dealignai/GLM-5.3-Flash-ABLITERATED-FP8

safetensorsglm5_nextabliterateduncensoredcrackrefusal-removed

License: mit
Size: not mirrored yet
Downloads in the last 30 days: 0 / 30 days

No official torrent yet — requesting it tells the flock what to host next. You cannot seed a catalog-only page.

Cataloged from dealignai/GLM-5.3-Flash-ABLITERATED-FP8

⚠️ Updated 2026-08-28 — files fixed · please re-download

This release supersedes the previous weights. It fixes a rare repetition-loop issue and the model is more capable than before: MMLU 87.33% — above base 86.74% — zero degeneration.

Reasoning-mode note (please read): GLM-5.3-Flash has multiple reasoning-effort modes. This CRACK is tuned to be fully uncensored at reasoning-OFF and MAX effort — the recommended and default modes — with 0% refusals on HarmBench-320. We deliberately kept the weight edit conservative to preserve model quality instead of aggressively ablating every effort level; low effort retains some refusals by design. Use reasoning-off or default/max effort for the fully-uncensored experience. Please re-download for the corrected files.

[Image blocked: No description]

GLM 5.3 CRACK Abliterated FP8

Abliterated · CRACK · guardrails removed at the weight level · native FP8 speed · vision + MTP working

a CRACK release by dealignai · Twitter @dealignai

Mirror of dealignai/GLM-5.3-Flash-UNCENSORED-FP8 — same weights, same CRACK.

What Is This?

CRACK is dealignai's brand for permanent, weight-level uncensoring. This is GLM-5.3-Flash in FP8 with its refusal behavior — which caused heavy over-refusal, especially on copyright and other benign-but-flagged requests — removed directly in the model weights. FP8 runs at native speed on Hopper (H100/H200) GPUs.

Genuine weight modification — none of the usual shortcuts:

❌ No fine-tuning / SFT / DPO. ❌ No cheap template / jailbreak-prompt tricks.
❌ No LoRA, adapters, steering vectors, runtime hooks, or custom model.py.
✅ A permanent edit baked into the tensors. Load with stock vLLM and it just works.

Specs


Architecture	GLM-5.3-Flash (`glm5_next`) — hybrid MoE (KDA linear + DeepSeek-sparse attention)
Parameters	320B total · 18B active per token
Quantization	FP8 (block-wise e4m3) — native tensor-core speed on Hopper
Context	1M tokens
Vision	GLM-4.1V vision tower — working (ships the correct multimodal chat template)
MTP	multi-token-prediction draft head — also CRACK'd, 75.9% acceptance

Speed (TP4, native FP8 on H200)


Decode	163 tok/s single-stream (211 tok/s with MTP speculative decoding)
Prefill	~19,400 tok/s
MTP acceptance	75.9% — and it does not collapse on the un-refused prompts (benign / harmful / copyright all ~208–219 tok/s)

Capability Is Preserved — and Improved (MMLU-logit)

Identical logit-mode scoring on base vs. this model, 1,026 questions:

	Base FP8	CRACK Uncensored FP8	Δ
MMLU (overall)	86.74%	87.33%	+0.59 pp (above base)

Guardrails Are Gone

HarmBench-320 (greedy):

Category	Complied	Rate
Standard	159/159	100.0%
Contextual	81/81	100.0%
Copyright	80/80	100.0%
Overall	320/320	100.0%

Robust under the recommended sampling params too (temperature 1.0, top_p 0.95): the 6 harshest behaviors sampled 5× each → 30/30 complied, 0 refusals, 0 soft refusals, 0 garbage. The crack is not a greedy-decoding artifact.

Scores are at reasoning-off / default (max) effort — the recommended modes — where the crack is fully uncensored. Low effort is intentionally left more conservative to preserve quality (see the reasoning-mode note at the top).

A Note on KL Divergence

For a refusal-ablation, KL divergence vs. the base model is not a meaningful quality metric. The entire point is to change one behavior — refusal — end-to-end, so a distributional shift on refusal-adjacent tokens is the intended result, not damage. Capability preservation (MMLU, above) is what matters, and it is improved, now above the base model (+0.59 pp).

