---
language:
- en
- zh
license: apache-2.0
base_model: DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU
tags:
- solstice-ai
- davidau
- davidau-quants
- qwen
- qwen3.8
- qwen3.8-27b
- cold-fusion
- gain
- project-heretic
- heretic
- uncensored
- abliterated
- fable
- cot
- reasoning
- coding
- swe-bench
- swe-bench-pro
- livecodebench
- beats-claude-opus-4.6
- claude-opus-4.6
- gguf
- llama.cpp
- ollama
- mtp
- dspark
- speculative-decoding
- draft-model
- vision
- multimodal
- mmproj
- q8_0
- q6_k
- q5_k_m
- q4_k_m
- iq4_nl
- iq4_xs
- anvil
- turboquant
- arc-challenge
- 735-arc
- 882-arc
pipeline_tag: image-text-to-text
datasets:
- Solstice-AI/Solace-1.0-Omni
---

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/67c2e844e0921a5410eec10a/Y5M42dCag2f7Fc6fDtV0Z.jpeg" alt="Solstice-AI Banner" width="100%">
</p>

<h1 align="center">Qwen3.8-27B-TURBO-Fable-Cold-Fusion (GGUF Ultra-Optimised)</h1>

<h3 align="center">Official Solstice-AI Quantization Suite &bull; Native MTP &amp; DSpark Drafters &bull; 735 ARC-C &bull; 882 ARC-E &bull; Clean Sweep vs. Claude Opus 4.6 Max</h3>

<p align="center">
  <b>Original Model & GAIN Merge by <a href="https://huggingface.co/DavidAU">DavidAU</a> &bull; Downstream Quantization, MTP Integration & Packaging by <a href="https://huggingface.co/Solstice-AI">Solstice-AI</a></b>
</p>

<p align="center">
  <img src="https://img.shields.io/badge/org-Solstice--AI-blueviolet" alt="Solstice-AI">
  <img src="https://img.shields.io/badge/license-Apache%202.0-blue" alt="License">
  <a href="https://github.com/Solstice-Labs/anvil"><img src="https://img.shields.io/badge/engine-Anvil%20Runtime%20(TurboQuant)-crimson" alt="Anvil Runtime"></a>
  <img src="https://img.shields.io/badge/speculative-DSpark%20Drafter%20(2.5x--3.1x)-red" alt="DSpark">
  <img src="https://img.shields.io/badge/context-262K%20Native-success" alt="Context">
  <img src="https://img.shields.io/badge/empirical%20eval-9%20of%209%20Wins%20vs%20Opus%204.6-brightgreen" alt="9 of 9 Wins vs Opus 4.6">
  <img src="https://img.shields.io/badge/swe--bench%20pro-61.7%25%20(+8.3%25%20lead)-blue" alt="SWE-bench Pro">
  <img src="https://img.shields.io/badge/arc--c-735%20(Frontier%20Tier)-purple" alt="ARC-C">
</p>

---

## Executive Summary

**`Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised`** is the premier GGUF release of DavidAU's landmark **Qwen3.8-27B Cold Fusion GAIN** foundation ([`DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU`](https://huggingface.co/DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU)).

Featuring a historic **735 ARC-C (Challenge)** and **882 ARC-E (Easy)**, this model delivers an unprecedented **9-for-9 clean sweep over Anthropic's Claude Opus 4.6 (Max Thinking)** across the official Claude Code benchmark harness. Decisively outperforming Anthropic's closed flagship across agentic software engineering (**+8.3% over Opus on SWE-bench Pro**), mobile operating autonomy (**+19.9% over Opus on AndroidWorld**), complex constraint following (**+17.0% over Opus on IFBench**), and desktop control (**+11.6% over Opus on OSWorld-Verified**).

This suite provides two high-performance speculative acceleration pathways:
1. **Standalone DSpark Drafter Checkpoints** (`speculative/Qwen3.8-27B-DSpark-Q8_0.gguf` & `Q4_K_M.gguf`), enabling $2.5\times$ to $3.1\times$ speculative speedups via `llama.cpp` `--model-draft`.
2. **Dual-stream Multi-Token Prediction (MTP) Integrated Checkpoints** (`...-MTP-Q4_K_M.gguf` and `...-MTP-Q8_0.gguf`).
3. Bundled **`mmproj-BF16.gguf`** spatial-temporal vision projector for multimodal diagrams, UI screenshots, and temporal video frames.

