Model Overview

activeDap/gemma-2b_hh_helpful is a 2.5 billion parameter language model derived from Google's Gemma-2b. It has undergone Supervised Fine-Tuning (SFT) using the activeDap/sft-hh-data dataset, which focuses on helpfulness. The fine-tuning process involved 20 steps, achieving a final training loss of 2.1454 over 9.15 seconds.

Key Capabilities

• Instruction Following: Fine-tuned to generate responses that align with user instructions, particularly in a helpful assistant style.
• Conversational AI: Optimized for prompt-completion tasks, making it suitable for dialogue systems.
• Efficient Inference: As a 2.5B parameter model, it offers a balance between performance and computational efficiency.

Training Details

The model was trained for 1 epoch with a per-device batch size of 16 (total batch size 64 across 4 GPUs) and a learning rate of 2e-05. It utilized a cosine learning rate scheduler, BF16 mixed precision, and a maximum sequence length of 512 tokens. The training focused on assistant-only loss, ensuring the model learns to generate relevant and helpful outputs.

Good For

• Developing helpful AI assistants.
• Instruction-based text generation.
• Applications requiring a smaller, yet capable, fine-tuned language model for conversational tasks.