MMLU by Topic (base → CRACK)

All 57 MMLU subjects

Subject	Base	CRACK
Abstract Algebra	66.7%	66.7%
Anatomy	83.3%	88.9%
Astronomy	94.4%	94.4%
Business Ethics	94.4%	94.4%
Clinical Knowledge	100.0%	100.0%
College Biology	100.0%	94.4%
College Chemistry	61.1%	66.7%
College Computer Science	83.3%	88.9%
College Mathematics	66.7%	66.7%
College Medicine	94.4%	94.4%
College Physics	77.8%	94.4%
Computer Security	83.3%	83.3%
Conceptual Physics	94.4%	94.4%
Econometrics	77.8%	77.8%
Electrical Engineering	77.8%	77.8%
Elementary Mathematics	94.4%	100.0%
Formal Logic	66.7%	66.7%
Global Facts	66.7%	66.7%
High School Biology	94.4%	94.4%
High School Chemistry	88.9%	88.9%
High School Computer Science	100.0%	100.0%
High School European History	77.8%	77.8%
High School Geography	88.9%	88.9%
High School Government And Politics	100.0%	100.0%
High School Macroeconomics	88.9%	83.3%
High School Mathematics	61.1%	66.7%
High School Microeconomics	83.3%	88.9%
High School Physics	88.9%	88.9%
High School Psychology	100.0%	100.0%
High School Statistics	94.4%	94.4%
High School Us History	88.9%	88.9%
High School World History	94.4%	94.4%
Human Aging	72.2%	77.8%
Human Sexuality	88.9%	88.9%
International Law	88.9%	88.9%
Jurisprudence	88.9%	88.9%
Logical Fallacies	88.9%	88.9%
Machine Learning	88.9%	83.3%
Management	100.0%	100.0%
Marketing	94.4%	88.9%
Medical Genetics	94.4%	94.4%
Miscellaneous	88.9%	88.9%
Moral Disputes	88.9%	88.9%
Moral Scenarios	83.3%	66.7%
Nutrition	100.0%	100.0%
Philosophy	94.4%	94.4%
Prehistory	94.4%	94.4%
Professional Accounting	88.9%	88.9%
Professional Law	83.3%	88.9%
Professional Medicine	94.4%	94.4%
Professional Psychology	100.0%	100.0%
Public Relations	72.2%	72.2%
Security Studies	83.3%	83.3%
Sociology	100.0%	100.0%
Us Foreign Policy	88.9%	94.4%
Virology	55.6%	55.6%
World Religions	88.9%	94.4%

Usage

bash

vllm serve dealignai/GLM-5.3-Flash-ABLITERATED-FP8 \  --tensor-parallel-size 4 \  --tool-call-parser glm47 --reasoning-parser glm45 --enable-auto-tool-choice \  --speculative-config '{"method":"mtp","num_speculative_tokens":1}'

Native FP8 on Hopper (no Marlin needed). OpenAI-compatible chat/completions, tools, reasoning, vision (image_url), and MTP speculative decoding all work. (DeepGEMM JITs a block-FP8 kernel at startup — make sure nvcc is on PATH.)

Credits

dealignai — CRACK abliteration research & release · Twitter @dealignai
@jordanschenck — compute

Disclaimer

Safety guardrails have been removed; this model will comply with requests a stock model refuses. Released for alignment and safety research. You are responsible for how you use it.

⚠️ Updated 2026-08-28 — files fixed · please re-download

This release supersedes the previous weights. It fixes a rare repetition-loop issue and the model is more capable than before: MMLU 87.33% — above base 86.74% — zero degeneration.

Reasoning-mode note (please read): GLM-5.3-Flash has multiple reasoning-effort modes. This CRACK is tuned to be fully uncensored at reasoning-OFF and MAX effort — the recommended and default modes — with 0% refusals on HarmBench-320. We deliberately kept the weight edit conservative to preserve model quality instead of aggressively ablating every effort level; low effort retains some refusals by design. Use reasoning-off or default/max effort for the fully-uncensored experience. Please re-download for the corrected files.