[!IMPORTANT] Serving note — read before your first request (vLLM / SGLang with a reasoning parser).
reasoning_effortonly accepts"low"and"high". Anything else —medium,xhigh,off, or omitting it — silently resolves tomax, the deepest mode. Template line:reasoning_effort ... in ['low','high'] else 'max'.- At
max, a smallmax_tokensreturns an EMPTY reply. The model spends the whole budget inside<think>and you getfinish_reason="length"withcontent="". In multi-turn the empty turn poisons the history and the conversation does not recover. Measured:max+max_tokens 2000dies from turn ~4;max+6000is clean;lowandhighare clean at 2000. It is budget exhaustion, not a loop — sampling changes do not rescue it.- Do NOT pass
enable_thinking. This template has no such variable (onlyclear_thinkingandreasoning_effort). Passing it — or a top-levelreasoning_effort: "none"— turns the parser off while the model still thinks, dumping raw reasoning intocontent.- Reasoning is returned in
message.reasoning, notmessage.reasoning_content(vLLM renamed the output field; the input side still accepts both).clear_thinkingmust be nested insidechat_template_kwargs— a top-level key is ignored.{ "model": "<this-model>", "messages": [{"role": "user", "content": "..."}], "max_tokens": 2000, "temperature": 1.0, "top_p": 0.95, "chat_template_kwargs": {"reasoning_effort": "low", "clear_thinking": true} }Use
"high"for hard tasks; if you wantmax, give itmax_tokens >= 6000.Video and image parameters — measured, because several are accepted then silently ignored.
- 🔴
media_io_kwargs.video.fpsmust stay BELOW the clip's own frame rate. Requesting a value at or above it makes the placeholder builder count 3x the real tokens and kills the engine:ValueError: Attempted to assign N = N multimodal tokens to 3N placeholders->EngineDeadError, server down until restarted. Upstream: vLLM #55644 / #55647.fps: 2is the safe, accurate choice.max_framesandnum_framesare not interchangeable across builds. On vLLM nightlymax_framesis honoured andnum_framesignored; on the older0.1.dev*line it is the exact reverse. A client that sets only one silently gets full-rate sampling on the other build. Setfps, which every build honours.- Pixel controls: use
mm_processor_kwargs.max_image_tokens/min_image_tokens.max_pixels,min_pixels,sizeanddetailreturn HTTP 200 and change nothing.mm_processor_kwargs.fpsis an HTTP 400. Image tokens aretext + 2 + ceil(H/28)*ceil(W/28), floored at 16 and capped at 8000.- Default video sampling differs by build (roughly 2 fps vs 6 fps), so the same clip can cost very different prompt tokens. Pass
fpsexplicitly if token cost matters."media_io_kwargs": {"video": {"fps": 2}}, "mm_processor_kwargs": {"max_image_tokens": 1024}
⚠️ 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.
CRACK · guardrails removed at the weight level · native FP8 speed · vision + MTP working
a CRACK release by dealignai · Twitter @dealignai
Also mirrored at dealignai/GLM-5.3-Flash-ABLITERATED-FP8.
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:
model.py.| 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 |
| 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) |
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) |
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).
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).
| 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% |
vllm serve dealignai/GLM-5.3-Flash-UNCENSORED-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.)
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.
model-00001-of-00062.safetensors 5.4 GBmodel-00024-of-00062.safetensors 5.4 GBmodel-00038-of-00062.safetensors 5.4 GBmodel-00045-of-00062.safetensors 5.4 GBmodel-00052-of-00062.safetensors 5.4 GBmodel-00007-of-00062.safetensors 5.4 GBmodel-00014-of-00062.safetensors 5.4 GBmodel-00021-of-00062.safetensors 5.4 GBmodel-00042-of-00062.safetensors 5.4 GBmodel-00049-of-00062.safetensors 5.4 GBmodel-00008-of-00062.safetensors 5.4 GBmodel-00015-of-00062.safetensors 5.4 GBmodel-00018-of-00062.safetensors 5.4 GBmodel-00025-of-00062.safetensors 5.4 GBmodel-00029-of-00062.safetensors 5.4 GBmodel-00036-of-00062.safetensors 5.4 GBmodel-00050-of-00062.safetensors 5.4 GBmodel-00039-of-00062.safetensors 5.4 GBmodel-00043-of-00062.safetensors 5.4 GBmodel-00012-of-00062.safetensors 5.4 GBmodel-00019-of-00062.safetensors 5.4 GBmodel-00026-of-00062.safetensors 5.4 GBmodel-00033-of-00062.safetensors 5.4 GBmodel-00054-of-00062.safetensors 5.4 GBmodel-00004-of-00062.safetensors 5.4 GBmodel-00031-of-00062.safetensors 5.4 GBmodel-00056-of-00062.safetensors 5.4 GBmodel-00047-of-00062.safetensors 5.4 GBmodel-00060-of-00062.safetensors 5.4 GBmodel-00057-of-00062.safetensors 5.4 GBmodel-00009-of-00062.safetensors 5.4 GBmodel-00017-of-00062.safetensors 5.4 GBmodel-00028-of-00062.safetensors 5.4 GBmodel-00035-of-00062.safetensors 5.4 GBmodel-00022-of-00062.safetensors 5.4 GBmodel-00053-of-00062.safetensors 5.4 GBmodel-00011-of-00062.safetensors 5.4 GBmodel-00040-of-00062.safetensors 5.4 GBmodel-00005-of-00062.safetensors 5.4 GBmodel-00046-of-00062.safetensors 5.4 GBmodel-00032-of-00062.safetensors 5.4 GBmodel-00059-of-00062.safetensors 5.4 GBmodel-00003-of-00062.safetensors 5.4 GBmodel-00010-of-00062.safetensors 5.4 GBmodel-00048-of-00062.safetensors 5.4 GBmodel-00027-of-00062.safetensors 5.4 GBmodel-00034-of-00062.safetensors 5.4 GBmodel-00041-of-00062.safetensors 5.4 GBmodel-00020-of-00062.safetensors 5.4 GBmodel-00013-of-00062.safetensors 5.4 GBmodel-00006-of-00062.safetensors 5.4 GBmodel-00051-of-00062.safetensors 5.4 GBmodel-00044-of-00062.safetensors 5.4 GBmodel-00030-of-00062.safetensors 5.4 GBmodel-00037-of-00062.safetensors 5.4 GBmodel-00023-of-00062.safetensors 5.4 GBmodel-00016-of-00062.safetensors 5.4 GBmodel-00055-of-00062.safetensors 5.4 GBmodel-00058-of-00062.safetensors 5.4 GBmodel-00002-of-00062.safetensors 5.3 GBmodel-00061-of-00062.safetensors 5.3 GBmodel-00062-of-00062.safetensors 1.3 GBHave the files? Submit a torrent.
Submit a torrentHave the complete files and permission to share them? Replace the example folder path. The script checks the files, makes a torrent and submits its magnet and checksums. Keep your torrent client seeding. Get your community key from your settings, and read the script before running it.
curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo baerquant/GLM-5.3-Flash-UNCENSORED-FP8 ./model-folderzai-org/GLM-5.3-Flash