gromo-llama2-faq-model

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf using LoRA (Low-Rank Adaptation) on a custom dataset (gromo-faq-v1-0.csv). It achieves the following results on the evaluation set:

• Loss: 1.2682
• Perplexity: 3.55

Model description

This is a fine-tuned version of meta-llama/Llama-2-7b-hf using LoRA (Low-Rank Adaptation) on a custom dataset (gromo-faq-v1-0.csv). It is optimized for text generation tasks, particularly for FAQ-style responses, and achieves a loss of 1.2682 and a perplexity of 3.55 on the evaluation set.

Intended uses & limitations

• Intended uses: This model is intended for generating text based on FAQs or similar structured data. It is suitable for research, prototyping, or educational purposes.
• Limitations: The model may produce inaccurate or biased outputs due to the limited dataset it was trained on. It should not be used for production without further evaluation and fine-tuning.

Training and evaluation data

The model was trained on a custom dataset, gromo-faq-v1-0.csv, containing FAQ-style text data. Evaluation was performed on a separate validation subset of this dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Use paged_adamw_8bit with betas=(0.9, 0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
3.2528	1.0	172	1.5046
3.1911	2.0	344	1.4763
2.808	2.9854	513	1.4937

Note: The final evaluation after additional testing yielded a Validation Loss of 1.2682 and Perplexity of 3.55, reflecting improved performance.

Framework versions

• PEFT 0.14.0
• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0