Qwen3.8 Flash Next MTP needs about 3.9 GB of memory at Q4_K_M (a 2.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 1.9 GB (Q4_K_XL) to 7.8 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.3 GB.
Qwen3.8 Flash Next MTP – a multi‑token‑prediction head for Qwen3.8‑Flash‑Next
Qwen3.8 Flash Next MTP is a draft multi‑token‑prediction head exported from the bf16 checkpoint of Qwen3.8‑Flash‑Next. It is compatible with llama.cpp kernels and requires a patched build. The head is designed for efficient inference on large‑context workloads.
Good at
- Multi‑token prediction
- Efficient large‑context handling
- Compatible with patched llama.cpp
Limits
- Requires patched llama.cpp or specific branch
- No vision support
- License is other
Best for
- LLM inference with long context
- Research on draft‑head performance
- Testing Qwen4exp architecture
Who it’s for. Developers and researchers who need a multi‑token‑prediction head for Qwen3.8‑Flash‑Next and are willing to use a patched llama.cpp build. It is not a standalone model but a head that can be combined with existing Flash‑Next GGUFs.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_XL | 1.9 GB (2 parts) | 3.2 GB | Fits | Get file → | |
| Q4_K_M | 2.6 GB | 3.9 GB | Fits | Get file → | |
| BF16 | 7.8 GB | 9.1 GB | Too big | Get file → |
For 8 GB VRAM, use the Q4_K_XL quant (1.9 GB file, ~3.2 GB memory at 8 k context). For 12 GB VRAM, Q4_K_M (2.62 GB file, ~3.9 GB memory) is suitable. BF16 requires ~9.1 GB memory and 7.77 GB file size, suitable only for larger GPUs.
What GPU do you need for Qwen3.8 Flash Next 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: the best quant that fits is Q4_K_M (2.6 GB file, about 3.9 GB in use).
- 6 GB VRAM: the best quant that fits is Q4_K_M (2.6 GB file, about 3.9 GB in use).
- 8 GB VRAM: the best quant that fits is Q4_K_M (2.6 GB file, about 3.9 GB in use).
- 12 GB VRAM: the best quant that fits is BF16 (7.8 GB file, about 9.1 GB in use).
Which Qwen3.8 Flash Next MTP quant should you choose?
Q4_K_M is the usual starting point for Qwen3.8 Flash Next MTP: it balances file size and quality well. Here it is a 2.6 GB file.
If memory is tight, the smallest file is Q4_K_XL at 1.9 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Qwen3.8 Flash Next MTP
How much VRAM does Qwen3.8 Flash Next MTP need?
About 3.9 GB at Q4_K_M with an 8K context window. That is the 2.6 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 Flash Next MTP on an 8 GB GPU?
Yes, with Q4_K_M: the 2.6 GB file needs about 3.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Qwen3.8 Flash Next MTP GGUF?
It depends on the quant: Q4_K_XL is 1.9 GB, Q4_K_M is 2.6 GB, BF16 is 7.8 GB.
What context length does Qwen3.8 Flash Next 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 0.8 GB at 8K and 3.3 GB at 32K.
Where can I download Qwen3.8 Flash Next MTP GGUF?
The files are on Hugging Face in the dzannotti/Qwen3.8-Flash-Next-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 Flash Next MTP use?
Hugging Face lists the license as other. 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 Flash Next MTP on 8 GB VRAM?
Yes, using the Q4_K_XL quant requires about 3.2 GB memory at 8 k context, which fits within 8 GB VRAM.
Do I need a special version of llama.cpp?
Yes, a patched llama.cpp (tag b10612 or the v98aa9847 branch) is required to load the qwen4exp graph.
Is there a vision capability in this head?
No, the MTP head does not include vision support.