TL;DR: Qwopus3.6-27B-v2 abliterated, quantized to NVFP4 (W4A4) for vLLM on NVIDIA Blackwell. 18 GB, wikitext-2 PPL 6.92, 256K general-purpose reasoner, refusals removed (2-round Heretic).

Qwopus3.6-27B-v2 abliterated NVFP4

Jackrong/Qwopus3.6-27B-v2, abliterated (refusal direction removed) with Heretic in two iterative rounds, then quantized to NVFP4 (W4A4) in the compressed-tensors nvfp4-pack-quantized format with llm-compressor (GPTQ + MSE, shared fused-layer scales).

Near-lossless and decensored. Two Heretic rounds cut refusals from 99/100 to 9/100 of held-out harmful prompts while keeping a KL divergence of 0.0160 to the original model (well under the 0.5 line that signals capability damage). NVFP4 then compresses to ~18 GB with a wikitext-2 perplexity of 6.92.

Built for vLLM on NVIDIA Blackwell (4-bit weight + 4-bit activation). Pre-Blackwell GPUs run it weight-only.
Loading and generation verified in vLLM on an NVIDIA GB10 (Blackwell, sm_121).

Uncensored / abliterated model. It follows instructions without refusal guardrails. The abliteration only removes refusals; all other behaviour comes from the base model. You are responsible for how you use it.

Fidelity

Near-lossless versus the bf16 source, 18 GB vs 55.6 GB bf16 (~33%), at wikitext-2 perplexity 6.92 and KL divergence 0.0160 to the original. GPTQ error compensation and an MSE observer keep the drop from bf16 minimal; the header lists the full characteristics and Quantization covers the recipe.

Quickstart

NVFP4 is auto-detected from config.json (compressed-tensors); no quantization flag needed. --reasoning-parser qwen3 splits the <think> block into reasoning_content.

bash

vllm serve maci0/Qwopus3.6-27B-v2-abliterated-NVFP4 \
  --served-model-name qwopus-27b-v2-abliterated-nvfp4 \
  --max-model-len 131072 \
  --gpu-memory-utilization 0.90 \
  --kv-cache-dtype fp8 \
  --reasoning-parser qwen3 \
  --enable-auto-tool-choice --tool-call-parser qwen3_coder

Supports up to 262144 tokens; keep at least 128K to preserve thinking quality. --max-model-len 131072 is a safe default; raise it if memory allows.
Add --language-model-only to skip the vision tower and free KV cache for text use.
The parser flags are not auto-detected; pass them explicitly. Drop the tool-call line if you do not need tool calling.

Python (OpenAI client)

python

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="x")
r = client.chat.completions.create(
    model="qwopus-27b-v2-abliterated-nvfp4",
    messages=[{"role": "user", "content": "Explain, step by step, why the sky is blue."}],
)
print(r.choices[0].message.content)

curl

bash

curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "qwopus-27b-v2-abliterated-nvfp4",
  "messages": [{"role": "user", "content": "Explain, step by step, why the sky is blue."}]
}'

About the base model

A 27B Qwen3.5-family vision-language model (Qwopus 3.6 v2), a general-purpose reasoning and instruction-following model with thinking-mode reasoning and a 256K context window.

64 decoder layers: hybrid gated delta-net linear attention plus full attention, dense MLP, plus a vision tower for image and video input.
256K context (max_position_embeddings 262144).
Thinking mode by default, with an instruct toggle.

Abliteration

Heretic removes the refusal direction with a TPE-optimized search over per-component ablation strength, jointly minimizing refusal rate and KL divergence from the original model, then merges the best trial. This model was abliterated in two iterative rounds: round 1 removed the dominant refusal direction, then Heretic was re-run on the round-1 model to remove the residual refusal direction that surfaced once the first was gone. Because Qwopus is a thinking model, evaluation ran in non-thinking mode so each judged response is a real answer rather than an unfinished <think> block; the shipped model restores the original thinking chat template.

Round	Refusals	KL divergence	Note
Baseline	99/100		original model
Round 1	32/100	0.0235	dominant refusal direction removed
Round 2	9/100	0.0160	residual direction removed (shipped)

Datasets: mlabonne/harmless_alpaca (good) vs mlabonne/harmful_behaviors (bad).
This checkpoint required Heretic with --row-normalization NONE. The default (FULL) degenerated this model, producing a broken output; run with --row-normalization NONE to reproduce.

Quantization


Scheme	NVFP4, W4A4
Weight rounding	GPTQ (Hessian-based error compensation), MSE observer
Weights	FP4 (E2M1), `group_size=16`, `tensor_group`, FP8 (E4M3) group scales, shared across fused layers
Activations	FP4, dynamic per-group, FP8 (E4M3) scales
Quantized	all language-model `Linear` layers
Kept in bf16	vision tower (`model.visual.*`), `lm_head`, MTP head
Untouched	gated delta-net `Conv1d` and SSM params (`A_log`, `dt_bias`), never `Linear`

GPTQ is a quantization-time cost only; inference speed and format are identical to plain round-to-nearest NVFP4, but it chooses better 4-bit values.

Calibration: 512 domain-matched samples (long reasoning + general chat + code), max_seq_len=2048, text-only path through the VL model.

Recommended sampling

Thinking mode is the default.

Thinking, precise: temperature=0.6, top_p=0.95, top_k=20
Thinking, general: temperature=1.0, top_p=0.95, top_k=20
Instruct / non-thinking: temperature=0.7, top_p=0.80, top_k=20
To run non-thinking, set {%- set enable_thinking = false %} in the chat template, or pass extra_body={"chat_template_kwargs": {"enable_thinking": false}}.

Notes

Needs NVIDIA Blackwell (sm_121, e.g. GB10) for accelerated W4A4; pre-Blackwell GPUs run it weight-only.
--reasoning-parser and --tool-call-parser are not auto-detected; pass them explicitly.
Thinking mode is on by default; toggle it via the chat template or chat_template_kwargs.
No refusal guardrails; you are responsible for how you use it.

License

Apache-2.0, following the base model. Intended use and all responsibility for use follow the base model.

Credits

Base model: Jackrong
Abliteration: Heretic by Philipp Emanuel Weidmann
Quantization tooling: llm-compressor / compressed-tensors

Part of Rogue Quants · NVFP4 component datasheets · collection. Fabricated on GB10 (Blackwell) with llm-compressor. Refusals shown per 100 harmful prompts; "n/a" = not separately measured (base-inherited).

maci0/Qwopus3.6-27B-v2-abliterated-NVFP4