QW

mango qwen38 27b nvfp4mtp

QwenGGUF by isiliconUpdated 2026-10-30
Download on Hugging Face
Parameters27.3B
Context256K
Layers65
Architectureqwen35
Smallest quant19.7 GB
Quick answer

mango qwen38 27b nvfp4mtp needs about 22.4 GB of memory at its main file (a 19.7 GB file) with an 8K context window, so it fits on a 24 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 8.7 GB.

27B Qwen model with native NVFP4 quantization and MTP head

mango qwen38 27b nvfp4mtp is a 27.3B parameter Qwen model quantized to NVFP4. It retains the MTP speculative head and includes a BF16 vision projector. The model supports a 262,144 token context and runs on Blackwell GPUs.

Good at

  • Fast decoding with MTP
  • Native NVFP4 density
  • Vision support via BF16 projector

Limits

  • Requires Blackwell (sm_120) GPU
  • Approx. 22.4 GB VRAM at 8k context
  • No retraining or distillation needed

Best for

  • High‑speed inference on RTX 5090/5080
  • Multimodal tasks with image input
  • Large‑context reasoning

Who it’s for. Developers and researchers needing a fast, uncensored 27B Qwen model on Blackwell GPUs. The model is suitable for applications that benefit from speculative decoding and multimodal input.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—19.7 GB 22.4 GB
Too big Get file →

The only available GGUF file is 19.69 GB with an estimated 22.4 GB VRAM requirement at 8k context. For users with less VRAM, a lower‑precision quant is not available in this release.

What GPU do you need for mango qwen38 27b nvfp4mtp?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is the main file (19.7 GB file, about 22.4 GB in use).

Specifications

Architecture from the GGUF header

Architectureqwen35
Parameters27.3B
Layers65
Trained context262,144 tokens
KV cache (fp16)260 KB per token

Source

Quantized byisilicon
FamilyQwen
Licenseapache-2.0
Downloads (HF)1,102
Last updated2026-10-30

Frequently asked questions about mango qwen38 27b nvfp4mtp

How much VRAM does mango qwen38 27b nvfp4mtp need?

About 22.4 GB at its main file with an 8K context window. That is the 19.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run mango qwen38 27b nvfp4mtp on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 22.4 GB, so you would need a larger GPU or CPU offloading.

What is the file size of mango qwen38 27b nvfp4mtp GGUF?

It depends on the quant: file is 19.7 GB.

What context length does mango qwen38 27b nvfp4mtp support?

The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.2 GB at 8K and 8.7 GB at 32K.

Where can I download mango qwen38 27b nvfp4mtp GGUF?

The files are on Hugging Face in the isilicon/mango-qwen38-27b-nvfp4mtp repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does mango qwen38 27b nvfp4mtp use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run this model on an RTX 3080 with 10 GB VRAM?

No. The model requires a Blackwell GPU and at least 22.4 GB VRAM at 8k context.

Does the model support image input?

Yes, the BF16 vision projector is included; add mmproj for image or video frames.

What license applies to this model?

The model is released under the apache-2.0 license.

Similar models

Browse all →

More from isilicon

All by isilicon →

More Qwen models

All Qwen →

Other medium models

All medium models →