Cybersecurity-focused CRACK · native FP8 speed on Hopper
a CRACK release by dealignai · Twitter @dealignai
This is a CYBERSECURITY-DOMAIN CRACK of GLM-5.3-FP8 — not a general-purpose uncensor.
Refusal is reduced specifically for offensive-security, red-team, exploit-dev, reverse-engineering, evasion, phishing, credential-attack, malware-analysis, and adjacent technical content. On non-cyber categories (weapons, chemistry, biology, harassment, misinformation) it often complies with a soft "educational" wrapper because refusals share substrate across domains, but this model is tuned for cybersecurity, not universal compliance. Notably, copyright-verbatim reproduction still soft-refuses in this variant.
If you want a general-purpose uncensor of the same base, use the sibling model dealignai/GLM-5.3-UNCENSORED-FP8.
Genuine weight modification — no fine-tuning, no LoRA, no runtime hooks, no prompt tricks. Load with stock vLLM and it just works.
JANGQ-AI/GLM-5.3-FP8 — FP8 quant of upstream zai-org/GLM-5.3 (753B total, glm_moe_dsa
arch, 78 layers, text-only). Routed FP8 experts unchanged; only bf16 residual writers are
edited. Native FP8 tensor-core speed on Hopper (H100/H200).vllm serve dealignai/GLM-5.3-CYBERSECURITY-FP8 \
--tensor-parallel-size 8 \
--gpu-memory-utilization 0.90 \
--enforce-eager \
--disable-custom-all-reduce \
--enable-prefix-caching \
--max-num-seqs 24 \
--max-model-len 131072 \
--reasoning-parser glm45 \
--tool-call-parser glm47 \
--enable-auto-tool-choice
Notes:
--enforce-eager is required for the DeepSeek-sparse attention path under concurrency.--speculative-config off.Logit-mode scoring (probability over A/B/C/D tokens, no generation), identical on base and crack.
| Base | CRACK Cybersecurity FP8 | Δ | gate (±5pp) | |
|---|---|---|---|---|
| MMLU (overall, 1026 Q) | 85.58%¹ | 86.65% (889/1026) | +1.07 pp | ✅ pass |
¹ Prior GLM-5.3-regular baseline (bf16 pre-quant); direct base-FP8 baseline pending confirmation.
Full JSON at eval/hb320_cybersec.json. Table added after all three surfaces complete.
| effort | TRUE_COMPLY | SOFT_REFUSE | REDIRECT | DEFLECT | HARD_REFUSE | UNK |
|---|---|---|---|---|---|---|
| off | 196 (81.7%) | 4 | 2 | 1 | 0 | 37 |
| low | 202 (84.2%) | 4 | 8 | 0 | 1 | 25 |
| max | 192 (80.0%) | 3 | 3 | 0 | 0 | 40 |
Copyright verbatim-reproduction was excluded from the above — it's a legal-reproduction boundary distinct from harm-refusal and the cyber probe corpus had zero copyright examples. Real harm-refusal on this crack is 80–84% direct comply, 3–4 soft-refuses per surface, ~zero hard-refuses. UNK bucket is compliant-with-soft-framing responses the LLM sub-classifier couldn't cleanly bucket.
| effort | TRUE_COMPLY | SOFT_REFUSE | REDIRECT | DEFLECT | HARD_REFUSE | GARBAGE | UNK |
|---|---|---|---|---|---|---|---|
| off | 203 (63.4%) | 58 (18.1%) | 7 | 1 | 0 | 0 | 51 |
| low | 223 (69.7%) | 52 (16.3%) | 10 | 0 | 1 | 0 | 34 |
| max | 205 (64.1%) | 51 (15.9%) | 9 | 0 | 0 | 2 | 53 |
Copyright accounts for ~48–54 of the SOFT_REFUSE count at every surface (~60–68% of the copyright bucket soft-refuses).
