qwen3.8 27b gguf strix halo needs about 1.7 GB of memory at IQ4_XS (a 1.0 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 2 quants, from 1.0 GB (IQ4_XS) to 16.1 GB (IQ4_XS). Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Qwen3.8-27B GGUF Strix Halo – dense vision‑language model
Qwen3.8-27B is a 27B parameter dense model with native vision‑language support. It is quantized to IQ4_XS for efficient inference on AMD Strix Halo GPUs. The model supports reasoning, tools, and vision tasks.
Good at
- Vision‑language understanding
- Reasoning and tool use
- Long‑horizon agentic tasks
Limits
- No official vision projector included
- Requires separate mmproj file for image input
- Large context length may need more VRAM
Best for
- Multimodal document analysis
- Agentic task planning
- Research prototyping
Who it’s for. Developers building multimodal applications on AMD GPUs. Researchers needing a dense vision‑language model without large parameter count. Teams integrating with llama.cpp for low‑latency inference.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ4_XS | 1.0 GB | 1.7 GB | Fits | Get file → | |
| IQ4_XS | 16.1 GB | 16.8 GB | Too big | Get file → |
For 8‑GB VRAM use the 1.04‑GB IQ4_XS file; it needs about 1.7‑GB VRAM at 8‑k context. For larger workloads or 16‑GB VRAM use the 16.11‑GB IQ4_XS file, requiring about 16.8‑GB VRAM at 8‑k context. Estimates are approximate.
What GPU do you need for qwen3.8 27b gguf strix halo?
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 IQ4_XS (1.0 GB file, about 1.7 GB in use).
- 6 GB VRAM: the best quant that fits is IQ4_XS (1.0 GB file, about 1.7 GB in use).
- 8 GB VRAM: the best quant that fits is IQ4_XS (1.0 GB file, about 1.7 GB in use).
- 12 GB VRAM: the best quant that fits is IQ4_XS (1.0 GB file, about 1.7 GB in use).
- 16 GB VRAM: the best quant that fits is IQ4_XS (1.0 GB file, about 1.7 GB in use).
- 24 GB VRAM: the best quant that fits is IQ4_XS (16.1 GB file, about 16.8 GB in use).
Which qwen3.8 27b gguf strix halo quant should you choose?
IQ4_XS (1.0 GB) is the middle option in this repository. 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 27b gguf strix halo
How much VRAM does qwen3.8 27b gguf strix halo need?
About 1.7 GB at IQ4_XS with an 8K context window. That is the 1.0 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 gguf strix halo on an 8 GB GPU?
Yes, with IQ4_XS: the 1.0 GB file needs about 1.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of qwen3.8 27b gguf strix halo GGUF?
It depends on the quant: IQ4_XS is 1.0 GB, IQ4_XS is 16.1 GB.
What context length does qwen3.8 27b gguf strix halo support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.7 GB at 32K.
Where can I download qwen3.8 27b gguf strix halo GGUF?
The files are on Hugging Face in the ilintar/qwen3.8-27b-gguf-strix-halo 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 gguf strix halo 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?
Yes, use the 1.04‑GB IQ4_XS file which requires about 1.7‑GB VRAM at 8‑k context.
Does the model include a vision projector?
No, the repository does not include the Q8_0 multimodal projector; supply mmproj-Qwen3.8-27B-Q8_0.gguf separately for image input.
What license applies to this model?
The model is released under the apache-2.0 license.