Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP needs about 15.8 GB of memory at Q3_K_XL (a 13.1 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 16 quants, from 8.8 GB (IQ2_XS) to 54.7 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 8.7 GB.
Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP – a large vision‑capable LLM
Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP is a 28 billion‑parameter model based on Qwen‑3.5. It includes a vision tower and an MTP head for self‑speculative decoding. The model is quantized into multiple GGUF formats for different GPU memory budgets.
Good at
- Vision input and processing
- Self‑speculative decoding via MTP
- Large context handling up to 262 144 tokens
Limits
- No text‑only license restrictions
- Requires Blackwell‑architecture GPU for NVFP4
- MTP head only supported in llama.cpp builds with MTP support
Best for
- Image‑guided text generation
- Video captioning with vision input
- Research on self‑speculative decoding
Who it’s for. Researchers and developers needing a vision‑capable LLM with a self‑speculative decoding head. The model is suitable for experimentation on Blackwell GPUs and for projects that can handle the uncensored content.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XS | 8.8 GB | 11.5 GB | Too big | Get file → | |
| IQ2_S | 9.6 GB | 12.3 GB | Too big | Get file → | |
| Q2_K_XL | 9.8 GB | 12.5 GB | Too big | Get file → | |
| IQ3_XXS | 10.4 GB | 13.1 GB | Too big | Get file → | |
| IQ3_XXS | 10.9 GB | 13.6 GB | Too big | Get file → | |
| IQ3_S | 12.0 GB | 14.7 GB | Too big | Get file → | |
| IQ3_S | 12.1 GB | 14.8 GB | Too big | Get file → | |
| Q3_K_XL | 13.1 GB | 15.8 GB | Too big | Get file → | |
| IQ4_XS | 14.3 GB | 16.9 GB | Too big | Get file → | |
| Q4_K_S | 15.4 GB | 18.0 GB | Too big | Get file → | |
| Q4_K_XL | 17.6 GB | 20.2 GB | Too big | Get file → | |
| Q5_K_S | 18.7 GB | 21.3 GB | Too big | Get file → | |
| Q5_K_M | 19.8 GB | 22.5 GB | Too big | Get file → | |
| Q6_K_XL | 25.3 GB | 28.0 GB | Too big | Get file → | |
| Q8_K_XL | 31.5 GB | 34.1 GB | Too big | Get file → | |
| BF16 | 54.7 GB | 57.3 GB | Too big | Get file → |
For 8 GB VRAM, use IQ2_XS (file size 8.77 GB, estimated 11.5 GB memory). For 12 GB VRAM, IQ2_S (9.61 GB, 12.3 GB). For 16 GB VRAM, Q4_K_XL (17.56 GB, 20.2 GB). Larger budgets can use Q5_K_M or Q6_K_XL for higher quality.
What GPU do you need for Swift 1.5 Qwen3.8 27B Uncensored Dynamic 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 IQ2_XS (8.8 GB file, about 11.5 GB in use).
- 16 GB VRAM: the best quant that fits is Q3_K_XL (13.1 GB file, about 15.8 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_K_M (19.8 GB file, about 22.5 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K_XL (25.3 GB file, about 28.0 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_K_XL (31.5 GB file, about 34.1 GB in use).
Which Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP quant should you choose?
Q3_K_XL (13.1 GB) is the middle option in this repository. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
If memory is tight, the smallest file is IQ2_XS at 8.8 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP
How much VRAM does Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP need?
About 15.8 GB at Q3_K_XL with an 8K context window. That is the 13.1 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 Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 11.5 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP GGUF?
It depends on the quant: IQ2_XS is 8.8 GB, IQ2_S is 9.6 GB, Q2_K_XL is 9.8 GB, IQ3_XXS is 10.4 GB, IQ3_XXS is 10.9 GB, IQ3_S is 12.0 GB and 10 more.
What context length does Swift 1.5 Qwen3.8 27B Uncensored Dynamic 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 Swift 1.5 Qwen3.8 27B Uncensored Dynamic MTP GGUF?
The files are on Hugging Face in the ajgazin/Swift-1.5-Qwen3.8-27B-Uncensored-Dynamic-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 Swift 1.5 Qwen3.8 27B Uncensored Dynamic 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 Swift 1.5 Qwen3.8 27B on 8 GB VRAM?
Use the IQ2_XS quant; it requires about 11.5 GB of VRAM at 8 k context, so it will fit on 8 GB with some swapping.
Does the model support vision input?
Yes, the vision tower is included and works with image and video input through the projector.
Is the MTP head available in all GGUF files?
The MTP head is included in every main GGUF; no separate draft file is needed.