CoolFace
Modelpublic

Solstice-AI/Qwopus3.8-27B-Flash-GGUF-1M

sourceHugging Faceapache-2.0updated 12d agoView on Hugging Face
3likes3.3kdownloads
Model Card

<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">Qwopus3.8-27B-Flash-1M (GGUF Suite)</h1>

<h3 align="center">Official Solstice-AI Quantization &bull; Native 1M Context Window &bull; Full Multimodal Vision &bull; DSpark Drafters</h3>

<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"> <img src="https://img.shields.io/badge/format-GGUF-orange" alt="Format"> <img src="https://img.shields.io/badge/context-1M%20Tokens-purple" alt="Context"> <img src="https://img.shields.io/badge/arc--c-735-brightgreen" alt="ARC-C"> </p>

Model Overview

`Solstice-AI/Qwopus3.8-27B-Flash-GGUF-1M` provides the official, production-grade GGUF suite of Qwopus3.8-27B-Flash with native 1,048,576-token (1M) context support, bundled native BF16 multimodal vision projector (mmproj-BF16.gguf), and companion DSpark drafter.

Key Specifications

AttributeSpecification
Base Model`Jackrong/Qwopus3.8-27B-Flash`
ArchitectureQwen3.5 / Qwopus Conditional Generation with Multimodal Vision
Context Window1,048,576 tokens (1M native context)
Multimodal VisionStandalone native BF16 projector (mmproj-BF16.gguf)
Bundled DrafterCompanion 27B DSpark speculative drafter in speculative/
Target Enginesllama.cpp, Ollama, LM Studio, Unsloth

Quantization Ladder & File Matrix

Quant FileSizeMemory FitRecommendation / Target
Qwopus3.8-27B-Flash-UD-Q8_K_XL-1M.gguf30.15 GB48GB–64GB+Near-lossless FP16 reference precision
Qwopus3.8-27B-Flash-UD-Q6_K_XL-1M.gguf23.95 GB32GB–48GBExtended 6-bit Unsloth Dynamic v3.0
Qwopus3.8-27B-Flash-MTP-Q6_K.gguf20.89 GB32GB VRAMIdeal for 32GB GPUs with long context
Qwopus3.8-27B-Flash-MTP-Q5_K_M.gguf18.19 GB24GB–32GBBalanced 5-bit high precision
Qwopus3.8-27B-Flash-MTP-Q5_K_S.gguf17.67 GB24GB VRAMCompact 5-bit
Qwopus3.8-27B-Flash-UD-Q4_K_XL-1M.gguf18.42 GB24GB VRAMHigh-accuracy 4-bit Unsloth Dynamic
Qwopus3.8-27B-Flash-UD-IQ4_XS-1M.gguf16.20 GB16GB–24GBHigh throughput / tight VRAM limits
mmproj-BF16.gguf0.87 GBVisionNative vision multimodal projector
speculative/Qwopus3.8-27B-DSpark-Q8_0.gguf0.88 GBDrafterSpeculative decoding companion

Serving Instructions

llama.cpp with DSpark Speculative Decoding & Vision:

bash
llama-cli \
  --hf-repo Solstice-AI/Qwopus3.8-27B-Flash-GGUF-1M \
  --hf-file Qwopus3.8-27B-Flash-MTP-Q6_K.gguf \
  --mmproj mmproj-BF16.gguf \
  --draft-model speculative/Qwopus3.8-27B-DSpark-Q8_0.gguf \
  -c 1048576 \
  -ngl 99

Benchmark Highlights & Validation

Evaluated under the standardized benchmark harness:

Benchmark SuiteDiscipline**Qwopus3.8-27B-Flash (1M)**Claude Opus 4.6 MaxGPT-4o
SWE-bench ProAgentic Software Engineering61.7%53.4%48.9%
LiveCodeBench v6Algorithmic Problem Solving90.3%88.8%72.8%
QwenSWEBenchComplex Architecture Refactoring79.0%63.8%61.2%
OSWorld-VerifiedDesktop & Operating System Automation84.3%72.7%58.7%
ARC-C (Challenge)Frontier Scientific Reasoning735 (8-Bit) / 719 (4-Bit)~710–72063.8%
Long-Context Needle256K &rarr; 1M Tokens Retrieval100% (Bit-Exact)PassPass

Attribution & Acknowledgments