Zrald-AI Qwen 3.8 27B (zraldv1-cs - Compressed Size Tier)

Official high-efficiency GGUF release of Qwen3.8-27B optimized for Compressed Size Tier.

This repository contains zraldv1-cs.gguf (10.18 GiB / 10.93 GB), physically benchmarked on AMD Instinct MI300X hardware.

For cross-comparison tables against standard Q8_0, Q6_K, Q5_K, Q4_K, and Q2_K models, visit the master repository: 👉 Zrald/Zrald-AI-model-quant-qwen-3.8-27b

Key Performance Indicators

• Physical File Size: 10.18 GiB (10.93 GB)
• Accuracy Retention: 90.72%
• Prompt Processing Throughput: 1,089.6 tok/s
• Token Generation Velocity: 76.33 (⚡ +19.7% speedup) tok/s
• Recommended Minimum VRAM: 12 GB
• Profile: Ultra-compact tier with 62.3% size reduction from Q8_0 base. Solidly exceeds the >=90% accuracy floor at 76.3 tok/s.

Quickstart with llama.cpp

1. Download Model File

huggingface-cli download Zrald/Zrald-AI-qwen-3.8-27b-zraldv1-cs zraldv1-cs.gguf --local-dir ./models

2. Run Single Prompt

./build/bin/llama-cli \
    -m ./models/zraldv1-cs.gguf \
    -ngl 99 \
    -c 4096 \
    -p "<|im_start|>user\nSolve 15 * 14 step-by-step.<|im_end|>\n<|im_start|>assistant\n" \
    -n 128 \
    --single-turn

3. Run Interactive Chat

./build/bin/llama-cli \
    -m ./models/zraldv1-cs.gguf \
    -ngl 99 \
    -c 8192 \
    -cnv

4. Launch OpenAI-Compatible API Server

./build/bin/llama-server \
    -m ./models/zraldv1-cs.gguf \
    -ngl 99 \
    -c 16384 \
    --host 0.0.0.0 \
    --port 8080

Citation & Acknowledgments

• Base model by the Qwen Team (Alibaba) under Apache 2.0.
• Runtime by Georgi Gerganov and the llama.cpp community.