QW

Flashnext Abliterated

QwenGGUF by windowsxp811203Updated 2026-09-17
Download on Hugging Face
Parameters177B
Context256K
Layers48
Architectureqwen4exp
Smallest quant119.2 GB
Quick answer

Flashnext Abliterated needs about 120.5 GB of memory at Q4_K_M (a 119.2 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 5 quants, from 119.2 GB (Q4_K_M) to 354.0 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.2 GB.

Flashnext Abliterated – 190B Qwen model with refusal removal and 262k context

Flashnext Abliterated is a 190 billion‑parameter Qwen4exp model with a 262,144‑token context window. It removes refusal behavior via orthogonalization and retains the original vision tower and n‑gram table. The GGUF builds require a patched llama.cpp with qwen4_exp support.

Good at

  • Refusal removal
  • Large context handling
  • High‑parameter Qwen architecture

Limits

  • Requires patched llama.cpp
  • No MTP draft head support
  • Large memory footprint for high‑precision quants

Best for

  • Long‑form content generation
  • Research prototyping
  • Enterprise LLM deployment

Who it’s for. Developers and researchers needing a large Qwen model with refusal mitigation and extended context. Suitable for teams with access to a patched llama.cpp and sufficient GPU memory.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_K_M119.2 GB 120.5 GB
Too big Get file →
Q5_K_M134.1 GB 135.4 GB
Too big Get file →
Q6_K167.6 GB 168.9 GB
Too big Get file →
Q8_0188.2 GB 189.5 GB
Too big Get file →
BF16354.0 GB 355.3 GB
Too big Get file →

For 8 GB VRAM, use Q4_K_M (119 GB file, ~120 GB VRAM at 8k context). For 16 GB VRAM, Q5_K_M (134 GB file, ~135 GB VRAM). For 32 GB VRAM, Q6_K (168 GB file, ~169 GB VRAM). Higher‑precision quants increase memory proportionally.

What GPU do you need for Flashnext Abliterated?

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: no quant fits at an 8K context.
  • 32 GB VRAM: no quant fits at an 8K context.
  • 48 GB VRAM: no quant fits at an 8K context.

Which Flashnext Abliterated quant should you choose?

Q4_K_M is the usual starting point for Flashnext Abliterated: it balances file size and quality well. Here it is a 119.2 GB file.

The Q8_0 file is 188.2 GB, about 1.6 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architectureqwen4exp
Parameters177B (approximate, mixture of experts)
Layers48
Trained context262,144 tokens
KV cache (fp16)96 KB per token

Source

Quantized bywindowsxp811203
FamilyQwen
Licenseother
Downloads (HF)13,364
Last updated2026-09-17

Frequently asked questions about Flashnext Abliterated

How much VRAM does Flashnext Abliterated need?

About 120.5 GB at Q4_K_M with an 8K context window. That is the 119.2 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 Flashnext Abliterated on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 120.5 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Flashnext Abliterated GGUF?

It depends on the quant: Q4_K_M is 119.2 GB, Q5_K_M is 134.1 GB, Q6_K is 167.6 GB, Q8_0 is 188.2 GB, BF16 is 354.0 GB.

What context length does Flashnext Abliterated 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 3.2 GB at 32K.

Where can I download Flashnext Abliterated GGUF?

The files are on Hugging Face in the windowsxp811203/Qwen3.8-Flash-Next-Abliterated-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 Flashnext Abliterated 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 Flashnext Abliterated on 8 GB VRAM?

Only with the Q4_K_M quant, which still requires ~120 GB VRAM at 8k context; 8 GB is insufficient.

Does this model support vision tasks?

The vision tower is byte‑identical to the parent model, but the GGUF builds lack MTP draft head support.

What llama.cpp version is needed?

A patched llama.cpp with qwen4_exp support (PR #27742) is required; stock builds will not load the files.

Similar models

Browse all →

More from windowsxp811203

All by windowsxp811203 →

More Qwen models

All Qwen →

Other large models

All large models →