GE

Abliterated

GemmaGGUF by anlordUpdated 2026-10-10
Download on Hugging Face
Parameters4.63B
Context128K
Layers35
Architecturegemma4
Smallest quant3.3 GB
Quick answer

Abliterated needs about 4.5 GB of memory at Q4_K_M (a 3.4 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 12 quants, from 3.3 GB (Q4_0) to 9.3 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.3 GB.

Gemma-4-E2B-it Abliterated – text‑only GGUF model with reduced refusal rate

Gemma‑4‑E2B‑it Abliterated is a 4.63 B parameter text‑tower model derived from google/gemma‑4‑E2B‑it. The refusal behavior has been removed via directional ablation, lowering refusals from 99 / 100 to 4 / 100 on the standard evaluation set. All 35 decoder layers were modified; vision and audio towers remain unchanged.

Good at

  • Reasoning with [Start thinking] blocks
  • Tool usage via built‑in prompts
  • Low refusal rate on harmful prompts

Limits

  • Text tower only – no vision or audio support
  • Requires llama‑cpp Gemma4‑capable build
  • No fine‑tuning adapters included

Best for

  • Chatbots needing uncensored responses
  • Research on refusal mitigation
  • Low‑latency inference on GPUs with 8 GB VRAM

Who it’s for. Developers and researchers who want a Gemma‑4‑based model with minimal refusal behavior and no additional training. Suitable for local deployment in llama.cpp or compatible inference engines.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_03.3 GB 4.4 GB
Fits Get file →
Q4_K_S3.4 GB 4.4 GB
Fits Get file →
Q4_K_M3.4 GB 4.5 GB
Fits Get file →
Q4_13.5 GB 4.6 GB
Fits Get file →
Q5_03.6 GB 4.7 GB
Fits Get file →
Q5_K_S3.6 GB 4.7 GB
Fits Get file →
Q5_K_M3.6 GB 4.7 GB
Fits Get file →
Q5_13.7 GB 4.8 GB
Fits Get file →
Q6_K3.8 GB 4.9 GB
Fits Get file →
Q8_04.9 GB 6.0 GB
Fits Get file →
BF169.3 GB 10.4 GB
Too big Get file →
F169.3 GB 10.4 GB
Too big Get file →

For 8 GB VRAM, Q4_K_M (3.42 GB) or Q4_K_S (3.35 GB) fit comfortably; Q5_K_M (3.62 GB) also fits. For larger VRAM, Q6_K (3.83 GB) or Q8_0 (4.95 GB) provide higher quality. F16/BF16 require CPU offload.

What GPU do you need for 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: the best quant that fits is Q8_0 (4.9 GB file, about 6.0 GB in use).
  • 8 GB VRAM: the best quant that fits is Q8_0 (4.9 GB file, about 6.0 GB in use).
  • 12 GB VRAM: the best quant that fits is BF16 (9.3 GB file, about 10.4 GB in use).

Which Abliterated quant should you choose?

Q4_K_M is the usual starting point for Abliterated: it balances file size and quality well. Here it is a 3.4 GB file.

The Q8_0 file is 4.9 GB, about 1.4 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.

If memory is tight, the smallest file is Q4_0 at 3.3 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

Architecturegemma4
Parameters4.63B
Layers35
Trained context131,072 tokens
KV cache (fp16)70 KB per token

Source

Quantized byanlord
FamilyGemma
Licensegemma
Downloads (HF)1,011
Last updated2026-10-10

Frequently asked questions about Abliterated

How much VRAM does Abliterated need?

About 4.5 GB at Q4_K_M with an 8K context window. That is the 3.4 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 Abliterated on an 8 GB GPU?

Yes, with Q8_0: the 4.9 GB file needs about 6.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Abliterated GGUF?

It depends on the quant: Q4_0 is 3.3 GB, Q4_K_S is 3.4 GB, Q4_K_M is 3.4 GB, Q4_1 is 3.5 GB, Q5_0 is 3.6 GB, Q5_K_S is 3.6 GB and 6 more.

What context length does Abliterated support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.6 GB at 8K and 2.3 GB at 32K.

Where can I download Abliterated GGUF?

The files are on Hugging Face in the anlord/gemma-4-E2B-it-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 Abliterated use?

Hugging Face lists the license as gemma. 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 Q4_K_M on an 8 GB GPU?

Yes. Q4_K_M is 3.42 GB and fits in 8 GB VRAM with room for KV cache.

Does this model support vision or audio?

No. The GGUFs contain only the text tower; vision and audio towers are omitted.

Will the model refuse harmful prompts?

Refusal rate is reduced to 4 / 100 on the standard evaluation set, but it remains uncensored compared to the original.

Similar models

Browse all →

More from anlord

All by anlord →

More Gemma models

All Gemma →

Other small models

All small models →