HuggingFaceH4/zephyr-7b-gemma-v0.1

Zephyr 7B Gemma v0.1 Overview

HuggingFaceH4/zephyr-7b-gemma-v0.1 is an 8.5 billion parameter language model, developed by HuggingFaceH4 as the third iteration in the Zephyr series. It is a fine-tuned version of Google's gemma-7b base model, specifically designed to function as a helpful assistant.

Key Capabilities & Training

• Fine-tuning: The model was initially fine-tuned on the DEITA 10K dataset, comprising synthetic dialogues generated by ChatGPT. Further alignment was achieved using Direct Preference Optimization (DPO) on the argilla/dpo-mix-7k dataset, which contains 7,000 prompts and GPT-4 ranked model completions.
• Performance: Zephyr 7B Gemma v0.1 shows improved performance on the MT-Bench benchmark (7.81) compared to its base model, google/gemma-7b-it (6.38), indicating enhanced conversational abilities. It also performs competitively across other benchmarks like AGIEval, GPT4All, TruthfulQA, and BigBench.
• Primary Language: Primarily English.

Intended Use Cases

• Chat Applications: Optimized for chat and conversational AI tasks due to its DPO fine-tuning on diverse dialogue datasets.
• Assistant Roles: Designed to act as a helpful assistant, providing informative and coherent responses.

Limitations

• The model has not undergone extensive alignment for safety with human preferences (e.g., RLHF) and may produce problematic outputs, especially when prompted to do so. Users should exercise caution regarding potential biases or unsafe content.