Model Overview

Eric111/Yarn-Mistral-7b-128k-DPO is a 7 billion parameter language model that has undergone Direct Preference Optimization (DPO) fine-tuning. It is built upon the NousResearch/Yarn-Mistral-7b-128k base model, which itself is derived from the Mistral architecture. The DPO process utilized the Intel/orca_dpo_pairs dataset, aiming to align the model's outputs with human preferences.

Key Characteristics

• Base Model: NousResearch/Yarn-Mistral-7b-128k, a Mistral-based model.
• Parameter Count: 7 billion parameters.
• Fine-tuning Method: Direct Preference Optimization (DPO).
• Training Data: Fine-tuned with the Intel/orca_dpo_pairs dataset.
• Context Length: Features an extended context window of 8192 tokens, inherited from its base model.

Potential Use Cases

Given its DPO fine-tuning and extended context, this model is potentially well-suited for:

• Conversational AI: Maintaining longer dialogue histories and understanding complex multi-turn conversations.
• Text Summarization: Processing and summarizing lengthy documents or articles.
• Content Generation: Creating coherent and contextually relevant long-form text.
• Instruction Following: Executing complex instructions that require understanding of broader context, enhanced by DPO alignment.