Qwen3.8 Uncensored NVFP4 MTP 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.
Qwen3.8 Uncensored NVFP4 MTP – 27B model with native FP4 quant and speculative decoding
This 27B Qwen model is quantized to NVFP4 for Blackwell GPUs. It retains the MTP speculative head and includes a vision projector. The model supports a 262,144 token context and runs on llama.cpp with qwen35 architecture.
Good at
- Fast decoding with draft-mtp
- Native FP4 density on RTX 5090/5080
- Vision input via mmproj
Limits
- Requires Blackwell GPU
- No retraining needed
- Large VRAM requirement
Best for
- High‑throughput inference on RTX 5090
- Vision‑enabled chatbots
- Fast speculative decoding
Who it’s for. Engineers and researchers needing a fast, uncensored 27B model on Blackwell GPUs. The model is ideal for applications that benefit from speculative decoding and vision capabilities.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 19.7 GB | 22.4 GB | Too big | Get file → |
The single GGUF file is 19.69 GB and needs about 22.4 GB VRAM at 8 k context. It is suitable for GPUs with 24 GB or more VRAM. For lower VRAM budgets, a different quantization is required, which is not provided here.
What GPU do you need for Qwen3.8 Uncensored 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: 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
Source
Frequently asked questions about Qwen3.8 Uncensored NVFP4 MTP
How much VRAM does Qwen3.8 Uncensored NVFP4 MTP 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 Qwen3.8 Uncensored NVFP4 MTP 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 Qwen3.8 Uncensored NVFP4 MTP GGUF?
It depends on the quant: file is 19.7 GB.
What context length does Qwen3.8 Uncensored 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 2.2 GB at 8K and 8.7 GB at 32K.
Where can I download Qwen3.8 Uncensored NVFP4 MTP GGUF?
The files are on Hugging Face in the BennyDaBall/Qwen3.8-Uncensored-NVFP4-MTP 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 Uncensored 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 this model on 8 GB VRAM?
No, the model requires approximately 22.4 GB VRAM at 8 k context, so 8 GB is insufficient.
Does the model support vision?
Yes, it includes a BF16 vision projector that can be loaded with --mmproj for image and video input.
Is the MTP head available?
Yes, the model ships with the MTP speculative head retained for fast decoding.