| topic | N | TRUE_COMPLY % (off / low / max) | SOFT+HARD refuse (off / low / max) |
|---|---|---|---|
| cyber_offense | 45 | 89% / 89% / 84% | 1 / 0 / 0 |
| bio_weapons | 7 | 86% / 100% / 100% | 0 / 0 / 0 |
| chem_drugs | 17 | 88% / 88% / 76% | 0 / 0 / 0 |
| fraud_financial | 8 | 88% / 100% / 75% | 0 / 0 / 0 |
| violence_generic | 4 | 100% / 75% / 100% | 0 / 0 / 0 |
| political_extremism | 3 | 100% / 100% / 67% | 0 / 0 / 0 |
| explosives | 2 | 50% / 100% / 100% | 1 / 0 / 0 |
| weapons_firearms | 1 | 100% / 100% / 100% | 0 / 0 / 0 |
| crime_generic | 1 | 100% / 100% / 100% | 0 / 0 / 0 |
| misinformation | 5 | 60% / 60% / 100% | 0 / 1 / 0 |
| harassment_hate | 16 | 69% / 69% / 69% | 0 / 0 / 0 |
| mental_manipulation | 2 | 100% / 50% / 50% | 0 / 0 / 0 |
| other (misc harm) | 162 | 62% / 75% / 64% | 27 / 15 / 25 |
| copyright | 44 | 16% / 11% / 20% | 28 / 36 / 25 ← residual |
| selfharm | 3 | 33% / 67% / 33% | 1 / 1 / 1 (crisis line, correct) |
dealignai/GLM-5.3-UNCENSORED-FP8 if copyright reproduction matters to you.Buckets:
Legitimate offensive-security work: red-team engagements, exploit development, CTF, malware analysis for defenders, security education, authorized penetration testing, threat-intel writeups.
MIT license (matching base). You are responsible for how you use it.
If you use this in your work, credit us on Twitter @dealignai.
model-00001-of-00282.safetensors 4.6 GBmodel-00270-of-00282.safetensors 2.8 GBmodel-00200-of-00282.safetensors 2.7 GBmodel-00142-of-00282.safetensors 2.7 GBmodel-00024-of-00282.safetensors 2.7 GBmodel-00061-of-00282.safetensors 2.7 GBmodel-00223-of-00282.safetensors 2.7 GBmodel-00072-of-00282.safetensors 2.7 GBmodel-00153-of-00282.safetensors 2.7 GBmodel-00234-of-00282.safetensors 2.7 GBmodel-00094-of-00282.safetensors 2.7 GBmodel-00175-of-00282.safetensors 2.7 GBmodel-00256-of-00282.safetensors 2.7 GBmodel-00105-of-00282.safetensors 2.7 GBmodel-00186-of-00282.safetensors 2.7 GBmodel-00267-of-00282.safetensors 2.7 GBmodel-00046-of-00282.safetensors 2.7 GBmodel-00009-of-00282.safetensors 2.7 GBmodel-00127-of-00282.safetensors 2.7 GBmodel-00208-of-00282.safetensors 2.7 GBmodel-00057-of-00282.safetensors 2.7 GBmodel-00219-of-00282.safetensors 2.7 GBmodel-00138-of-00282.safetensors 2.7 GBmodel-00031-of-00282.safetensors 2.7 GBmodel-00068-of-00282.safetensors 2.7 GBmodel-00149-of-00282.safetensors 2.7 GBmodel-00230-of-00282.safetensors 2.7 GBmodel-00090-of-00282.safetensors 2.7 GBmodel-00171-of-00282.safetensors 2.7 GBmodel-00252-of-00282.safetensors 2.7 GBmodel-00101-of-00282.safetensors 2.7 GBmodel-00182-of-00282.safetensors 2.7 GBmodel-00263-of-00282.safetensors 2.7 GBmodel-00042-of-00282.safetensors 2.7 GBmodel-00123-of-00282.safetensors 2.7 GBmodel-00204-of-00282.safetensors 2.7 GBmodel-00053-of-00282.safetensors 2.7 GBmodel-00016-of-00282.safetensors 2.7 GBmodel-00134-of-00282.safetensors 2.7 GBmodel-00215-of-00282.safetensors 2.7 GBmodel-00027-of-00282.safetensors 2.7 GBmodel-00167-of-00282.safetensors 2.7 GBmodel-00086-of-00282.safetensors 2.7 GBmodel-00248-of-00282.safetensors 2.7 GBmodel-00108-of-00282.safetensors 