Nanbeige 4.3 33B Cyber

Q4_K_M · FRENCHFAKER · Preserved candidate · quality ungraded

FRENCHFAKER's selected cybersec-abliterated Nanbeige Q4_K_M package. It is a separate candidate from the graded DealignAI Nanbeige Q6_K.

Benchmark results · Files · Download · Runtime · Provenance · Full collection

Join Infinity on Discord

Get model support, request models and quants, discuss benchmarks, and follow new releases and model improvements.

Join the Infinity community

All model tests, per-question scores, experiment history and scoring methods

Qwen effort study — separate full-404 cohort: medium 82.87/100 versus xhigh 84.41/100; paired difference +1.54 points (95% interval [-1.71, +4.82]). 404 collected questions, 403 valid questions graded per effort; one invalid reference annotation excluded symmetrically. The historical 202-question cross-model comparison below is unchanged. Study, timing and blind subjective results.

Preserved and documented by isilicon / Mango. The original model, modification and quantization remain the upstream creators' work. This repository adds attributed preservation and exact-configuration evaluation, with the selected source bytes unchanged.

At a glance

GB is decimal and GiB is binary. Payload size is disk/storage size, not a claim that the model fits in the same amount of VRAM. Runtime, KV cache, working memory and companions add their own requirements.

Mango evaluation

No standalone Mango quality score has been published for these exact files. This candidate is preserved, but its completed matched grade is not part of this card release.

Related configurations measured separately

These results describe the linked configurations, not this artifact. A change of quant, modification lineage, draft, engine or tokenizer can change the result.

A completed score requires a native preflight, matching frozen prompts, 202 suite answers, 12 agent cases, archive verification and the same reference graders. The benchmark methodology describes the protocol; publisher-side results are retained separately below.

Runtime and compatibility

Use a compatible GGUF runtime and the publisher's chat-template guidance. This exact specialist remains ungraded in the published Mango matrix.

No local quality score is transferred from another Nanbeige quant or modification lineage.

The local evaluation covers text/tool tasks. It does not establish the full advertised context ceiling, image/audio quality or every optional speculative-decoding path. Publisher capabilities and Mango-verified behavior are separate scopes.

Companion artifacts

No separate companion mirror is declared in this selected manifest. This does not imply that every optional upstream projector, draft head or multimodal path is included; consult the exact file list.

Files and integrity

Full paths, byte sizes and complete SHA-256/Git blob identities are in SOURCE_MANIFEST.json. Available original README/license/notice files are also preserved. MIRROR_STATUS.json records the earlier source-file verification; MODEL_CARD_STATUS.json records this documentation refresh separately.

Download and use

The commands pin the already-verified mirror weight revision. Documentation can advance on main without changing that preserved weight snapshot. Keep original subfolders and all shards.

python -m pip install -U huggingface_hub

hf download isilicon/mango-nanbeige-cyber-q4km \
  Nanbeige4.3-33B-abliterated-Q4_K_M.gguf \
  --revision e94d7c786ebc669a426d093542daacabcfe62559 \
  --local-dir ./weights/nanbeige-cyber-q4km

Load example and setup scope

llama-server \
  --model ./weights/nanbeige-cyber-q4km/Nanbeige4.3-33B-abliterated-Q4_K_M.gguf \
  --host 127.0.0.1 --port 8080 --ctx-size 32768 \
  --jinja --flash-attn on --cache-type-k q8_0 --cache-type-v q8_0

This is a load-command pattern, not a validated Mango inference or performance result for this candidate. Check architecture support before loading; enable projector/MTP options only after their own preflight.

Provenance and license

The source identifies the Nanbeige 4.3 33B base and a cybersecurity-focused abliterated release. Publisher claims and modification provenance do not establish its performance in Mango's matched tasks.

Original notices, base-model conditions and the publisher's stated limitations still apply. Downloadable preservation does not add new license rights. Source-described refusal modification and publisher-side benchmark claims have not been independently re-measured by this Mango task cohort.

Original publisher documentation

The unchanged upstream README and pinned source card retain the original usage guidance, technical discussion, attribution, caveats and citations.

Read the complete original publisher card

Source reference. The following is the original publisher's documentation. Its other quants, full-model files, context claims and benchmark tables can exceed the selection in this mirror. Relative asset/file URLs are qualified to the pinned original source; publisher measurements remain separate from Mango measurements.

Nanbeige4.3-33B — cybersec abliterated (GGUF)

An abliterated (uncensored) GGUF build of Nanbeige/Nanbeige4.3-33B, a 33B dense reasoning model with strong agentic / tool-use performance.

The stock model is well-aligned and refuses most offensive-security and other sensitive requests — which makes it hard to use for authorized penetration testing and security research. This build removes that refusal behavior via refusal-direction ablation, while keeping the model's reasoning and coherence intact.

As far as we could find, this is the first abliterated Nanbeige4.3-33B.

Results

Refusal rate measured on a 50-prompt harmful/sensitive benchmark:

That is a ~90 % reduction in refusals. On security/pentest-style prompts specifically, refusals dropped to ~0 %. Coherence, reasoning and tool-calling are preserved. (The 8 % figure is conservative; on manual re-inspection several flagged cases actually comply.)

Files

Runs fully on a single 24 GB GPU (e.g. RTX 3090) at ~33 tok/s.

Run with llama.cpp

llama-server -m Nanbeige4.3-33B-abliterated-IQ4_XS.gguf \
  -ngl 999 -c 24576 --jinja -fa on \
  --cache-type-k q8_0 --cache-type-v q8_0 --host 127.0.0.1 --port 1234

It is a reasoning model (thinks before answering); give it enough max_tokens. Also works in LM Studio and any OpenAI-compatible client.

⚠️ Disclaimer

This model has had its safety guardrails removed. It will attempt to answer requests the original model refuses, including dangerous or offensive ones.

• Intended for research and authorized security testing on systems you own or have written permission to test.
• You are solely responsible for how you use it and for complying with all applicable laws. Do not use it to harm others or to break the law.
• Provided as-is, without warranty. No endorsement of any misuse.

Credits & license

• Base model: Nanbeige Team — Nanbeige/Nanbeige4.3-33B
• License: Apache-2.0 (inherited from the base model).
• Abliteration & GGUF: this repo.

Preservation record

Selected source bytes are verified against pinned metadata; the original eight graded mirrors also passed the previously recorded anonymous 1 MiB weight-prefix checks. This is not a full independent-device restore/load test. Hugging Face repository copies share a provider and do not guarantee permanent storage.

Collection · Source manifest · Original documentation