activeDap/Llama-3.1-8B_hh_harmful
The activeDap/Llama-3.1-8B_hh_harmful model is an 8 billion parameter Llama 3.1-based causal language model developed by activeDap, fine-tuned on the activeDap/sft-harm-data dataset. This model is specifically trained using Supervised Fine-Tuning (SFT) to address harmful content, leveraging a maximum sequence length of 512 tokens. It is designed for applications requiring a robust Llama 3.1 variant with enhanced safety characteristics.
Loading preview...
Model Overview
activeDap/Llama-3.1-8B_hh_harmful is an 8 billion parameter language model built upon the meta-llama/Llama-3.1-8B architecture. It has been specifically fine-tuned by activeDap using Supervised Fine-Tuning (SFT) on the activeDap/sft-harm-data dataset.
Key Characteristics
- Base Model: Meta Llama 3.1 (8B parameters).
- Fine-tuning: Supervised Fine-Tuning (SFT) using the Transformers and TRL libraries.
- Dataset: Trained on
activeDap/sft-harm-data, indicating a focus on safety or harmful content mitigation. - Training Configuration: Utilized a maximum sequence length of 512, a learning rate of 2e-05, and BF16 mixed precision.
- Training Efficiency: Achieved a final training loss of 1.8782 over 35 steps, with a runtime of approximately 40 seconds.
Intended Use Cases
This model is suitable for applications where a Llama 3.1-8B variant with specific training on harmful content data is beneficial. Developers can integrate it using the Hugging Face transformers library for causal language modeling tasks, particularly those requiring a model that has been exposed to and fine-tuned on data related to harmful content.