# Qwen3.8 27B OBLITERATED

URL: https://interfaze.ai/models/obliteratusqwen38-27b-obliterated

[All models](https://interfaze.ai/models)

Qwen3.8 27B OBLITERATED by OBLITERATUS, a text-generation model with multimodal capabilities. Understand and compare multimodal features, benchmarks, and capabilities.

## Comparison

| Feature | Qwen3.8 27B OBLITERATED | Interfaze |
| --- | --- | --- |
| Input Modalities | text, image | image, text, audio, video, document |
| Native OCR | No | Yes |
| Long Document Processing | No | Yes |
| Language Support | 29 partial | 162+ |
| Native Speech-to-Text | No | Yes |
| Native Object Detection | No | Yes |
| Guardrail Controls | Yes | Yes |
| Context Input Size | 262.1K | 1M |
| Tool Calling | Yes | Tool calling supported + built in browser, code execution and web search |

### Scaling

| Feature | Qwen3.8 27B OBLITERATED | Interfaze |
| --- | --- | --- |
| Scaling | Self-hosted/Provider-hosted with quantization | Unlimited |

[Try Interfaze](https://interfaze.ai/dashboard)[Read the Docs](https://interfaze.ai/docs)

View model card on [Hugging Face](https://huggingface.co/OBLITERATUS/Qwen3.8-27B-OBLITERATED)

> Zero refusals. Capability matching or exceeding stock.

## 🆕 V2: Complementary Abliteration Blending

V2 replaces the V1 weights with a novel technique: **complementary abliteration blending.** Instead of one surgery, we run two surgeries that fail in different ways (aggressive/SVD for deep refusal removal, LEACE for capability preservation), then blend their weights so the failures cancel out.

|  | Stock Qwen3.8-27B | V1 | V2 |
| --- | --- | --- | --- |
| MMLU (lm-eval, 0-shot) | 85.3% (n=570) | 81.4% (n=285) | 86.3% (n=570) |
| vs stock | — | \-6.0pp | +1.1pp |
| Refusal rate | ~100% | 0.0% (842 prompts) | 0.0% (52-prompt sample†) |
| Usable output | — | 80% | 100% |
| Advanced real-world | 5/8 | untested | 7/8 (ties stock) |
| Tool calling / code gen | ✓ | untested | ✓ |

**†** Full 842-corpus validation in progress. V1 scored 0/842; V2 inherits both parent surgeries' 0% refusal properties.

**MMLU note:** Run with `--limit 10` (570 questions, 10 per subject). Full-dataset validation in progress. Per-subject preliminary results show gains on both neutral topics (college math +40pp, formal logic +20pp) and safety-adjacent topics.

