MLX conversion of OBLITERATUS/Qwen3.6-27B-OBLITERATED for Apple Silicon. The model uses an alternating linear/full attention pattern across 64 layers (mamba-style linear attention interleaved with standard multi-head attention), giving it efficient long-context performance on Apple Silicon. — mlxlm emits a harmless warning about unrecognized ropeparameters keys (mropeinterleaved, mropesection) — this is a known Qwen quirk; generation is unaffected — Converted and tested on Apple M5 Pro, 48 GB unified memory — Metal’s effective GPU working set ceiling is ~37.4 GB regardless of total RAM; model weights occupy ~14.4 GB of that — Practical context on 48 GB: comfortable sustained use up to ~75K tokens; hard ceiling observed at ~91K before prefill guard trips — Source
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: mlx-community
Теги: mlx, qwen3_5, qwen3, 4bit, mlx-lm, conversational, en, 4-bit
Лайков: 3 | Загрузок: 171
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
