Fully quantized NVFP4 Qwen3.8-27B with QUASAR QAD

Article automatically generated from technical news.

We're releasing a fully quantized NVFP4 version of Qwen3.8-27B. The checkpoint was trained using quantization-aware distillation (QAD) with QUASAR, our new QAT algorithm. We used the original BF16 model as the teacher and distilled the quantized model for 2,446 steps. The checkpoint supports vLLM on NVIDIA Blackwell GPUs: vllm serve QUASAR-QAT/Qwen3.8-27B-QUASAR-NVFP4 \ --max-model-len 262144 \ --gpu-memory-utilization 0.85 This model uses an aggressive quantization con