barozp/Qwen3.8-27B-MTP-GGUF overview
Qwen3.8 27B MTP GGUF NOTE Placeholder — not yet available. This repo is reserved ahead of the Qwen/Qwen3.8 27B release and will be filled in once that model sh…
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| Browse files on Hugging Face | ||||
Model Details
Model README
---
license: apache-2.0
tags:
- gguf
- llama.cpp
- qwen
---
Qwen3.8-27B-MTP-GGUF
> [!NOTE]
> Placeholder — not yet available. This repo is reserved ahead of the
> Qwen/Qwen3.8-27B release and will be filled in once that model ships.
> Watch this repo or check back after the Qwen3.8 announcement.
GGUF of Qwen/Qwen3.8-27B with its Multi-Token Prediction (MTP) head preserved for self-speculative decoding in llama.cpp. Dense architecture (not MoE), so this is a straightforward tensor extraction, no expert remapping involved.
Plain + MTP — base model with self-speculative decoding enabled.
Related releases in this line
- Qwen3.8-27B-GGUF
- Qwen3.8-27B-MTP-GGUF (this repo)
- Qwen3.8-27B-Opus-Distill-GGUF
- Qwen3.8-27B-Opus-Distill-MTP-GGUF
Prior work from this account
This follows the same recipe (REAP-style efficiency work, Opus reasoning
distillation, MTP grafting, full GGUF quant ladders with measured
speed/quality benchmarks) used for the Qwen3.6 line:
Run barozp/Qwen3.8-27B-MTP-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models