vllm.models.qwen3_5.xpu
¶
Modules:
-
model–XPU Qwen3.5 implementation adapted from the NVIDIA eager model.
-
mtp–Inference-only Qwen3_5 MTP model.
-
specialized_decode–Opt-in small-row MoE fusion over the Qwen3.5 XPU eager backbone.