What license applies to tooltd/Qwen3.6-27B-MTP-TENSORS-ONLY-GGUF?

License: apache-2.0. Verify terms on Hugging Face before commercial use.

How do I run tooltd/Qwen3.6-27B-MTP-TENSORS-ONLY-GGUF locally?

Download a GGUF file from this page and load it in guIDE or llama.cpp. Pipeline task: text-generation.

Model Intelligence Sheet

tooltd/Qwen3.6-27B-MTP-TENSORS-ONLY-GGUF overview

Qwen3.6 27B MTP Tensor GGUF Blocks This repository contains standalone Multi Token Prediction MTP tensor blocks extracted from Unsloth's Qwen3.6 27B MTP GGUFs …

ggufqwen3.6mtpspeculative-decodingmtp only27blicense:apache-2.0region:us

Runs locally from ~199.2 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).

Downloads

113

Likes

Pipeline

—

Author

tooltd

Repository Files & Downloads

3 GGUF files detected

Direct downloads for local inference

File	Type	Quantization	Size	Link
Qwen3.6-27B_MTP-TENSORS-ONLY-IQ4_XS.gguf	GGUF	IQ4_XS	253.5 MB	Download
Qwen3.6-27B_MTP-TENSORS-ONLY-UD-IQ3_XXS.gguf	GGUF	IQ3_XXS	199.2 MB	Download
Qwen3.6-27B_MTP-TENSORS-ONLY-UD-Q3_K_XL.gguf	GGUF	Q3_K_XL	222.3 MB	Download

Model Details

Model ID	tooltd/Qwen3.6-27B-MTP-TENSORS-ONLY-GGUF
Author	tooltd
Pipeline	—
License	apache-2.0
Base model	—
Last modified	2026-06-27T13:09:03.000Z

Model README

---

license: apache-2.0

tags:

qwen3.6
mtp
speculative-decoding
mtp only
gguf
27b

---

Qwen3.6-27B MTP Tensor GGUF Blocks

This repository contains standalone Multi-Token Prediction (MTP) tensor blocks extracted from Unsloth's Qwen3.6 27B MTP GGUFs.

These blocks are used to quickly integrate MTP capabilities into Qwen3.6 27B GGUFs that do not natively include them.

Here I've extracted small versions of only 200 MB for those who need to save VRAM.

After inserting them into model, results worked well, speed increased by approximately 2x.

How to Use

You can merge these MTP blocks into your target GGUF file using the specialized injection script available on GitHub Gist:

🔗 Script: Gist: Qwen MTP GGUF Injector

python convert.py Qwen3.6-27B-ABCXYZ-IQ4_XS.gguf Qwen3.6-27B_MTP-TENSORS-ONLY-IQ4_XS.gguf Qwen3.6-27B-ABCXYZ-IQ4_XS-MTP.gguf

Tensor Maps

UD_IQ3_XXS

  NAME                                  SHAPE          TYPE        SIZE
  ------------------------------------  -------------  ----------  ----------
  blk.64.attn_k.weight                  [5120, 1024]   Q4_K_S      1.7 MB
  blk.64.attn_k_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_norm.weight               [5120]         F32         20.5 KB
  blk.64.attn_output.weight             [6144, 5120]   Q4_K_M      13.5 MB
  blk.64.attn_q.weight                  [5120, 12288]  Q4_K_S      20.6 MB
  blk.64.attn_q_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_v.weight                  [5120, 1024]   Q4_K_M      2.3 MB
  blk.64.ffn_down.weight                [17408, 5120]  Q4_K_M      38.3 MB
  blk.64.ffn_gate.weight                [5120, 17408]  Q4_K_M      38.3 MB
  blk.64.ffn_up.weight                  [5120, 17408]  Q4_K_M      38.3 MB
  blk.64.nextn.eh_proj.weight           [10240, 5120]  Q8_0        55.7 MB
  blk.64.nextn.enorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.hnorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.shared_head_norm.weight  [5120]         F32         20.5 KB
  blk.64.post_attention_norm.weight     [5120]         F32         20.5 KB
                                                                   ----------
                                                                   208.8 MB

UD_Q3_K_XL

  NAME                                  SHAPE          TYPE        SIZE
  ------------------------------------  -------------  ----------  ----------
  blk.64.attn_k.weight                  [5120, 1024]   Q4_K_M      2.3 MB
  blk.64.attn_k_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_norm.weight               [5120]         F32         20.5 KB
  blk.64.attn_output.weight             [6144, 5120]   Q5_K_S      17.7 MB
  blk.64.attn_q.weight                  [5120, 12288]  Q4_K_M      27.0 MB
  blk.64.attn_q_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_v.weight                  [5120, 1024]   Q5_K_M      3.6 MB
  blk.64.ffn_down.weight                [17408, 5120]  Q5_K_S      50.1 MB
  blk.64.ffn_gate.weight                [5120, 17408]  Q4_K_M      38.3 MB
  blk.64.ffn_up.weight                  [5120, 17408]  Q4_K_M      38.3 MB
  blk.64.nextn.eh_proj.weight           [10240, 5120]  Q8_0        55.7 MB
  blk.64.nextn.enorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.hnorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.shared_head_norm.weight  [5120]         F32         20.5 KB
  blk.64.post_attention_norm.weight     [5120]         F32         20.5 KB
                                                                   ----------
                                                                   233.1 MB

IQ4_XS

  NAME                                  SHAPE          TYPE        SIZE
  ------------------------------------  -------------  ----------  ----------
  blk.64.attn_k.weight                  [5120, 1024]   Q5_K_S      2.9 MB
  blk.64.attn_k_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_norm.weight               [5120]         F32         20.5 KB
  blk.64.attn_output.weight             [6144, 5120]   Q5_K_S      17.7 MB
  blk.64.attn_q.weight                  [5120, 12288]  Q5_K_S      35.4 MB
  blk.64.attn_q_norm.weight             [256]          F32         1.0 KB
  blk.64.attn_v.weight                  [5120, 1024]   Q5_K_M      3.6 MB
  blk.64.ffn_down.weight                [17408, 5120]  Q5_K_S      50.1 MB
  blk.64.ffn_gate.weight                [5120, 17408]  Q5_K_S      50.1 MB
  blk.64.ffn_up.weight                  [5120, 17408]  Q5_K_S      50.1 MB
  blk.64.nextn.eh_proj.weight           [10240, 5120]  Q8_0        55.7 MB
  blk.64.nextn.enorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.hnorm.weight             [5120]         F32         20.5 KB
  blk.64.nextn.shared_head_norm.weight  [5120]         F32         20.5 KB
  blk.64.post_attention_norm.weight     [5120]         F32         20.5 KB
                                                                   ----------
                                                                   265.9 MB

Run tooltd/Qwen3.6-27B-MTP-TENSORS-ONLY-GGUF with guIDE

Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.

Download guIDE → · Browse 524k+ models · Compare models

Source: Hugging Face · Compare models