jan-ai/Pandora-10.7B-v1 Overview
jan-ai/Pandora-10.7B-v1 is a 10.7 billion parameter language model developed by Jan, utilizing a unique 'passthrough' merge method. This model combines specific layers from two high-performing 7B models identified on the OpenLLM Leaderboard:
- viethq188/LeoScorpius-7B-Chat-DPO (layers 0-24)
- GreenNode/GreenNodeLM-7B-v1olet (layers 8-32)
This merging strategy aims to leverage the strengths of its constituent models. The model supports a 4096 token context length and is designed for general language understanding and generation tasks.
Key Capabilities & Features
- Merged Architecture: Combines layers from two distinct 7B models for enhanced performance.
- Offline Operation: Optimized for local execution, particularly with the Jan application, ensuring data privacy.
- OpenAI Compatible Endpoints: When run locally via Jan, it provides OpenAI-compatible API endpoints.
- ChatML Prompt Template: Uses the standard ChatML format for structured conversations.
Usage and Recommendations
This model is suitable for developers looking for a capable language model that can be run entirely offline. It is particularly recommended for use with the Jan application on Mac, Windows, or Linux, which facilitates local deployment and management. For optimal performance within Jan, the Pandora-v1-10.7B-GGUF version is advised.