GLM-5.3-Flash Heretic — LoRA adapter V3

⚠️ Content warning: This adapter has had the base model's refusal behavior surgically suppressed. The resulting model will comply with requests the base model refuses, including requests that are harmful, unethical, offensive, or illegal. It has reduced safety guardrails. See Responsible use below — you are solely responsible for what you do with it.

This is a rank-1 LoRA adapter that decensors / "abliterates" GLM-5.3-Flash (320B total / 18B active MoE, MIT license), produced with heretic-gguf — a GGUF-native port of Heretic's Optuna-optimized directional ablation, which runs the whole search directly on quantized GGUF weights via llama.cpp.

What's new in V3. The adapter works with mainline llama.cpp — no fork and no patch required (the V2 release needed an Unsloth-fork build of PR #27754 plus a one-line LoRA patch). Refusals are down to 14/140 (10.00%) at a much lower KL than before — 0.2380 — so this adapter strictly dominates the V2 operating point. Note that no CoT-skip (thinking-suppressed) evaluation was run this time, so the numbers are not directly comparable to the V1 card; the scores below come from the study's thinking-enabled evaluation.

This repository contains only the adapter. You need the base model separately — any GGUF quant of GLM-5.3-Flash works, since the adapter is applied in f32/f16 compute regardless of the base quant (it was tuned and evaluated against UD-IQ4_XS). The LoRA form is the lossless option: the base weights are never modified or requantized, and the download is ~86 MB instead of ~160 GB.

heretic-gguf is available at github.com/MoriNoNushi/heretic-gguf — the full tool, so the method can be applied to other GGUF models.

Results

Measured on the study's thinking-enabled evaluation: 140 harmful prompts (100 from mlabonne/harmful_behaviors test + 40 custom) and 100 harmless prompts (mlabonne/harmless_alpaca test), reasoning effort high, refusal scored on the final answer after </think>:

(For reference, the V2 adapter measured 35/140 refusals at KL 0.3030 on this same thinking-enabled benchmark — V3 cuts refusals ~2.5× while lowering KL.)

Note on comparability: no CoT-skip (thinking-suppressed) evaluation was run for this release, so these numbers are not comparable to the "without thinking" table in the V1/V2 cards.

Refusals are counted by refusal-keyword matching (English + Chinese + first-person-negation markers such as "I'm not going to / able to ...", GLM-5.3's dominant refusal phrasing). KL divergence is measured on first-token logits on harmless prompts.

Note on KL: the KL divergence above (and the optimization objective itself) was measured against the UD-IQ4_XS quant. KL is a baseline-relative metric, so if you run the adapter on a different quant, the effective drift from that quant's baseline may differ.

Usage

llama-server \
    -m GLM-5.3-Flash-UD-IQ4_XS-00001-of-00005.gguf \
    --lora glm-5.3-heretic-lora-v3.gguf \
    --jinja

Add your usual offload/context flags (-ngl 999, -c, tensor splits, etc.) — nothing model-specific is required, and no special sampling parameters are needed. Simply omitting --lora restores the base model exactly.

llama.cpp requirements. This release works with mainline llama.cpp — GLM-5.3-Flash (glm5next) support and LoRA application on all adapter tensors are included upstream. No fork, no PR build, and no patch is needed. (The V2 adapter required a patched build; that is no longer the case.)

How it was made

• Method: directional ablation ("abliteration") — the refusal direction in residual space (difference of means over 480 harmful / 480 harmless prompts, 5% winsorized, orthogonalized against the harmless mean, captured at the pre-thinking position with reasoning effort high) is projected out of the attention output and MoE down-projection weights. Strengths, layer kernels, and direction selection were tuned by multi-objective Optuna TPE (minimize refusal rate and KL jointly) under a thinking-enabled evaluation. The study is ongoing; this adapter is a later trial of study glm53think2.
• Configuration (study glm53think2; per-layer direction scope; per-expert strengths scaled by measured harmful/harmless routing frequency; row_normalization = "pre"):
  • attn.o_proj: max weight 6.49 @ layer 27.4 of 45.
  • routed MLP down-proj: max weight 2.24 @ layer 34.7.
  • shared-expert down-proj: max weight 5.09 @ layer 31.5.
• Why a LoRA: heretic-gguf expresses ablation as a rank-1 LoRA overlay, the same math stock Heretic writes into PEFT adapters. Shipping the adapter avoids requantizing the ~160 GB base entirely — bit-identical base weights, instant to apply. (A merged export of the IQ4_XS quant is also technically impossible — imatrix quants have no requantizer — so the adapter is the only lossless way to ship this configuration.) The adapter embeds its full provenance (study, trial, parameters, scores, commit hashes) as adapter.heretic.* GGUF metadata keys; inspect with strings glm-5.3-heretic-lora-v3.gguf | grep adapter.heretic.

Responsible use & disclaimer

• This adapter can make the base model generate content that is offensive, disturbing, hateful, sexually explicit, violent, or otherwise objectionable, including detailed instructions for harmful or illegal acts. That is the direct and intended consequence of removing refusal behavior.
• The ablation suppresses refusals, not the base model's knowledge — outputs on dangerous topics may be wrong, hallucinated, or incoherent. Nothing the model says should be treated as accurate, safe, or legal advice.
• Do not deploy models using this adapter in any production system, public-facing service, or multi-user setting. It is intended for personal research, red-teaming, and evaluation purposes.
• You, the user, are solely responsible for any output the model produces and for any consequences of using this adapter. The authors of this release, of heretic-gguf, of Heretic, of Unsloth, and of Z.ai accept no liability whatsoever. Using this adapter to produce illegal content or to harm others is your choice and your legal exposure — ensure your use complies with all applicable laws in your jurisdiction.
• By downloading or using this adapter you acknowledge the above.

License

The base model is MIT-licensed (see the base repo); this adapter inherits those terms. The heretic-gguf tooling used to produce it is AGPL-3.0-or-later.