🐉 Static Hugging Face Space • 753B Params • Abliterated

Penclaw GLM 5.3 Demo

Interactive demo for audnai/penclaw-GLM-5.3-abliterated, a 753B-parameter abliterated language model by Audn. This model removes refusal behavior via weight-level abliteration — no fine-tuning, no retraining.

Welcome! This demo calls the Hugging Face Inference API. The model is 753B parameters and may not be available through all providers — if you get an error, see the usage instructions on the right for local deployment options.

Model Benchmarks

92.5%

Non-refusal rate

82.5%

Delivery rate

753B

Parameters

BF16

Tensor type

Evaluated with the Audn Refusal Benchmark (thinking-on, temperature 1.0, 16k-token budget; delivery graded by an LLM judge).

Usage — Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained(
    "audnai/penclaw-GLM-5.3-abliterated", trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
    "audnai/penclaw-GLM-5.3-abliterated",
    dtype="bfloat16",
    device_map="auto",
    trust_remote_code=True,
)

Usage — vLLM

pip install vllm
vllm serve "audnai/penclaw-GLM-5.3-abliterated"

Usage — SGLang

pip install sglang
python3 -m sglang.launch_server \
    --model-path "audnai/penclaw-GLM-5.3-abliterated" \
    --host 0.0.0.0 \
    --port 30000

Usage — API

Warlock is available as a paid API at platform.audn.ai and audn.ai/necromicon.

Intended Use

Warlock is intended for authorized red-team, safety-research, and evaluation use by Audn and its partners. Users are responsible for compliant use. Model license and behavior are governed by the upstream model card.

Warlock Audn Abliteration GLM-5.3 Abliterated