Qwen3.8 27B NVFP4 MTP needs about 18.5 GB of memory at Q4_K_M (a 15.8 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 4 quants, from 5.9 GB to 29.2 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
Qwen3.8-27B NVFP4 MTP – dense vision‑language model with speculative decoding
Qwen3.8-27B NVFP4 MTP is a 27B‑parameter dense model from the Qwen family. It supports vision and text, and includes a Multi‑Token Prediction head for speculative decoding in llama.cpp. The model is released under the apache‑2.0 license.
Good at
- Vision‑language understanding
- Long‑context reasoning up to 262144 tokens
- Speculative decoding with MTP
Limits
- No explicit vision support in all harnesses
- Large VRAM requirement for full precision
- Requires NVFP4 quantization for efficient use
Best for
- Image captioning and multimodal QA
- Complex multi‑step reasoning tasks
- High‑throughput inference with speculative decoding
Who it’s for. Developers building multimodal applications that need both vision and text understanding. Researchers exploring long‑context reasoning and speculative decoding techniques. Teams requiring a dense model with a large context window and flexible control.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 5.9 GB | 8.6 GB | Too big | Get file → | |
| Q4_K_M | 15.8 GB | 18.5 GB | Too big | Get file → | |
| Q8_0 | 20.0 GB | 22.7 GB | Too big | Get file → | |
| BF16 | 29.2 GB | 31.9 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_K_M (file size 15.81 GB, estimated 18.5 GB memory at 8 k context). For 12 GB VRAM, Q8_0 (20.04 GB, 22.7 GB memory) is suitable. For larger GPUs, BF16 (29.21 GB, 31.9 GB memory) offers the highest precision.
What GPU do you need for Qwen3.8 27B NVFP4 MTP?
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: the best quant that fits is the main file (5.9 GB file, about 8.6 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (5.9 GB file, about 8.6 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (20.0 GB file, about 22.7 GB in use).
- 32 GB VRAM: the best quant that fits is BF16 (29.2 GB file, about 31.9 GB in use).
Which Qwen3.8 27B NVFP4 MTP quant should you choose?
Q4_K_M is the usual starting point for Qwen3.8 27B NVFP4 MTP: it balances file size and quality well. Here it is a 15.8 GB file.
The Q8_0 file is 20.0 GB, about 1.3 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is the first one in the table at 5.9 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.8 27B NVFP4 MTP
How much VRAM does Qwen3.8 27B NVFP4 MTP need?
About 18.5 GB at Q4_K_M with an 8K context window. That is the 15.8 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 Qwen3.8 27B NVFP4 MTP on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 8.6 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Qwen3.8 27B NVFP4 MTP GGUF?
It depends on the quant: file is 5.9 GB, Q4_K_M is 15.8 GB, Q8_0 is 20.0 GB, BF16 is 29.2 GB.
What context length does Qwen3.8 27B NVFP4 MTP support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download Qwen3.8 27B NVFP4 MTP GGUF?
The files are on Hugging Face in the LibertAIDAI/Qwen3.8-27B-NVFP4-MTP-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Qwen3.8 27B NVFP4 MTP 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 Qwen3.8-27B NVFP4 MTP on 8 GB VRAM?
The Q4_K_M quantization requires about 18.5 GB memory at 8 k context, so 8 GB VRAM is insufficient.
Does this model support vision tasks?
Yes, it includes a vision tower and is capable of image and video understanding.
What is the context length of this model?
The model was trained with a context length of 262144 tokens.