mlx-community/RavenX-CyberAgent-Qwen3.6-35B-A3B-Opus-4.7-OpenMythos-Pentester-BugHunter-RATH-mlx-4bit-msq

Mixed-precision MLX quantization of deadbydawn101/RavenX-CyberAgent-Qwen3.6-35B-A3B-Opus-4.7-OpenMythos-Pentester-BugHunter-RATH-mlx, quantized with MLX Smart Quantize (MSQ) — my own sensitivity-based mixed-precision quantization method for Apple Silicon. It measures per-layer NMSE and assigns optimal bit widths automatically, combining architecture knowledge with measured data. — Type: Text (LLM) — Average: 4.50 bits per weight — Method: MLX Smart Quantize (MSQ) — AWQ scaling: applied to 50 groups

Модальности:
Генерация текста Компьютерное зрение

Области применения:
Генерация кода Диалог / чат

Задача: Генерация текста
Автор: mlx-community
Теги: mlx, qwen3_5_moe, security, cybersecurity, pentest, pentesting, bug-bounty, red-team
Лайков: 3  |  Загрузок: 327

Открыть на HuggingFace →

Описание основано на материалах HuggingFace. Перевод выполнен автоматически.