Qwen3.5-2B-MTP-GGUF

Unsloth Dynamic 2.0 achieves superior accuracy & outperforms other leading quants. NEW: MTP speculative decoding for ~1.5-2x faster generation - build llama.cpp from the MTP PR branch - has been merged 16th May 2026! > **Note:** llama.cpp renamed `--spec-type mtp` to `--spec-type draft-mtp` on 2026-05-13

번역 준비 중입니다. 원문을 표시합니다.