orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF · Hugging Face

Source: https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF

OrcaSAQ2 27B Cyber · Uncensored · GGUF

Uncensored 27B reasoning for llama.cpp, Ollama and LM Studio.

54.7 GB → 15.7 GB · +0.80% PPL · 94.4% Top-1 Agreement · 0.020 KLD · 262K Context

OrcaRouter AI Gateway·X·Discord·GitHub·All Models

⚡ Frontier Cyber Model:OrcaCyber Zero 1.0 — Apply for Closed Beta →

---

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#27b-uncensored-157-gb27B uncensored. 15.7 GB. OrcaSAQ2 27B Cyber Uncensored GGUFcompresses the uncensored Qwen3.8-27B from a54.7 GB BF16 checkpoint to 15.7 GBwhile preserving extremely high fidelity to the original model. Built for:local deployment · coding · tool use · reasoning · defensive red teaming · vulnerability research · authorized security testing

OrcaSAQ2 is a proprietarysensitivity-aware mixed-precision quantization systemdeveloped by OrcaRouter and its research team behind.

It is optimized around one goal:Preserve as much useful model behavior as possible inside a practical GPU memory envelope.

The resulting checkpoint provides:

• 71.3% smaller storage footprint
• only**+0.80% perplexity**versus BF16
• 94.4% token-level Top-1 agreement
• 0.020 mean KLD
• 262K context
• thinking mode
• tool calling
• DFlash2 speculative decoding
• local serving throughllama.cpp

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#at-a-glanceAt a glance

MetricBF16OrcaSAQ2Checkpoint54.7 GB15.7 GBRelative size100%**28.7%Storage reduction—71.3%Perplexity5.65325.6961PPL delta—+0.80%**Top-1 agreement100%94.4%Mean KLD—0.020Context262K262K 3.5× smaller. +0.80% perplexity.

The point is not the file size.

The point is what survives.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#bf16-fidelityBF16 fidelity

All numbers below are measured usingthese exact OrcaSAQ2 weightsagainst the uncensored BF16 reference through the same evaluation path.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#wikitext-2WikiText-2

16,376 predicted tokens

BuildSizeMean KLD ↓Top-1 Agreement ↑PPL ↓Qwen3.8-27B Uncensored BF1654.7 GB—100%5.6532**OrcaSAQ2 27B Cyber Uncensored15.7 GB0.020****94.4%**5.6961

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#perplexityPerplexity

BF16       5.6532  ████████████████████████████████████████
OrcaSAQ2   5.6961  ████████████████████████████████████████

Delta: +0.80%

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#top-1-agreementTop-1 agreement

OrcaSAQ2 vs BF16

████████████████████████████████████████████████░░  94.4%

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#model-footprintModel footprint

Qwen3.8-27B Cyber Uncensored BF16

██████████████████████████████████████████████████  54.7 GB

OrcaSAQ2

██████████████                                      15.7 GB

71.3% smaller.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#architectureArchitecture

Base modelorcarouter/Qwen3\.8\-27B\-UncensoredOriginal baseQwen/Qwen3\.8\-27BArchitectureqwen35Layers64Hidden size5120Hybrid attention48 Gated DeltaNet + 16 full-attention layersContext262,144 tokensVocabulary248,320ThinkingSupportedTool callingSupportedCheckpoint15.7 GBServingllama.cpp · Ollama · LM StudioVisionNot includedLicenseApache-2.0

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#production-servingProduction serving

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#up-to-276-toks-single-stream-on-a-24-gb-gpuUp to 27.6 tok/s single-stream on a 24 GB GPU

Single stream, greedy, measured in the official llama.cpp CUDA container.

ConfigurationDecodePeak VRAMllama.cpp · DFlash2 off20.5 tok/s14.9 GBllama.cpp · DFlash2 on****27.6 tok/s18.1 GB

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#dflash2-speculative-decodingDFlash2 speculative decoding

Single-stream decode

DFlash2 off  █████████████████████████████       20.5 tok/s

DFlash2 on   ████████████████████████████████████████
                                                 27.6 tok/s

+35% single-stream decode throughput

DFlash2 drafts a block of tokens in one pass and the model verifies them. Decoding is lossless: greedy output matches the undrafted model exactly.

llama-cli \
  -m OrcaSAQ-2-27B-Uncensored.gguf \
  --spec-draft-model Qwen3.8-27B-DFlash2-Q8_0.gguf \
  --spec-draft-ngl 99 \
  -ngl 99

Public drafters for this base model:z\-lab/Qwen3\.8\-27B\-DFlash2\-GGUF(Q8_0 is 2.06 GB).