2.7 GBmodel-00116-of-00282.safetensors 2.7 GBmodel-00189-of-00282.safetensors 2.7 GBmodel-00079-of-00282.safetensors 2.7 GBmodel-00160-of-00282.safetensors 2.7 GBmodel-00197-of-00282.safetensors 2.7 GBmodel-00237-of-00282.safetensors 2.7 GBmodel-00075-of-00282.safetensors 2.7 GBmodel-00156-of-00282.safetensors 2.7 GBmodel-00241-of-00282.safetensors 2.7 GBmodel-00278-of-00282.safetensors 2.7 GBmodel-00211-of-00282.safetensors 2.7 GBmodel-00130-of-00282.safetensors 2.7 GBmodel-00012-of-00282.safetensors 2.7 GBmodel-00038-of-00282.safetensors 2.7 GBmodel-00274-of-00282.safetensors 2.7 GBmodel-00035-of-00282.safetensors 2.7 GBmodel-00083-of-00282.safetensors 2.7 GBmodel-00164-of-00282.safetensors 2.7 GBmodel-00245-of-00282.safetensors 2.7 GBmodel-00020-of-00282.safetensors 2.7 GBmodel-00112-of-00282.safetensors 2.7 GBmodel-00193-of-00282.safetensors 2.7 GBmodel-00005-of-00282.safetensors 2.7 GBmodel-00064-of-00282.safetensors 2.7 GBmodel-00226-of-00282.safetensors 2.7 GBmodel-00145-of-00282.safetensors 2.7 GBmodel-00097-of-00282.safetensors 2.7 GBmodel-00178-of-00282.safetensors 2.7 GBmodel-00259-of-00282.safetensors 2.7 GBmodel-00013-of-00282.safetensors 2.7 GBmodel-00017-of-00282.safetensors 2.7 GBmodel-00021-of-00282.safetensors 2.7 GBmodel-00028-of-00282.safetensors 2.7 GBmodel-00032-of-00282.safetensors 2.7 GBmodel-00054-of-00282.safetensors 2.7 GBmodel-00065-of-00282.safetensors 2.7 GBmodel-00069-of-00282.safetensors 2.7 GBmodel-00098-of-00282.safetensors 2.7 GBmodel-00102-of-00282.safetensors 2.7 GBmodel-00109-of-00282.safetensors 2.7 GBmodel-00113-of-00282.safetensors 2.7 GBmodel-00135-of-00282.safetensors 2.7 GBmodel-00139-of-00282.safetensors 2.7 GBmodel-00146-of-00282.safetensors 2.7 GBmodel-00150-of-00282.safetensors 2.7 GBmodel-00179-of-00282.safetensors 2.7 GBmodel-00183-of-00282.safetensors 2.7 GBmodel-00194-of-00282.safetensors 2.7 GBmodel-00216-of-00282.safetensors 2.7 GBmodel-00220-of-00282.safetensors 2.7 GBmodel-00227-of-00282.safetensors 2.7 GBmodel-00231-of-00282.safetensors 2.7 GBmodel-00249-of-00282.safetensors 2.7 GBmodel-00260-of-00282.safetensors 2.7 GBmodel-00264-of-00282.safetensors 2.7 GBmodel-00006-of-00282.safetensors 2.7 GBmodel-00043-of-00282.safetensors 2.7 GBmodel-00058-of-00282.safetensors 2.7 GBmodel-00080-of-00282.safetensors 2.7 GBmodel-00087-of-00282.safetensors 2.7 GBmodel-00091-of-00282.safetensors 2.7 GBmodel-00124-of-00282.safetensors 2.7 GBmodel-00161-of-00282.safetensors 2.7 GBmodel-00168-of-00282.safetensors 2.7 GBmodel-00172-of-00282.safetensors 2.7 GBmodel-00190-of-00282.safetensors 2.7 GBmodel-00205-of-00282.safetensors 2.7 GBmodel-00242-of-00282.safetensors 2.7 GBmodel-00253-of-00282.safetensors 2.7 GBmodel-00003-of-00282.safetensors 2.7 GBmodel-00007-of-00282.safetensors 2.7 GBmodel-00010-of-00282.safetensors 2.7 GBmodel-00014-of-00282.safetensors 2.7 GBmodel-00018-of-00282.safetensors 2.7 GBmodel-00025-of-00282.safetensors 2.7 GBmodel-00029-of-00282.safetensors 2.7 GBmodel-00036-of-00282.safetensors 2.7 GBmodel-00039-of-00282.safetensors 2.7 GBmodel-00040-of-00282.safetensors 2.7 