Model Overview

CharlesLi/llama_2_sky_o1_5_full is a 7 billion parameter language model derived from the meta-llama/Llama-2-7b-chat-hf base model. It has been specifically fine-tuned on a generator dataset, indicating its primary intended use for text generation tasks. During its training, the model achieved a final validation loss of 0.5805.

Training Details

The model was trained using the following key hyperparameters:

• Learning Rate: 2e-05
• Batch Size: 4 (train), 4 (eval) with 2 gradient accumulation steps, resulting in a total train batch size of 32.
• Optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08.
• Scheduler: Cosine learning rate scheduler with a warmup ratio of 0.1.
• Epochs: 1

The training utilized a multi-GPU setup with 4 devices. The training process showed a consistent decrease in validation loss, reaching 0.5809 by the end of the training steps.

Framework Versions

This model was developed using:

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1