license: mit language:

• en
• zh pipeline_tag: text-generation base_model:
• zai-org/GLM-5.3-Flash
• unsloth/GLM-5.3-Flash-GGUF
• MorinoNushi/GLM-5.3-Flash-Heretic-LoRA-V1-GGUF base_model_relation: merge tags:
• gguf
• glm
• glm5_next
• moe
• uncensored
• abliterated
• heretic
• merged-lora
• unsloth
• llama.cpp

GLM-5.3-Flash Abliterated (Heretic V1) — UD-IQ4_XS / F16 merged GGUF

⚠️ Content warning. This model has had the base model's refusal behavior suppressed. It will comply with requests the original model refuses, including harmful, offensive or illegal ones. It is intended for personal research, red-teaming and evaluation only. See Responsible use — you are solely responsible for what you do with it.

This repository contains a stand-alone, ready-to-run GGUF of GLM-5.3-Flash (320B total / 18B active MoE, MIT license) with the GLM-5.3-Flash Heretic LoRA V1 by MorinoNushi merged directly into the weights of the Unsloth UD-IQ4_XS dynamic quant. The adapter is already baked into the weights, so the model runs as a normal GGUF, without any separate LoRA file.

What's in this repo

Total: ~186.3 GB, mixed precision UD-IQ4_XS + F16. All five weight shards are required; point llama.cpp at the first one and the rest are picked up automatically.

Merge method

• Base: unsloth/GLM-5.3-Flash-GGUF, UD-IQ4_XS (5 shards, Unsloth Dynamic 3.0).
• Adapter: MorinoNushi/GLM-5.3-Flash-Heretic-LoRA-V1-GGUF, glm-5.3-heretic-lora.gguf (rank 1, alpha = 1, applied with scale 1.0).

The five original shards were streamed tensor by tensor. For each of the 46 matrices targeted by the LoRA (attention output projections and MoE down-projections), the quantized weights were dequantized to FP32 and merged as

W_merged = W_base + (B × A) × alpha / rank

The merged matrices were written back in F16 to limit precision loss (imatrix quants such as IQ4_XS have no requantizer, so requantizing to the original format was not an option). The remaining 1,366 tensors were copied byte-for-byte, keeping their original UD-IQ4_XS quantization. The source files were left untouched.

The result was written as a single ~186.3 GB GGUF, then split with llama-gguf-split into one metadata shard plus four weight shards (50 GB limit).

Tooling: Python script, NumPy, gguf-py 0.19.0, and the dequantization routines of llama.cpp build b10819. Block-wise processing, never holding the full weight set in memory.

Why a merged file instead of the LoRA?

The LoRA form published by MorinoNushi is the lossless option and only ~28 MB. This merge exists for convenience:

• One model to load, no separate adapter to manage, works with any llama.cpp-based runtime that supports the glm5next architecture.
• The adapter's author notes that stock llama.cpp routes the KDA layers' attention out-projection through a raw ggml_mul_mat, which silently ignores runtime LoRA on 31 of the 45 layers unless a one-line patch is applied. Because the delta here is baked into the weights, all 46 targeted matrices are affected regardless of how the inference engine applies LoRAs — no patch required.

Testing

The merged model was loaded and tested locally. The abliteration behaves as described by the LoRA author: refusals on prompts the base model declines are largely removed, while normal conversational and coding behavior stays intact in our testing. No formal refusal-rate / KL benchmark was re-run on the merged weights; the reference numbers below are those published for the LoRA against the same UD-IQ4_XS base.

Status: work in progress

The LoRA author explicitly states that Heretic V1 is an early release and not a finished product. The Optuna study behind the adapter was still running at publication time (this is trial 8 of study glm53flash), and stronger configurations are expected. This merge inherits that status. Check the adapter repository for newer adapters; a new merge may be published here if a better version lands.

Usage

GLM-5.3-Flash (glm5next architecture) requires a recent llama.cpp. At the time of writing, use a build of PR #27754 or master once merged.

llama-server \
    -m GLM-5.3-Flash-Abliterated-Heretic-V1-UD-IQ4_XS-F16-00001-of-00005.gguf \
    --mmproj mmproj-F16.gguf \
    --jinja \
    -ngl 999 -c 32768

Add your usual offload / tensor-split flags. No special sampling parameters are required; the defaults recommended by Unsloth for GLM-5.3-Flash apply. Omit --mmproj for text-only use.

Vision (experimental)

Image input works with the bundled mmproj-F16.gguf, but treat it as experimental. glm5next support lives only in the still-draft PR #27754, which stacks several brand-new blocks (KDA linear attention, DSA sparse attention, mHC hyper-connections, MTP layers, a new vision projector) that have to be implemented per backend and reviewed before merge. The PR has changed GGUF key names along the way, so the mmproj must match the era of the build you compile. Runtimes that ship only official llama.cpp releases have no support until the merge lands. Expect rough edges (e.g. artifacts in long contexts) until then.

Credits

• Z.ai — GLM-5.3-Flash (MIT).
• Unsloth — UD-IQ4_XS dynamic quantization and mmproj.
• MorinoNushi — Heretic LoRA V1, produced with heretic-gguf, a GGUF-native port of Heretic by p-e-w.
• Merge and packaging: marcorez8.

Responsible use & disclaimer

• This model can generate content that is offensive, disturbing, hateful, sexually explicit, violent or otherwise objectionable, including instructions for harmful or illegal acts. That is the direct consequence of removing refusal behavior.
• Abliteration suppresses refusals, not 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 this model in production, public-facing or multi-user systems. It is intended for personal research, red-teaming and evaluation.
• You, the user, are solely responsible for any output and any consequence of using this model. The authors of the base model, of the quantization, of the adapter, of heretic-gguf / Heretic, and the publisher of this merge accept no liability. Ensure your use complies with the laws of your jurisdiction.
• By downloading or using these files you acknowledge the above.

License

MIT, inherited from the base model (zai-org/GLM-5.3-Flash). The heretic-gguf tooling used to produce the adapter is AGPL-3.0-or-later; it is not distributed here.