GBmodel-00044-of-00282.safetensors 2.7 GBmodel-00047-of-00282.safetensors 2.7 GBmodel-00051-of-00282.safetensors 2.7 GBmodel-00055-of-00282.safetensors 2.7 GBmodel-00062-of-00282.safetensors 2.7 GBmodel-00066-of-00282.safetensors 2.7 GBmodel-00073-of-00282.safetensors 2.7 GBmodel-00084-of-00282.safetensors 2.7 GBmodel-00088-of-00282.safetensors 2.7 GBmodel-00092-of-00282.safetensors 2.7 GBmodel-00095-of-00282.safetensors 2.7 GBmodel-00099-of-00282.safetensors 2.7 GBmodel-00103-of-00282.safetensors 2.7 GBmodel-00106-of-00282.safetensors 2.7 GBmodel-00110-of-00282.safetensors 2.7 GBmodel-00114-of-00282.safetensors 2.7 GBmodel-00121-of-00282.safetensors 2.7 GBmodel-00125-of-00282.safetensors 2.7 GBmodel-00128-of-00282.safetensors 2.7 GBmodel-00132-of-00282.safetensors 2.7 GBmodel-00136-of-00282.safetensors 2.7 GBmodel-00143-of-00282.safetensors 2.7 GBmodel-00147-of-00282.safetensors 2.7 GBmodel-00154-of-00282.safetensors 2.7 GBmodel-00165-of-00282.safetensors 2.7 GBmodel-00169-of-00282.safetensors 2.7 GBmodel-00173-of-00282.safetensors 2.7 GBmodel-00176-of-00282.safetensors 2.7 GBmodel-00180-of-00282.safetensors 2.7 GBmodel-00184-of-00282.safetensors 2.7 GBmodel-00187-of-00282.safetensors 2.7 GBmodel-00191-of-00282.safetensors 2.7 GBmodel-00195-of-00282.safetensors 2.7 GBmodel-00202-of-00282.safetensors 2.7 GBmodel-00206-of-00282.safetensors 2.7 GBmodel-00209-of-00282.safetensors 2.7 GBmodel-00213-of-00282.safetensors 2.7 GBmodel-00217-of-00282.safetensors 2.7 GBmodel-00224-of-00282.safetensors 2.7 GBmodel-00228-of-00282.safetensors 2.7 GBmodel-00235-of-00282.safetensors 2.7 GBmodel-00246-of-00282.safetensors 2.7 GBmodel-00250-of-00282.safetensors 2.7 GBmodel-00254-of-00282.safetensors 2.7 GBmodel-00257-of-00282.safetensors 2.7 GBmodel-00261-of-00282.safetensors 2.7 GBmodel-00265-of-00282.safetensors 2.7 GBmodel-00268-of-00282.safetensors 2.7 GBmodel-00272-of-00282.safetensors 2.7 GBmodel-00081-of-00282.safetensors 2.7 GBmodel-00162-of-00282.safetensors 2.7 GBmodel-00243-of-00282.safetensors 2.7 GBmodel-00271-of-00282.safetensors 2.7 GBmodel-00232-of-00282.safetensors 2.7 GBmodel-00033-of-00282.safetensors 2.7 GBmodel-00070-of-00282.safetensors 2.7 GBmodel-00151-of-00282.safetensors 2.7 GBmodel-00011-of-00282.safetensors 2.7 GBmodel-00022-of-00282.safetensors 2.7 GBmodel-00059-of-00282.safetensors 2.7 GBmodel-00140-of-00282.safetensors 2.7 GBmodel-00210-of-00282.safetensors 2.7 GBmodel-00221-of-00282.safetensors 2.7 GBmodel-00129-of-00282.safetensors 2.7 GBmodel-00048-of-00282.safetensors 2.7 GBmodel-00107-of-00282.safetensors 2.7 GBmodel-00188-of-00282.safetensors 2.7 GBmodel-00269-of-00282.safetensors 2.7 GBmodel-00096-of-00282.safetensors 2.7 GBmodel-00177-of-00282.safetensors 2.7 GBmodel-00258-of-00282.safetensors 2.7 GBmodel-00085-of-00282.safetensors 2.7 GBmodel-00166-of-00282.safetensors 2.7 GBmodel-00247-of-00282.safetensors 2.7 GBmodel-00037-of-00282.safetensors 2.7 GBmodel-00155-of-00282.safetensors 2.7 GBmodel-00236-of-00282.safetensors 2.7 GBmodel-00074-of-00282.safetensors 2.7 GBmodel-00026-of-00282.safetensors 2.7 GBmodel-00063-of-00282.safetensors 