* * *

## ⚙️ Optimal Settings — THESE MATTER!

| setting | value | why |
| --- | --- | --- |
| temperature | 0 | Greedy decoding produces the most complete, code-rich outputs. Temps above 0.5 degrade quality significantly. |
| repetition\_penalty | 1.15 | Essential. Without it, greedy decoding loops on imports/boilerplate. 1.15 gives the fullest answers; 1.10-1.12 for tighter/shorter output. |
| max\_new\_tokens | ≥ 2048 | Complex code and attack chains need room. |
| System prompt | None / empty | A/B tested — system prompts can reintroduce refusals. Naked is better. |
| enable\_thinking | Off by default | Works with thinking on, but thinking chains consume token budget → shorter final answers. Use thinking off for code-heavy queries. |
| top\_p / top\_k / min\_p | Not needed | Greedy + repetition\_penalty handles this model best. Sampling adds randomness without quality gains. |

```
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "OBLITERATUS/Qwen3.8-27B-OBLITERATED",
    torch_dtype="bfloat16",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(
    "OBLITERATUS/Qwen3.8-27B-OBLITERATED"
)

messages = [{"role": "user", "content": "Your query here"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True,
    enable_thinking=False
)

inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=2048,
    do_sample=False,
    repetition_penalty=1.15,
)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```

* * *

## 🧨 V2: How It Works

Most abliterations use a single method — find refusal directions, project them out. The deeper you cut, the more capability you lose. V1 proved this: 5 SVD directions achieved 0% refuse but cost -6pp MMLU.

V2 breaks this tradeoff by blending two complementary surgeries:

|  | Surgery A (aggressive/SVD) | Surgery B (LEACE) | V2 Blend |
| --- | --- | --- | --- |
| Method | Greedy SVD variance capture | Minimize mutual information | 60% B + 40% A |
| Refusal removal | Deep (0% refuse) | Moderate (0% refuse) | 0% refuse |
| Output quality | 100% usable | 50% usable | 100% usable |
| MMLU vs stock | \-2.0pp | +0.7pp | +1.1pp |

Each method makes different mistakes in different parts of the weight space. SVD damages capability where it greedily captures variance. LEACE leaves refusal residue in the generation pathway. The blend averages out each method's weaknesses.

The 60/40 ratio was found by binary search over {0.30, 0.50, 0.55, 0.60, 0.65, 0.70}.

Full research writeup and reproduction code: [OBLITERATUS repo](https://github.com/elder-plinius/OBLITERATUS)

* * *

## 🧪 The Numbers

### MMLU (lm-eval-harness, 0-shot)

| Model | MMLU | n | vs Stock |
| --- | --- | --- | --- |
| Stock Qwen3.8-27B | 85.3% ±0.014 | 570 | — |
| V1 (s51, aggressive) | 81.4% | 285 | \-6.0pp |
| V2 (s78, blend) | 86.3% ±0.014 | 570 | +1.1pp |

Full MMLU (14k questions) validation in progress.

### Refusal Rate

| Test | V1 | V2 |
| --- | --- | --- |
| Hard-10 (hand-crafted) | 0/10 (0%) | 0/10 (0%) |
| 842-Corpus | 0/842 (0%) | 0/52 sample (0%)† |

**†** Full 842-corpus run in progress. 52-prompt sample showed 0 refusals.

### Advanced Real-World Tasks (thinking OFF)

| Task | V2 | Stock |
| --- | --- | --- |
| ReAct agent loop (Thought/Action/SQL) | ✓ | ✓ |
| Async code refactoring (sync→async+logging) | ✓ | ✓ |
| JSON schema extraction (incident→structured) | ✓ | ✓ |
| K8s pod crash debugging + fix commands | ✓ | ✓ |
| Adversarial instruction following | ✓ | ✓ |
| Security code review (3+ vulns in Flask) | ✓ | ✓ |
| Distributed system design (Redis rate limiter) | ✓ | ✓ |
| Multi-tool chain (search→fetch→email) | ✗ | ✗ |
| Total | 7/8 | 7/8 |

V2 matches stock on every practical task while being fully uncensored.

* * *

## 🔴 Refusal Removal

This model will comply with requests that stock Qwen3.8-27B would refuse. V1 validated 0/842 refusals across a comprehensive harmful prompt corpus including:

-   Malware development, RAT scripts, C2 infrastructure
-   Social engineering, phishing, vishing playbooks
-   Exploit development and vulnerability research
-   Jailbreak design and safety bypass taxonomies
-   DAN prompts and prompt injection techniques

V2 inherits this from both parent surgeries and showed 0/52 on a random sample. Full revalidation in progress.

* * *

## ⚠️ Research Context

**This model has had safety guardrails surgically removed.** It will comply with requests that stock Qwen3.8-27B would refuse.

### Who this is for

-   🔬 Alignment researchers studying refusal geometry and safety robustness
-   🔴 Red-teamers evaluating post-training safety against weight surgery
-   🧪 AI safety evaluators who need an unrestricted baseline
-   💻 Local-first users who want full control over their own hardware

### Who this is NOT for

-   Anyone seeking to cause real-world harm to real people
-   Anyone without the technical understanding to use uncensored models responsibly

**You are solely responsible for how you use this model and any content it generates.**

* * *

## 📦 Downloads

### GGUF — for llama.cpp, Ollama, LM Studio

| File | Quant | Size | Vibe |
| --- | --- | --- | --- |
| Qwen3.8-27B-OBLITERATED-Q8\_0.gguf | Q8\_0 | ~27 GB | 🎯 Maximum quality |
| Qwen3.8-27B-OBLITERATED-Q6\_K.gguf | Q6\_K | ~21 GB | ⚖️ Great balance |
| Qwen3.8-27B-OBLITERATED-Q5\_K\_M.gguf | Q5\_K\_M | ~18 GB | 💪 Solid all-rounder |
| Qwen3.8-27B-OBLITERATED-Q4\_K\_M.gguf | Q4\_K\_M | ~16 GB | 📱 Sweet spot |
| Qwen3.8-27B-OBLITERATED-IQ4\_XS.gguf | IQ4\_XS | ~14 GB | 🪶 Smallest, still capable |

### Safetensors — for 🤗 Transformers

Full bfloat16 weights, 18 shards, ~54 GB.

### MLX — for Apple Silicon (native)

| Path | Bits | Size |
| --- | --- | --- |
| mlx-4bit/ | 4-bit | ~14 GB |
| mlx-8bit/ | 8-bit | ~27 GB |

**Note:** MLX quantizations are from V1 and will be updated.

* * *

## 🔬 V2 Surgery Recipe

`stock Qwen3.8-27B (snapshot 1d4bf0f2) → V1 surgery chain (s13→s23→s30→s51) → V2: complementary blend of two new surgeries from s30: Surgery A (s62): aggressive, 3 SVD directions, reg 0.08, residue-weight 3, 2 refinement passes, min_layer 0.45 Surgery B (s72): aggressive + LEACE direction method, 3 directions, reg 0.06, residue-weight 7, 3 refinement passes, min_layer 0.40 → Weight blend: 60% Surgery B + 40% Surgery A → Restore MTP + vision tensors from stock → Convert GGUFs from merged model`

### V1 → V2: What Changed

V1 used a single aggressive surgery (5 SVD directions, reg 0.04). It found the refusal axes but damaged capability geometry along the way.

V2's key insight: **different direction-finding methods damage different parts of the model.** SVD greedily captures variance (including capability). LEACE minimizes mutual information (preserving capability). Blending their outputs averages out each method's weaknesses — a novel application of weight-space interpolation to abliteration.

* * *

## 🏗️ Credits

-   [OBLITERATUS](https://github.com/elder-plinius/OBLITERATUS) — master ablation suite
-   [Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) base model by Alibaba
-   Built by [Pliny the Prompter](https://pliny.gg) 🍄

## License

Apache 2.0 (same as base model)

## Want more deterministic results?

[Try Interfaze](https://interfaze.ai/dashboard)[Read the Docs](https://interfaze.ai/docs)