
• Language(s): English
• License: apache-2.0
• Finetuned from model: meta-llama/Llama-3.2-1B-Instruct

How to Get Started with the Model

To use these LoRA adapters, load the base model (meta-llama/Llama-3.2-1B-Instruct) and then apply the adapters from this repository.

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

# Set base model ID and adapter path
base_model_id = "meta-llama/Llama-3.2-1B-Instruct"
adapter_id = "NilayR/llama32-iterative-dpo-iter1"

# Configure BitsAndBytes for 4-bit quantization
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

# Load the base model with quantization
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
)

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
tokenizer.pad_token = tokenizer.eos_token

# Load and apply the PEFT adapters
model = PeftModel.from_pretrained(base_model, adapter_id)

# --- Generate a response ---
prompt = "What are the key benefits of meditation?"
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": prompt}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=200,
    do_sample=True,
    temperature=0.7,
    top_p=0.95
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response.split("assistant")[-1].strip())

Training Details

Training Data

The model was trained on a preference dataset generated by the meta-llama/Llama-3.2-1B-Instruct model itself.

• Data Generation Process:
  1. Instructions: 20 instructions were selected from the LIMA dataset.
  2. Response Generation: The base model generated multiple diverse responses for each instruction.
  3. Preference Labeling: A custom LLM Judge powered by GPT-3.5-Turbo was used to compare pairs of the generated responses, creating a dataset of 56 chosen/rejected pairs.

Training Procedure

The model was trained for one epoch using the TRL library's DPOTrainer.

Training Hyperparameters

• Framework: trl.DPOTrainer
• Epochs: 1
• Batch Size: 1
• Gradient Accumulation Steps: 2
• Optimizer: paged_adamw_8bit
• Learning Rate: 2e-5
• DPO Beta (β): 0.1
• Max Steps: 50
• Final Training Loss: 0.6405

LoRA Configuration

• Rank (r): 16
• Alpha (lora_alpha): 32
• Target Modules: q_proj, k_proj, v_proj, o_proj
• Dropout: 0.05

Compute Infrastructure

• Hardware: 1x NVIDIA A100 40GB GPU
• Cloud Provider: Google Colab
• Software: transformers, peft, trl, bitsandbytes