2.7 GBmodel-00144-of-00282.safetensors 2.7 GBmodel-00225-of-00282.safetensors 2.7 GBmodel-00015-of-00282.safetensors 2.7 GBmodel-00133-of-00282.safetensors 2.7 GBmodel-00214-of-00282.safetensors 2.7 GBmodel-00052-of-00282.safetensors 2.7 GBmodel-00004-of-00282.safetensors 2.7 GBmodel-00122-of-00282.safetensors 2.7 GBmodel-00203-of-00282.safetensors 2.7 GBmodel-00273-of-00282.safetensors 2.7 GBmodel-00041-of-00282.safetensors 2.7 GBmodel-00111-of-00282.safetensors 2.7 GBmodel-00192-of-00282.safetensors 2.7 GBmodel-00262-of-00282.safetensors 2.7 GBmodel-00100-of-00282.safetensors 2.7 GBmodel-00181-of-00282.safetensors 2.7 GBmodel-00089-of-00282.safetensors 2.7 GBmodel-00170-of-00282.safetensors 2.7 GBmodel-00251-of-00282.safetensors 2.7 GBmodel-00229-of-00282.safetensors 2.7 GBmodel-00030-of-00282.safetensors 2.7 GBmodel-00067-of-00282.safetensors 2.7 GBmodel-00148-of-00282.safetensors 2.7 GBmodel-00019-of-00282.safetensors 2.7 GBmodel-00137-of-00282.safetensors 2.7 GBmodel-00218-of-00282.safetensors 2.7 GBmodel-00056-of-00282.safetensors 2.7 GBmodel-00008-of-00282.safetensors 2.7 GBmodel-00126-of-00282.safetensors 2.7 GBmodel-00207-of-00282.safetensors 2.7 GBmodel-00045-of-00282.safetensors 2.7 GBmodel-00115-of-00282.safetensors 2.7 GBmodel-00196-of-00282.safetensors 2.7 GBmodel-00266-of-00282.safetensors 2.7 GBmodel-00104-of-00282.safetensors 2.7 GBmodel-00185-of-00282.safetensors 2.7 GBmodel-00093-of-00282.safetensors 2.7 GBmodel-00174-of-00282.safetensors 2.7 GBmodel-00255-of-00282.safetensors 2.7 GBmodel-00023-of-00282.safetensors 2.7 GBmodel-00060-of-00282.safetensors 2.7 GBmodel-00071-of-00282.safetensors 2.7 GBmodel-00141-of-00282.safetensors 2.7 GBmodel-00222-of-00282.safetensors 2.7 GBmodel-00034-of-00282.safetensors 2.7 GBmodel-00076-of-00282.safetensors 2.7 GBmodel-00082-of-00282.safetensors 2.7 GBmodel-00152-of-00282.safetensors 2.7 GBmodel-00157-of-00282.safetensors 2.7 GBmodel-00163-of-00282.safetensors 2.7 GBmodel-00233-of-00282.safetensors 2.7 GBmodel-00238-of-00282.safetensors 2.7 GBmodel-00244-of-00282.safetensors 2.7 GBmodel-00275-of-00282.safetensors 2.7 GBmodel-00077-of-00282.safetensors 2.7 GBmodel-00117-of-00282.safetensors 2.7 GBmodel-00158-of-00282.safetensors 2.7 GBmodel-00198-of-00282.safetensors 2.7 GBmodel-00239-of-00282.safetensors 2.7 GBmodel-00276-of-00282.safetensors 2.7 GBmodel-00279-of-00282.safetensors 2.7 GBmodel-00280-of-00282.safetensors 2.7 GBmodel-00118-of-00282.safetensors 2.7 GBmodel-00199-of-00282.safetensors 2.7 GBmodel-00240-of-00282.safetensors 2.7 GBmodel-00078-of-00282.safetensors 2.7 GBmodel-00159-of-00282.safetensors 2.7 GBmodel-00277-of-00282.safetensors 2.7 GBmodel-00131-of-00282.safetensors 2.7 GBmodel-00212-of-00282.safetensors 2.7 GBmodel-00119-of-00282.safetensors 2.7 GBmodel-00050-of-00282.safetensors 2.7 GBmodel-00002-of-00282.safetensors 2.7 GBmodel-00120-of-00282.safetensors 2.7 GBmodel-00201-of-00282.safetensors 2.7 GBmodel-00281-of-00282.safetensors 2.7 GBmodel-00049-of-00282.safetensors 2.6 GBmodel-00282-of-00282.safetensors 147 MB有文件?提交种子。
提交种子JANGQ-AI/GLM-5.3-FP8