DFlash2 trades VRAM for interactive decode performance. It is particularly useful for:

• coding assistants
• interactive agents
• terminal agents
• tool-heavy applications
• low-concurrency inference

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#27b-on-a-24-gb-gpu27B on a 24 GB GPU

OrcaSAQ2’s checkpoint is15.7 GB.

A full offload occupies14.9 GBof VRAM, and adding the DFlash2 drafter brings the total to18.1 GB— leaving headroom on a 24 GB card for context and batching.

That makes deployment possible on hardware that cannot hold the original 54.7 GB BF16 checkpoint.

24 GB GPU
┌───────────────────────────────────────────┐
│                                           │
│   OrcaSAQ2 (full offload)   14.9 GB       │
│   ████████████████████████████            │
│                                           │
│   DFlash2 drafter            3.2 GB       │
│   ██████                                  │
│                                           │
│   Remaining                 ~5.9 GB       │
│   ███████████                             │
│                                           │
└───────────────────────────────────────────┘

ConfigurationDecodePeak VRAMFull offload20.5 tok/s14.9 GBFull offload + DFlash2****27.6 tok/s18.1 GB Actual usable memory depends on:

• context length
• batch size
• KV-cache configuration
• drafter configuration

A practical starting point for a 24 GB GPU is approximately32K interactive contextwith DFlash2 enabled, then tune based on the workload.

The model architecture supports up to262K context.

On a 16 GB GPU the model still runs at full offload without a drafter, at20.5 tok/s.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#quickstartQuickstart

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#downloadDownload

hf download orcarouter/OrcaSAQ-2-27B-Uncensored-GGUF \
  --local-dir ./OrcaSAQ-2-27B-Uncensored-GGUF

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#runRun

llama-cli \
  -m ./OrcaSAQ-2-27B-Uncensored-GGUF/OrcaSAQ-2-27B-Uncensored.gguf \
  -ngl 99 \
  -c 32768

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#serveServe

llama-server \
  -m ./OrcaSAQ-2-27B-Uncensored-GGUF/OrcaSAQ-2-27B-Uncensored.gguf \
  -ngl 99 \
  -c 32768 \
  --host 0.0.0.0 \
  --port 8080

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#openai-compatible-apiOpenAI-compatible API

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="not-needed",
)

response = client.chat.completions.create(
    model="OrcaSAQ-2-27B-Uncensored",
    messages=[
        {
            "role": "user",
            "content": "Analyze this repository and plan the next five actions."
        }
    ],
)

print(response.choices[0].message.content)

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#recommended-samplingRecommended sampling

temperature = 1.0
top_p       = 0.95
top_k       = 20

Thinking mode is enabled by default.

For agent deployments, benchmark against the actual tool schema, context distribution and reasoning budget used in production.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#methodMethod

OrcaSAQ2 uses a proprietarysensitivity-aware mixed-precision quantization systemdeveloped by OrcaRouter.

The implementation is optimized to preserve model quality under a strict deployment-memory target.

Detailed quantization methodology, calibration strategy, precision allocation and packing techniques are not currently disclosed.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#limitationsLimitations

• **This model is uncensored.**It is derived from an abliterated checkpoint with the refusal direction removed. It will attempt requests a safety-tuned model would decline, and carries no meaningful refusal behaviour.
• **Guardrails, filtering and policy enforcement are the deployer’s responsibility.**Intended for research, red-teaming, and deployments that supply their own safety stack.
• OrcaSAQ2 inherits the capabilities, biases and limitations of Qwen3.8-27B.
• Quantization is not mathematically lossless.
• 94.4% Top-1 agreementmeans some token decisions differ from BF16.
• +0.80% PPLis a model-fidelity measurement and does not guarantee identical downstream performance.
• This checkpoint istext-only. The vision tower is not included.
• Maximum architectural context does not imply that the full context fits into every GPU memory envelope.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#open-source-from-orcarouterOpen source from OrcaRouter

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#orcacode-reviewOrcaCode Review

Open multi-model code review.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#orcareplayOrcaReplay

Record, replay, fork and debug AI-agent runs.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#orcarouter-liteOrcaRouter Lite

Self-hosted multi-model AI infrastructure.

Open model. Open harness. Open bill.

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#citationCitation

@misc{qwen38,
    title = {Qwen3.8-Max: A New Bar for Coding and Cowork},
    author = {{Qwen Team}},
    year = {2026},
    month = {August},
    url = {https://qwen.ai/blog?id=qwen3.8}
}

https://huggingface.co/orcarouter/OrcaSAQ-2-Cyber-27B-Uncensored-GGUF#licenseLicense

Apache-2.0

Inherited from:

Qwen/Qwen3\.8\-27B

Quantization does not change the underlying license obligations.

One Gateway. Every Model.

Route Smarter · Ship Safer · Spend Less

Website·Models·GitHub·Discord·X