Qwen3.8 27B abliterated ROCmFPX needs about 25.2 GB of memory at Q6_0 (a 22.5 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 7 quants, from 14.6 GB (Q4_0) to 28.7 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 8.7 GB.
Qwen3.8 27B abliterated ROCmFPX – 27B dense model with vision support
This model is a 27B parameter Qwen variant that has undergone ablation to reduce refusals. It retains vision capabilities and supports a context length of up to 262,144 tokens. The model is available in multiple ROCmFPX quantization tiers for AMD hardware.
Good at
- Vision tasks with CLIP + Qwen3VL merger
- Large context generation up to 256K tokens
- Research on ablation effects
Limits
- No safety filtering – may produce sensitive content
- Requires strix.cpp or llama.cpp-rocm for ROCmFPX
- Large VRAM needed for higher quant tiers
Best for
- Vision‑enabled research prototypes
- Large‑context text generation
- Testing ablation techniques
Who it’s for. Researchers and developers who need a vision‑capable Qwen model with reduced safety filtering for controlled experiments. It is suitable for environments where outputs can be monitored and reviewed.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_0 | 14.6 GB | 17.2 GB | Too big | Get file → | |
| Q4_0 | 14.9 GB | 17.6 GB | Too big | Get file → | |
| Q4_0 | 15.0 GB | 17.7 GB | Too big | Get file → | |
| Q6_0 | 22.5 GB | 25.2 GB | Too big | Get file → | |
| Q6_0 | 25.3 GB | 28.0 GB | Too big | Get file → | |
| Q8_0 | 28.2 GB | 30.9 GB | Too big | Get file → | |
| Q8_0 | 28.7 GB | 31.3 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_0_ROCMFP4_FAST (≈14.56 GB file, ~17.2 GB VRAM). For 12 GB VRAM, Q6_0_ROCMFPX (≈22.53 GB file, ~25.2 GB VRAM). For 16 GB VRAM, Q8_0_ROCMFPX (≈28.19 GB file, ~30.9 GB VRAM). Estimates are approximate.
What GPU do you need for Qwen3.8 27B abliterated ROCmFPX?
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 Q4_0 (15.0 GB file, about 17.7 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (28.7 GB file, about 31.3 GB in use).
Which Qwen3.8 27B abliterated ROCmFPX quant should you choose?
Q6_0 (22.5 GB) is the middle option in this repository. Very close to the original quality for most uses, at a clearly larger size than 4-bit.
If memory is tight, the smallest file is Q4_0 at 14.6 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 27B abliterated ROCmFPX
How much VRAM does Qwen3.8 27B abliterated ROCmFPX need?
About 25.2 GB at Q6_0 with an 8K context window. That is the 22.5 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 abliterated ROCmFPX on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 17.2 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Qwen3.8 27B abliterated ROCmFPX GGUF?
It depends on the quant: Q4_0 is 14.6 GB, Q4_0 is 14.9 GB, Q4_0 is 15.0 GB, Q6_0 is 22.5 GB, Q6_0 is 25.3 GB, Q8_0 is 28.2 GB and 1 more.
What context length does Qwen3.8 27B abliterated ROCmFPX 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 27B abliterated ROCmFPX GGUF?
The files are on Hugging Face in the ArtomYuan/Qwen3.8-27B-abliterated-ROCmFPX 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 abliterated ROCmFPX 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, using the Q4_0_ROCMFP4_FAST tier which requires about 17.2 GB VRAM at 8K context, so it is close to the limit and may need lower context.
Does this model support vision?
Yes, it includes a bundled mmproj with CLIP and Qwen3VL vision tower.
What license applies to this model?
The model is released under the Apache‑2.0 license.