dgomes03/Ministral-8B-Instruct-mixed-4-6-bit

Ministral-8B-Instruct-2410 quantized with mixed precision: This is a Ministral-8B-Instruct model where the embedding layer and output (head) layer are quantized to 6-bit precision, while the rest of the model uses 4-bit quantization. This mixed-precision approach aims to balance model size and inference speed with improved precision in critical layers.