Model Overview

hamxea/StableBeluga-7B-activity-fine-tuned-v2 is an instruction-tuned language model developed by Stability AI. It is a 7 billion parameter variant derived from the Llama2 70B architecture, fine-tuned using an internal Orca-style dataset. This model is designed to follow instructions exceptionally well, making it suitable for a wide range of conversational and task-oriented applications.

Key Capabilities

• Instruction Following: Excels at understanding and executing user instructions, leveraging its Orca-style fine-tuning.
• Conversational AI: Capable of engaging in coherent and helpful dialogue.
• General Purpose: Suitable for various text generation tasks where clear instruction adherence is crucial.

Training Details

The model was fine-tuned using supervised learning on an Orca-style dataset, employing mixed-precision (BF16) training and optimized with AdamW. The training involved specific hyperparameters for different parts of the Orca dataset, including varying batch sizes and learning rates with cosine decay.

Usage Considerations

Stable Beluga 2 requires a specific prompt format for optimal performance:

### System:
This is a system prompt, please behave and help the user.

### User:
Your prompt here

### Assistant:
The output of Stable Beluga 2

Developers should perform safety testing tailored to their specific applications, as the model's outputs cannot be predicted in advance and may occasionally produce inaccurate or biased responses.