Zrald-AI Qwen 3.8 27B (zraldv1-ac - Accuracy Priority Tier)

Official high-efficiency GGUF release of Qwen3.8-27B optimized for Accuracy Priority Tier.

This repository contains zraldv1-ac.gguf (17.08 GiB / 18.34 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: 17.08 GiB (18.34 GB)
• Accuracy Retention: 99.68%
• Prompt Processing Throughput: 1,344.4 tok/s
• Token Generation Velocity: 69.4 tok/s
• Recommended Minimum VRAM: 20 GB
• Profile: Highest precision retention (99.68%). Matches or beats Q6_K and Q8_0 in benchmark accuracy while saving ~10 GB VRAM.

Quickstart with llama.cpp

1. Download Model File

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

2. Run Single Prompt

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

4. Launch OpenAI-Compatible API Server

./build/bin/llama-server \
    -m ./models/zraldv1-ac.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.