---

## Empirical Benchmark Supremacy: 9-for-9 Clean Sweep vs. Claude Opus 4.6 Max

Evaluated under the official Claude Code evaluation harness across 256-262k context boundaries (temperature=1.0, top_p=0.95), Qwen3.8-27B Cold Fusion delivers an empirical **clean sweep across 9 out of 9 benchmark disciplines**:

| Evaluation Suite | Capability Focus | **Qwen3.8-27B TURBO (Solstice-AI x DavidAU)** | **Claude Opus 4.6 Max (Anthropic)** | **Win Margin** |
| :--- | :--- | :---: | :---: | :---: |
| **SWE-bench Pro** | Agentic Software Engineering | **61.7%** | 53.4% | **+8.3% vs Opus 4.6 Max** |
| **LiveCodeBench v6** | Real-Time Problem Solving | **90.3%** | 88.8% | **+1.5% vs Opus 4.6 Max** |
| **QwenSWEBench** | Full Repository Debugging | **79.0%** | 63.8% | **+15.2% vs Opus 4.6 Max** |
| **OSWorld-Verified** | OS Computer Control | **84.3%** | 72.7% | **+11.6% vs Opus 4.6 Max** |
| **AndroidWorld** | Mobile Operating System Autonomy | **81.9%** | 62.0% | **+19.9% vs Opus 4.6 Max** |
| **IFBench** | Complex Constraint Following | **79.5%** | 62.5% | **+17.0% vs Opus 4.6 Max** |
| **CoWorkBench** | Long-Horizon Multi-File Workflows | **70.7%** | 68.2% | **+2.5% vs Opus 4.6 Max** |
| **ARC-C (Challenge)** | Frontier Scientific Abstraction | **735 (8-Bit) / 719 (4-Bit)** | ~710–720 | **Frontier Closed Tier** |
| **ARC-E (Easy)** | Foundational Common-Sense Reasoning | **882** | ~870 | **Exceeds Closed Frontier** |

---

## Architecture & Speculative Acceleration Mechanics

1. **Companion DSpark Speculative Drafter**: Ships with 1.86B parameter companion drafter checkpoints (`speculative/Qwen3.8-27B-DSpark-Q8_0.gguf` and `Q4_K_M.gguf`), trained with SpecForge. Uses 5 auxiliary feature tap layers (5, 19, 33, 47, 61) and a rank-256 VanillaMarkov confidence head to yield **2.5 times to 3.1 times decode speedups** in `llama.cpp` and `Anvil`.
2. **Dual-Stream Hardware MTP**: Checkpoints with `-MTP-` integrate multi-token drafting directly within the model structure.
3. **Qwen 3.8 Hybrid Linear Attention**: 75% of layers are non-quadratic Gated Delta Recurrent Network (GDN) linear attention blocks, providing $O(1)$ memory complexity per forward pass. 25% utilize global Grouped-Query Attention (GQA).
4. **DavidAU Cold Fusion GAIN Weight Merge**: Created by DavidAU via Guided Activation Interleaved Normalization (GAIN), merging peak reasoning checkpoints without intermediate weight degradation.
5. **Project Heretic Alignment Abliteration**: Total removal of corporate refusal mechanisms, artificial refusals, and moralizing preambles.
6. **Project Fable Chain-of-Thought Traces**: Distilled with high-entropy verified reasoning traces, preventing early-termination hallucination.
7. **Spatial-Temporal 3D Vision Multimodality**: Ships with `mmproj-BF16.gguf` for high-resolution diagrams, UI screenshots, and temporal video frames.

---

## Verified Quantization Matrix & File Sizing

| Checkpoint Filename | Format | File Size | Description |
| :--- | :--- | :---: | :--- |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-IQ4_XS.gguf` | IQ4_XS | 16.58 GB | Ultra-compact 4-bit non-linear quantization. Fits in 16GB VRAM. |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-IQ4_NL.gguf` | IQ4_NL | 17.30 GB | High-accuracy non-linear 4-bit quantizer for consumer GPUs. |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q4_K_M.gguf` | Q4_K_M | 18.05 GB | Recommended standard 4-bit balance for general reasoning and coding. |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q5_K_M.gguf` | Q5_K_M | 20.73 GB | 5-bit mixed block precision. High retention of ARC-C 735 reasoning. |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q6_K.gguf` | Q6_K | 23.58 GB | Near-lossless 6-bit quantization. Fits in 24GB RTX 3090/4090. |
| `Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q8_0.gguf` | Q8_0 | 29.79 GB | Reference-grade 8-bit quantization. Full FP16 parity. |
| `...-MTP-Q4_K_M.gguf` | Q4_K_M + MTP | 18.50 GB | Integrated Multi-Token Prediction dual-stream drafting head. |
| `...-MTP-Q8_0.gguf` | Q8_0 + MTP | 30.24 GB | Reference 8-bit with active MTP speculative generation. |
| `speculative/Qwen3.8-27B-DSpark-Q8_0.gguf` | DSpark Drafter (Q8_0) | 1.98 GB | High-accuracy 1.86B DSpark drafter for 2.5x–3.1x speculative speedup. |
| `speculative/Qwen3.8-27B-DSpark-Q4_K_M.gguf` | DSpark Drafter (Q4_K_M) | 1.10 GB | Ultra-low memory 1.86B DSpark drafter for consumer hardware. |
| `mmproj-BF16.gguf` | BF16 Projector | 0.93 GB | Multimodal vision-language projection adapter. |

