Intel LLM-Scaler-vLLM PV 1.5 Brings Big Improvements

{"heading":"vLLM-Scaler-vLLM PV 1.5 Improvements","subheading":"vLLM Coverage and Performance Enhancements","paragraph_1":"The vLLM-Scaler-vLLM PV 1.5 update includes support for more Qwen and Gemma models, as well as online FP8 and symmetric INT4 quantization.","paragraph_2":"The update also features improved MTP decoding performance, faster time-to-first-token for Qwen and Gemma 4 models, and improved LoRA serving and FP8 KV cache performance.","paragraph_3":"The release notes mention bug fixes, with the vLLM version jumping from 0.14 to 0.26 and PyTorch jumping from v2.10 to v2.12.","paragraph_4":"The Intel software stack has also seen performance improvements, including faster time-to-first-token for Qwen and Gemma 4 models, improved MTP decoding performance, and improved LoRA serving and FP8 KV cache performance."}