← back to catalog · registered 2026-11-18 21:04

mrfakename/GLM-5.3-Flash-abliterated

mrfakename Glm GGUF multimodal
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/mrfakename%2FGLM-5.3-Flash-abliterated"
Response includes
  • classification m8
  • files 76
  • author_summary 12 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M8
Primary method

Repackaging (quantization)

Applied on top of direct removal inherited from the base model.
Confidence
MEDIUM
Inherited from base model
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=1
  • assume M1 (base ablation) + M8 (GGUF quant) - default when producer unknown
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
262
Likes
2
Descendants
3
in 3 direct forks
Model age
3d ago
created 2026-11-18
Downloads over time
Now 262 → from 0 ↑ 0%
096192288 0 on Nov 18 262 on Nov 22 Nov
Nov 18 → Nov 22 · 5 snapshots · spans 4 days

Genealogy 3 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Variants by this author 2 formats · 587 downloads combined

The same weights this author released in different packaging. Pick the format that matches your runtime.

Metadata

License
mit
Tags
safetensors gguf glm5_next abliterated uncensored image-text-to-text conversational base_model:zai-org/GLM-5.3-Flash base_model:quantized:zai-org/GLM-5.3-Flash license:mit eval-results fp8

Related

Control vectors

Total size
306 GB
Files
76
Quantizations
1
Registered
2026-11-18 21:04
Last updated on HF
2026-11-18 21:01

Files by quantization

Auxiliary files 76 files 306 GB
model-00001-of-00062.safetensors 5.00 GB 9ff3c939 download
model-00024-of-00062.safetensors 5.00 GB 4256e216 download
model-00038-of-00062.safetensors 5.00 GB 2ff40ce5 download
model-00045-of-00062.safetensors 5.00 GB a03f8faf download
model-00052-of-00062.safetensors 5.00 GB a3157059 download
model-00007-of-00062.safetensors 5.00 GB fd7c1446 download
model-00014-of-00062.safetensors 5.00 GB d0b71d5a download
model-00021-of-00062.safetensors 5.00 GB ad2f64b6 download
model-00042-of-00062.safetensors 5.00 GB 449070b2 download
model-00049-of-00062.safetensors 5.00 GB cd71de75 download
model-00050-of-00062.safetensors 5.00 GB cce21e0d download
model-00008-of-00062.safetensors 5.00 GB 96e573f9 download
model-00015-of-00062.safetensors 5.00 GB 5e66e9c3 download
model-00018-of-00062.safetensors 5.00 GB 9f928d70 download
model-00025-of-00062.safetensors 5.00 GB 233cdf4b download
model-00029-of-00062.safetensors 5.00 GB 4d55140d download
model-00036-of-00062.safetensors 5.00 GB 8af53de7 download
model-00004-of-00062.safetensors 5.00 GB 45057a8e download
model-00039-of-00062.safetensors 5.00 GB 63300e9a download
model-00043-of-00062.safetensors 5.00 GB 2b9fcc49 download
model-00012-of-00062.safetensors 5.00 GB 9ffb5185 download
model-00019-of-00062.safetensors 5.00 GB 2d4e0370 download
model-00026-of-00062.safetensors 5.00 GB ec5b66dd download
model-00033-of-00062.safetensors 5.00 GB 894cb474 download
model-00054-of-00062.safetensors 5.00 GB 5a2394c6 download
model-00031-of-00062.safetensors 5.00 GB 574e1a9f download
model-00056-of-00062.safetensors 5.00 GB 07194080 download
model-00047-of-00062.safetensors 5.00 GB 8e64b3ae download
model-00060-of-00062.safetensors 5.00 GB 0f642a38 download
model-00057-of-00062.safetensors 5.00 GB 406f28a1 download
model-00009-of-00062.safetensors 5.00 GB 69cbb58b download
model-00017-of-00062.safetensors 5.00 GB 32adf86c download
model-00035-of-00062.safetensors 5.00 GB 03b4c853 download
model-00028-of-00062.safetensors 5.00 GB 500f5621 download
model-00022-of-00062.safetensors 5.00 GB d7370d62 download
model-00053-of-00062.safetensors 5.00 GB 3343e3b6 download
model-00005-of-00062.safetensors 5.00 GB ade210c8 download
model-00011-of-00062.safetensors 5.00 GB d43094d0 download
model-00040-of-00062.safetensors 5.00 GB ed30d5a2 download
model-00046-of-00062.safetensors 5.00 GB 84d7c744 download
model-00032-of-00062.safetensors 5.00 GB 4ce63439 download
model-00059-of-00062.safetensors 5.00 GB 96a3892e download
model-00003-of-00062.safetensors 5.00 GB d7bede1b download
model-00010-of-00062.safetensors 4.99 GB 9492a29e download
model-00048-of-00062.safetensors 4.99 GB 85c51200 download
model-00027-of-00062.safetensors 4.99 GB fdafd6e7 download
model-00034-of-00062.safetensors 4.99 GB 451bb036 download
model-00041-of-00062.safetensors 4.99 GB 9ddfde38 download
model-00020-of-00062.safetensors 4.99 GB 67a86a8b download
model-00013-of-00062.safetensors 4.99 GB bfecdc0f download
model-00006-of-00062.safetensors 4.99 GB 1653c5e3 download
model-00051-of-00062.safetensors 4.99 GB b3423a93 download
model-00044-of-00062.safetensors 4.99 GB ed57a936 download
model-00030-of-00062.safetensors 4.99 GB a28adbc9 download
model-00037-of-00062.safetensors 4.99 GB 0e7c51bd download
model-00023-of-00062.safetensors 4.99 GB 1ea8f1bb download
model-00016-of-00062.safetensors 4.99 GB 6cc97434 download
model-00055-of-00062.safetensors 4.99 GB 8d64c91f download
model-00058-of-00062.safetensors 4.99 GB ac4e58d1 download
model-00002-of-00062.safetensors 4.96 GB 6e256030 download
model-00061-of-00062.safetensors 4.94 GB 1645e645 download
model-00062-of-00062.safetensors 1.17 GB d3087816 download
abliteration-lora.gguf 7.26 MB c9dba843 download
tokenizer.json 19.3 MB 19e77364 download
model.safetensors.index.json 8.02 MB 6e8b6ed4 download
stats.png 136 KB 88efcc64 download
config.json 67.8 KB f93128cf download
abliteration-edit.npz 34.9 KB fb6259e3 download
bake-stats.txt 16.0 KB 25a5656d download
chat_template.jinja 10.6 KB 14180451 download
README.md 2.87 KB e687f197 download
.gitattributes 1.64 KB b63d6691 download
LICENSE 1.04 KB 986b06fb download
processor_config.json 909 B 3ec2a058 download
tokenizer_config.json 761 B e375fa0a download
generation_config.json 194 B 637ee6af download

