Underdog Saluki 27B 1.0 MTP needs about 11.0 GB of memory at its main file (a 8.3 GB file) with an 8K context window, so it fits on a 12 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 8.7 GB.
Underdog Saluki 27B 1.0 MTP – a 27 B Qwen3.8‑based model with multi‑token‑prediction for llama.cpp
Underdog Saluki 27B 1.0 MTP is a 27 B parameter model based on Qwen3.8‑27B. It adds a multi‑token‑prediction head for self‑speculative decoding. The GGUF file is 8.35 GB and requires 11 GB VRAM at 8 k context.
Good at
- Tool calling with Qwen3.8 chat template
- Self‑speculative decoding via MTP head
- Stock llama.cpp compatibility
Limits
- Requires 11 GB VRAM at 8 k context
- No vision support in base file
- License is Apache‑2.0
Best for
- Local tool‑calling agents
- LLM inference on 12 GB GPUs
- Testing MTP drafting in llama.cpp
Who it’s for. Developers and researchers who need a Qwen3.8‑based model that runs on stock llama.cpp with MTP support. The model is suitable for local inference on GPUs with at least 12 GB VRAM.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 8.3 GB | 11.0 GB | Too big | Get file → |
The only available quant is the original 8.35 GB file. It needs about 11 GB VRAM at 8 k context. For lower VRAM use a different quant from the parent model, but it is not provided here.
What GPU do you need for Underdog Saluki 27B 1.0 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 (8.3 GB file, about 11.0 GB in use).
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Underdog Saluki 27B 1.0 MTP
How much VRAM does Underdog Saluki 27B 1.0 MTP need?
About 11.0 GB at its main file with an 8K context window. That is the 8.3 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 Underdog Saluki 27B 1.0 MTP on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 11.0 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Underdog Saluki 27B 1.0 MTP GGUF?
It depends on the quant: file is 8.3 GB.
What context length does Underdog Saluki 27B 1.0 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 Underdog Saluki 27B 1.0 MTP GGUF?
The files are on Hugging Face in the Kujira/Underdog-Saluki-27B-1.0-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 Underdog Saluki 27B 1.0 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 Underdog Saluki 27B 1.0 MTP on 8 GB VRAM?
No. The model requires about 11 GB VRAM at 8 k context, so 8 GB is insufficient.
Does the model support vision?
The base GGUF is text only. Add the 0.6–0.9 GB vision add‑on to enable image input.
What license applies to this model?
Apache‑2.0. The model is built from Qwen3.8‑27B and inherits the same license.