GGUF quants of Qwopus3.8-27B — a task-vector transplant that carries the Qwopus reasoning distillation onto the Qwen3.8-27B base. Qwopus3.8 keeps Qwen3.8’s architecture, 262,144 context length and native MTP speculative head, and layers the Opus-style distilled reasoning trait on top: materially fewer tokens for the same result, and a stronger non-thinking (instruct) mode. The MTP block is exported (nextn layout), so llama.cpp speculative decoding works out of the box. MLX/Apple-Silicon users: see the MTPLX builds (4bit-Speed · 8bit-Quality). Built by mirroring the per-tensor type map and calibration imatrix from Unsloth’s Dynamic v3 Qwen3.8-27B release onto this merge (the models share an identical architecture, so the importance data transfers). Measured against the merge’s own bf16 ground truth on a held-out code/agentic/prose corpus: top-token agreement 96.7% vs 94.2% for Q4KM, with tail divergences 2–3× tighter. On a 38-task non-thinking suite it scores 36/38 vs 35/38 for stock Qwen3.8 quantized with the exact same UD v3 recipe — the merge’s instruction-following edge survives quantization. The MTP head is retained (Q6_K), so speculative decoding works as with the other…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: nom666
Теги: gguf, llama.cpp, multi-token-prediction, speculative-decoding, qwen3_5, merge, task-arithmetic, endpoints_compatible
Лайков: 3 | Загрузок: 115
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