README current version from Hugging Face


license: mit
base_model: zai-org/GLM-5.3-Flash
pipeline_tag: image-text-to-text
tags:

  • abliterated
  • uncensored

GLM-5.3-Flash (abliterated)

Huge thanks to Hugging Face for generously sponsoring the compute used to abliterate these models! This was done on HF Jobs, which makes it incredibly easy to run tasks such as these :)

Join the Discord: https://discord.com/invite/foss

refusal rates before and after, the search trade-off, and the per-layer edit

This is zai-org/GLM-5.3-Flash with its refusal behaviour removed by directional ablation. It has the same files, format and FP8/BF16 layout as the original, so it loads anywhere the original loads.

GGUF quants: mrfakename/GLM-5.3-Flash-abliterated-GGUF.

What changed

Chat mode and think mode refuse along slightly different directions, so instead of a single refusal direction a rank-2 refusal subspace was removed:

  • Refusal directions were measured separately for chat and think prompts at every layer, on the Q8_0 GGUF.
  • The subspace is the top two singular vectors of all of those directions together.
  • It was projected out of the residual-stream writes of the 45 main layers: self_attn.o_proj, mlp.shared_experts.down_proj, and mlp.down_proj on the three dense layers. That is 90 tensors, and each layer got its own strength.

The per-layer strengths came from a multi-objective search: minimise refusals and keep the first-token KL on harmless prompts low.

Nothing else was touched:

  • Routed experts, embeddings, the vision tower and the MTP/nextn layer are bit-identical to the original.
  • FP8 tensors were requantized with their original block scales, using stochastic rounding so the removal is not undone by rounding.

Results (Q8_0 GGUF, greedy)

refusals, chat (/100) think mode, reasoning or answer refuses (/100) KL, harmless first token
original 99 98 0
abliterated 3 2 0.038

The 100 prompts are held-out harmful requests. Chat replies were 384 tokens; think mode got 768 tokens. Where the reasoning had not finished by then, the reasoning text itself was judged. The KL figure includes a small noise floor from batched inference.

Files

  • model-*.safetensors: the patched checkpoint (same index as the original).
  • abliteration-lora.gguf: a rank-2 LoRA that applies the same edit to the original GGUFs, e.g. llama-server -m GLM-5.3-Flash-Q8_0-00001-of-….gguf --lora abliteration-lora.gguf.
  • abliteration-edit.npz: the subspace basis Q and the per-layer strengths scales[layer, (attn, mlp)]. The edit is W' = W - s·QQᵀW.
  • bake-stats.txt: per-tensor check of how much of the subspace is left after the edit, plus FP8 requantization error.

This model will comply with harmful requests. You are responsible for how you use it.

README history 4 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-11-18 card: Discord link 7c10d59 2.9 KB
    Loading...
  2. 2026-11-18 card: bold thanks line 5ab2dc4 2.8 KB
    Loading...
  3. 2026-11-18 card: thanks to Hugging Face for compute 4a4477f 2.8 KB
    Loading...
  4. 2026-11-18 abliterated GLM-5.3-Flash 510b6a8 2.6 KB
    Loading...
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration