Laguna-S-2.1-8bit-Att-6bit-Ex
This repository contains an MLX conversion of poolside/Laguna-S-2.1.
Conversion Details
- Original model:
poolside/Laguna-S-2.1 - Model family: Laguna S 2.1
- Model size: 118B total parameters, approximately 8B activated parameters per token
- Conversion: MLX-LM conversion using the Laguna architecture implementation
- Quantization: 8-bit affine attention/base weights and 6-bit affine routed experts
- Quantization policy: 8-bit attention/base weights with 6-bit routed experts
- Group size: 64
- Published MLX package size: 89.93 GiB
Benchmarks
No comparative benchmarks are available yet. This repository does not currently provide quality, speed, memory, or benchmark comparisons against the original weights or other quantizations.
License
This is a converted and quantized derivative of the original checkpoint. It retains the upstream OpenMDW-1.1 license and applicable notices of origin.