Q4_K_M (Optimal balance of speed and intelligence) or Q8_0 (Higher accuracy, requires ~1GB RAM).
The model is distributed across different platforms depending on your preferred deployment ecosystem. 1. Hugging Face (Standard & Quantized Weights)
Note: Always verify the checksums and sources when downloading open-source model weights. Aurora 0.7b.2 Download
Unlike its larger counterparts (e.g., Llama 2 7B or GPT-3.5), Aurora 0.7b.2 is optimized for:
While Phi-2 is larger and more capable on complex reasoning, Aurora 0.7b.2 is 3.5x faster and uses 60% less memory, making it better for real-time edge applications. Q4_K_M (Optimal balance of speed and intelligence) or
A functional installation of (v1.2 or higher) to execute initial local file transfers.
Go to the official Hugging Face repository for Aurora [1]. Llama 2 7B or GPT-3.5)
After installing Aurora 0.7b.2: