lodrick-the-lafted/Hermes-Instruct-7B-100K

Hermes-Instruct-7B-100K Overview

Hermes-Instruct-7B-100K is a 7 billion parameter instruction-tuned model developed by lodrick-the-lafted. It is built upon the robust Mistral-7B-Instruct-v0.2 architecture and has been further fine-tuned using 100,000 rows from the teknium/openhermes dataset, formatted in the Alpaca style. This training approach aims to enhance its ability to understand and follow instructions effectively.

Key Capabilities and Features

• Instruction Following: Optimized for responding to a wide range of user instructions, leveraging its fine-tuning on the OpenHermes dataset.
• Flexible Prompting: Supports both the default Mistral-Instruct prompt format (<s>[INST] {sys_prompt} {instruction} [/INST]) and the Alpaca format ({sys_prompt}\n\n### Instruction:\n{instruction}\n\n### Response:\n). The tokenizer defaults to Alpaca.
• Performance: Achieves an average score of 64.96 on the Open LLM Leaderboard, with notable scores including 82.84 on HellaSwag (10-Shot) and 60.95 on MMLU (5-Shot). Detailed evaluation results are available on the Open LLM Leaderboard.

Ideal Use Cases

• General Conversational AI: Suitable for chatbots and virtual assistants requiring strong instruction adherence.
• Text Generation: Effective for generating creative and coherent text based on specific prompts.
• Research and Development: A solid base model for further fine-tuning on domain-specific instruction datasets.