Hugging Face Overview
Overview
Einstein-v7-Qwen2-7B is a 7.6 billion parameter language model developed by Weyaxi, built upon the Qwen/Qwen2-7B architecture. It has been extensively fine-tuned over two epochs using 8xMI300X compute resources and the axolotl framework, leveraging a wide array of diverse datasets. The model is designed to operate with a ChatML prompt template, facilitating structured conversational interactions.
Key Capabilities
- Extensive Context Handling: Features a 131,072 token context length, enabling processing and generation of long and complex texts.
- Diverse Training: Fine-tuned on a broad spectrum of datasets including Airoboros, Wild Chat, Capybara, ShareGPT, and Alpaca variants, enhancing its general conversational abilities.
- ChatML Compatibility: Optimized for the ChatML prompt template, ensuring consistent and effective communication in chat-based applications.
- Quantized Versions Available: Community-contributed GGUF and ExLlamaV2 quantized versions are available for efficient deployment on various hardware.
Performance Insights
Evaluations on the Open LLM Leaderboard v2 show an average score of 24.01. Specific metrics include 41.00 for IFEval (0-Shot), 32.84 for BBH (3-Shot), and 34.40 for MMLU-PRO (5-shot), indicating a balanced performance across different reasoning and knowledge-based tasks.
Good for
- General-purpose conversational AI and chatbots.
- Applications requiring processing of long documents or complex dialogues due to its large context window.
- Developers looking for a Qwen2-7B based model with enhanced instruction-following capabilities from diverse fine-tuning.