GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
logicauro
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
logicauro
Model Tree
allura-org/Qwen3.8-27B-Dominatrix
Input Modalities
Output Modalities
Supported Functionality
Linear layers (Qwen3_5ForConditionalGeneration)lm_head, vision tower, GDN linear_attn internals, norms, embeddings~51 GB BF16 → ~26 GB NVFP4A16.
bash
vllm serve logicauro/Qwen3.8-27B-Dominatrix-NVFP4A16 \--served-model-name Qwen3.8-27B-Dominatrix \--language-model-only \--quantization compressed-tensors \--kv-cache-dtype fp8 \--max-model-len 98304 \--max-num-seqs 1 \--reasoning-parser qwen3 \--chat-template chat_template.jinja \--trust-remote-code
Temperature 1.0–1.25 with min_p 0.1 or top_p 0.95. Chat-template defaults: preserve_thinking off, reasoning_effort medium.
Original finetune: allura-org/Qwen3.8-27B-Dominatrix (Fizz / Allura), base Qwen/Qwen3.8-27B.