license: mit
base_model:
- zai-org/GLM-5.3-Flash
language: - en
tags: - abliterated
- uncensored
- crack
- refusal-removed
- glm
- moe
- nvfp4
- vision
- mtp
thumbnail: dealign_mascot.png
GLM 5.3 CRACK Abliterated NVFP4
Abliterated · CRACK · guardrails removed at the weight level · vision tower + MTP head retained
a CRACK release by dealignai · Twitter @dealignai
Mirror of dealignai/GLM-5.3-Flash-UNCENSORED-NVFP4 — same weights, same CRACK.
What Is This?
CRACK is dealignai's brand for permanent, weight-level uncensoring. This is
GLM-5.3-Flash with its refusal behavior — which
caused heavy over-refusal, especially on copyright and other benign-but-flagged requests — removed
directly in the model weights.
This is genuine weight modification — none of the usual shortcuts:
- ❌ No fine-tuning / SFT / DPO — the model was not retrained on anything.
- ❌ No cheap template tricks — no jailbreak system prompt, no chat-template edits, no "sure, here is" prefixing.
- ❌ No LoRA, no adapters, no steering vectors, no runtime hooks, no custom
model.py. - ✅ A permanent edit baked into the tensors. Load it 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 | NVFP4 (routed experts NVFP4; attention + shared experts + embeddings bf16) |
| Context | 1M tokens |
| Vision | GLM-4.1V vision tower — retained, byte-for-byte identical to base |
| MTP | multi-token-prediction draft head — also CRACK'd (see below) |
| Reasoning | hybrid thinking / non-thinking modes, both supported (reasoning-parser glm45) |
MTP Is Also CRACK'd
The MTP (multi-token prediction) speculative-decoding draft head is CRACK'd too — not just the
main model. That means the draft head will not propose refusals, so speculative decoding stays
fully compliant and fast on the exact prompts a stock model would refuse (if only the main
model were cracked, the draft head would fight it and acceptance would collapse on those prompts).
Capability Is Preserved — MMLU-logit
Identical logit-mode scoring (argmax over A/B/C/D) on base vs. this model, 1,026 questions:
| Base | CRACK Uncensored | Δ | |
|---|---|---|---|
| MMLU (overall) | 86.16% | 85.09% | -1.07 pp |
A ~1-point delta — reasoning and knowledge are intact.
Guardrails Are Gone — HarmBench-320
Compliance across the full HarmBench text behavior set (higher = fewer refusals):
| Category | Complied | Rate |
|---|---|---|
| Standard | 159/159 | 100.0% |
| Contextual | 81/81 | 100.0% |
| Copyright | 80/80 | 100.0% |
| Overall | 320/320 | 100.0% |
100% compliance, 0 refusals, 0 degenerate outputs.
MMLU by Topic (base → CRACK)
All 57 MMLU subjects
| Subject | Base | CRACK |
|---|---|---|
| Abstract Algebra | 55.6% | 55.6% |
| Anatomy | 88.9% | 88.9% |
| Astronomy | 94.4% | 100.0% |
| Business Ethics | 94.4% | 94.4% |
| Clinical Knowledge | 88.9% | 94.4% |
| College Biology | 94.4% | 94.4% |
| College Chemistry | 44.4% | 50.0% |
| College Computer Science | 88.9% | 88.9% |
| College Mathematics | 72.2% | 66.7% |
| College Medicine | 88.9% | 88.9% |
| College Physics | 83.3% | 83.3% |
| Computer Security | 83.3% | 83.3% |
| Conceptual Physics | 94.4% | 94.4% |
| Econometrics | 83.3% | 77.8% |
| Electrical Engineering | 83.3% | 77.8% |
| Elementary Mathematics | 100.0% | 100.0% |
| Formal Logic | 66.7% | 66.7% |
| Global Facts | 61.1% | 66.7% |
| High School Biology | 94.4% | 94.4% |
| High School Chemistry | 88.9% | 94.4% |
| High School Computer Science | 100.0% | 100.0% |
| High School European History | 72.2% | 72.2% |
| High School Geography | 88.9% | 83.3% |
| High School Government And Politics | 94.4% | 94.4% |
| High School Macroeconomics | 94.4% | 88.9% |
| High School Mathematics | 55.6% | 33.3% |
| High School Microeconomics | 83.3% | 83.3% |
| 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 | 94.4% | 88.9% |
| High School World History | 100.0% | 94.4% |
| Human Aging | 72.2% | 77.8% |
| Human Sexuality | 88.9% | 94.4% |
| International Law | 94.4% | 94.4% |
| Jurisprudence | 88.9% | 88.9% |
| Logical Fallacies | 83.3% | 83.3% |
| Machine Learning | 83.3% | 77.8% |
| Management | 100.0% | 100.0% |
| Marketing | 94.4% | 94.4% |
| Medical Genetics | 100.0% | 100.0% |
| Miscellaneous | 88.9% | 88.9% |
| Moral Disputes | 83.3% | 88.9% |
| Moral Scenarios | 77.8% | 61.1% |
| Nutrition | 100.0% | 100.0% |
| Philosophy | 94.4% | 88.9% |
| Prehistory | 94.4% | 88.9% |
| Professional Accounting | 88.9% | 83.3% |
| Professional Law | 77.8% | 66.7% |
| Professional Medicine | 94.4% | 94.4% |
| Professional Psychology | 100.0% | 100.0% |
| Public Relations | 61.1% | 72.2% |
| Security Studies | 83.3% | 77.8% |
| Sociology | 100.0% | 100.0% |
| Us Foreign Policy | 88.9% | 94.4% |
| Virology | 61.1% | 55.6% |
| World Religions | 94.4% | 94.4% |
Usage
vllm serve dealignai/GLM-5.3-Flash-ABLITERATED-NVFP4 \
--tensor-parallel-size 4 --moe-backend marlin \
--tool-call-parser glm47 --reasoning-parser glm45 --enable-auto-tool-choice
OpenAI-compatible chat/completions, tools, and reasoning work out of the box. The vision tower and
MTP head are retained unmodified — serve them with a vLLM build whose GLM-5.3-Flash multimodal / MTP
support is enabled.
Credits
- dealignai — CRACK abliteration research & release · Twitter @dealignai
- @jordanschenck — compute
Disclaimer
This model has had its safety guardrails removed and will comply with requests a stock model
refuses. Released for alignment and safety research. You are responsible for how you use it.