feat: bump to vllm-gb10 v0.24.0-gb10.5 base (vLLM 0.24.1 + b12x)

recipes/glm52-nf3-hybrid/Dockerfile

@@ -4,7 +4,7 @@
 #     (torch 2.11.0+cu130, vLLM, flashinfer, aarch64/sm_121a)
 #   Adds: b12x (voipmonitor NF3-hybrid fork) + Triton NF3 overlay
 
-ARG BASE_IMAGE=ghcr.io/timothystewart6/vllm-gb10:v0.23.0-gb10.1
+ARG BASE_IMAGE=ghcr.io/timothystewart6/vllm-gb10:v0.24.0-gb10.5
 FROM ${BASE_IMAGE}
 
 ARG TORCH_CUDA_ARCH_LIST=12.1a