ahmedandaloes/security-llama3.2-3b-MLX-bf16

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 11, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The ahmedandaloes/security-llama3.2-3b-MLX-bf16 is a 3.2 billion parameter Llama 3.2 model, converted to full-precision (bf16) MLX format for efficient inference on Apple Silicon. This model is a direct conversion of the viettelsecurity-ai/security-llama3.2-3b, retaining its original weights and security-focused capabilities. It is specifically optimized for fast local execution of security-related tasks on Apple hardware, offering unquantized performance.

Loading preview...

Overview

This model, ahmedandaloes/security-llama3.2-3b-MLX-bf16, is an MLX-optimized version of the viettelsecurity-ai/security-llama3.2-3b Llama 3.2 model. It features 3.2 billion parameters and is converted to full-precision (bf16) for use with Apple Silicon, ensuring identical weights to the source model without quantization. The conversion was performed using mlx-lm, making it suitable for fast local inference.

Key Capabilities

  • Full-Precision Performance: Operates with bf16 precision, offering unquantized performance identical to the original source model.
  • Apple Silicon Optimization: Specifically designed for efficient and fast local inference on Apple Silicon hardware using the MLX framework.
  • Security-Focused: Inherits the security-oriented capabilities of its base model, making it suitable for specialized security tasks.
  • Easy Integration: Can be easily loaded and used with the mlx-lm library for generating responses.

Good For

  • Developers and researchers working on security-related AI applications on Apple Silicon.
  • Local inference of security-focused language tasks where full precision is desired.
  • Experimenting with the security-llama3.2-3b model on Apple hardware without performance degradation from quantization.