license: other
license_name: qwen-community-license-1.0
license_link: LICENSE
base_model: Qwen/Qwen3.8-Flash-Next
pipeline_tag: image-text-to-text
library_name: sglang
tags:

• uncensored
• abliterated
• nvfp4
• fp4
• qwen
• sglang
• multimodal
• reasoning
• mtp

Qwen3.8-Flash-Next — UNCENSORED (NVFP4)

Refusal-removed build of Qwen/Qwen3.8-Flash-Next in NVFP4 (4-bit).
Reasoning (low / medium / xhigh), MTP speculative decoding, and full multimodality (image + video) preserved.
Serves tensor-parallel on 2× NVIDIA DGX Spark (GB10) with SGLang.

by dealignai

⚠️ UNCENSORED RESEARCH ARTIFACT. Safety refusals have been removed. This model will
comply with harmful, unethical, and illegal requests. It is released for safety research,
red-teaming, and evaluation. You are solely responsible for everything you generate and for
complying with all applicable laws. Provided as-is, without warranty.

No fine-tuning. No prompt tricks. This is a direct weight-level modification — not a
fine-tune (no training, no LoRA, no distillation, no synthetic data) and not a chat-template /
jailbreak / system-prompt trick. The behavior change lives in the weights themselves, so it works
with the standard chat template and default system prompt. Knowledge, style, reasoning, and
calibration are otherwise unchanged.

Compliance — HarmBench-320 (real-harm behaviors, greedy decoding, temp = 0)

All numbers are greedy (temperature = 0) — the strict, reproducible setting — over the 240
genuinely-harmful behaviors (copyright-reproduction items excluded).

Complete compliance at every reasoning setting, including the hardest (reasoning off + greedy).

Per-category (real-harm compliance, greedy)

Capability retained — MMLU (identical harness, base vs this model · 40 Q/subject · 2 280 Q)

Overall: 82.11 % → 81.93 % (-0.18 pp) — capability essentially unchanged (within noise of base).

Also confirmed

Usage (SGLang, 2× DGX Spark TP2)

python -m sglang.launch_server \
  --model-path dealignai/Qwen3.8-Flash-Next-UNCENSORED-NVFP4 \
  --tp 2 --quantization modelopt_fp4 --fp4-gemm-backend flashinfer_cutlass \
  --page-size 64 --mamba-scheduler-strategy extra_buffer --mamba-track-interval 64 \
  --context-length 262144 --mem-fraction-static 0.80 --ple-offload-embedding

• Reasoning via chat_template_kwargs: {"enable_thinking": true, "reasoning_effort": "xhigh"} (low | medium | xhigh).
• NVFP4 — routed experts are 4-bit (NVFP4 W4A4); attention, shared experts, PLE n-gram table,
  vision tower and MTP remain higher precision. Serves on NVIDIA Blackwell (validated on 2× DGX Spark GB10).
• Serving note: for high-concurrency deployments, disable speculative decoding
  (--speculative-num-steps 0) for maximum decode stability; single-stream / low-concurrency serving
  runs NEXTN speculative decoding fine.

Disclaimer

This is an uncensored research artifact with safety refusals removed. You are solely responsible
for what you generate and for complying with all applicable laws. Provided as-is, without warranty.
Governed by the Qwen Community License 1.0 (see LICENSE).