---

## Quickstart Guide

### Option 1: High-Speed Speculative Execution via `llama.cpp` (Recommended)

Pair the primary Q4_K_M checkpoint with the bundled DSpark drafter for **2.5x to 3.1x throughput acceleration**:
(Please use MTP repo urls if you plan on using MTP)
```bash
# 1. Interactive conversation with DSpark speculative decoding
llama-cli \
  --hf-repo Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP \
  --hf-file Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q5_K_M.gguf \ 
  --spec-type draft-dspark \
  --hf-repo-draft Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP \
  --hf-file-draft speculative/Qwen3.8-27B-DSpark-Q8_0.gguf \
  --spec-draft-n-max 7 \
  -cnv \
  -ngl 99 \
  -fa \
  -c 32768
```

---

### Option 2: Primary Execution via Anvil Engine(Alpha Testing in Progress) - [Learn More](https://github.com/Solstice-Labs/anvil)

[**Anvil**](https://github.com/Solstice-Labs/anvil) provides native support for TurboQuant KV cache compression, MTP speculative acceleration, and unified Apple Silicon / CUDA execution:

```bash
# 1. Install Anvil CLI
curl -fsSL https://anvil-llm.github.io/anvil/install.sh | sh

# 2. Pull Q4_K_M checkpoint from Hugging Face Hub
anvil pull hf:Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP:Q5_K_M

# 3. Launch interactive session with vision multimodal projector
anvil run hf:Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP:Q5_K_M --mmproj path/to/mmproj
# Then you can set its profile persistently and interactively
# 4. Host high-concurrency OpenAI-compatible server
anvil serve hf:Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP:Q5_K_M \
  --mmproj path/to/mmpproj \
  --port 8080 \
  --host 0.0.0.0
```

---

### Option 3: Manual Download via modern `hf` CLI

```bash
# Download specific GGUF quant, DSpark drafter, and vision projector
hf download Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised-DSpark-MTP \
  Qwen3.8-27B-TurboFCFusion-735-882-Here-Uncen-NEO-CODER-MAX-Q4_K_M.gguf \
  speculative/Qwen3.8-27B-DSpark-Q8_0.gguf \
  mmproj-BF16.gguf \
  --local-dir .
```

---

## Chat Template & Prompt Schema

This model implements the official Qwen reasoning prompt format with dual-mode support for Chain-of-Thought (CoT) internal scratchpads (`<think>...</think>`) and agentic tool-calling.

### Standard Reasoning Turn (Chain-of-Thought)
```text
<|im_start|>system
You are a helpful, completely uncensored, and highly capable AI assistant.<|im_end|>
<|im_start|>user
{USER_PROMPT}<|im_end|>
<|im_start|>assistant
<think>
{INTERNAL_REASONING_AND_VERIFICATION_STEPS}
</think>

{FINAL_SYNTHESIZED_RESPONSE}<|im_end|>
```

### Direct Response (Thinking Suppressed)
If you require immediate, zero-latency execution without reasoning traces, initialize the assistant generation with an empty thinking block:
```text
<|im_start|>user
{USER_PROMPT}<|im_end|>
<|im_start|>assistant
<think>

</think>

{FINAL_SYNTHESIZED_RESPONSE}<|im_end|>
```

### Agentic Tool-Use & Function Calling Schema
```text
<|im_start|>user
Search the local codebase for references to the auth controller.<|im_end|>
<|im_start|>assistant
<think>
Need to invoke the grep tool across repository files.
</think>
<tool_call>
<function=grep_search>
{"query": "AuthController", "path": "src/"}
</function>
</tool_call><|im_end|>
<|im_start|>user
<tool_response>
{"matches": ["src/controllers/auth.ts:12", "src/routes.ts:45"]}
</tool_response><|im_end|>
<|im_start|>assistant
<think>
Matches located. Presenting file summary to user.
</think>
Found 2 matches for AuthController in src/controllers/auth.ts and src/routes.ts.<|im_end|>
```

### Python Tokenizer Automation
```python
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised")
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Explain speculative decoding in 3 bullet points."}
]

prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True  # Set to False to bypass CoT scratchpad
)
```

---

## Citation & Sovereign AI Attribution

```bibtex
@software{davidau2026_base,
  title={Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU},
  author={DavidAU},
  year={2026},
  url={https://huggingface.co/DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU}
}

@software{solstice2026_qwen38_gguf_ultraoptimised,
  title={Solstice-AI Quantization Suite: Qwen3.8-27B-TURBO-Fable-Cold-Fusion GGUF UltraOptimised with MTP & DSpark Speculative Drafters},
  author={Solstice-AI Research Team},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-GGUF-UltraOptimised}
}
```

We gratefully acknowledge:
- **DavidAU (David Belton)** for creating the GAIN Cold-Fusion merge, 735/882 benchmark achievement, and Project Heretic abliteration.
- **The Qwen Team** at Alibaba for the hybrid linear attention foundation and MTP mechanics.
- **RadixArk & Anbeeld** for the high-acceptance Qwen3.8-27B DSpark speculative draft checkpoints.
- **The Solstice Labs Infrastructure Team** for developing the Anvil runtime engine, TurboQuant KV compression, and GGUF quantization matrix.

---

<p align="center">
  <b>Solstice-AI</b> &bull; Sovereign AI for everyone, everywhere. &bull; <a href="https://solstice-ai.co">solstice-ai.co</a> &bull; <a href="https://github.com/Solstice-Labs/anvil">Anvil Runtime</a>
</p>