Model Card : gdpr_gemma-2-2b

This model is a fine-tuned version of google/gemma-2b-it on GDPR compliance data using Direct Preference Optimization (DPO).

Model Details

• Developed by: cycloevan
• Model type: Causal Language Model
• Language(s): English
• License: Apache 2.0 (inherited from base model)
• Finetuned from model: google/gemma-2b-it
• GitHub Code: gdpr-gemma2

Uses

This model is designed to assist with GDPR compliance queries and provide information related to data protection regulations.

Training Details

Training Data

The model was fine-tuned on the sims2k/GDPR_QA_instruct_dataset.

Training Procedure

• Fine-tuning method: Direct Preference Optimization (DPO)
• Optimizer: AdamW (paged_adamw_32bit)
• Learning rate: 5e-6
• Batch size: 1
• Gradient accumulation steps: 3
• Number of epochs: 10
• LR scheduler: Cosine
• Warmup steps: 2
• Training regime: LoRA (Low-Rank Adaptation)

LoRA Hyperparameters

• r: 16
• lora_alpha: 32
• lora_dropout: 0.05
• Target modules: all-linear

Limitations and Bias

• The model's knowledge is limited to the training data and may not cover all aspects of GDPR or recent updates.
• The model may occasionally generate incorrect or inconsistent information.