Serving GLM-5.3-Flash on 4× DGX Spark: the attention route is the whole problem

GLM-5.3-Flash looked blocked on GB10: the sparse-MLA kernel path wanted a rope dimension the model does not have, and every workaround wedged the GPU. The fix was to stop patching that path and take a different one. Decode is now flat from 800 to 93K tokens of context.

• GLM-5.3-Flash
• vLLM
• multi-node
• TP4
• DGX Spark
• FP8
• MoE
• linear attention
• speculative decoding
• long context

• time~30 min per node, plus 10–15 min first boot
• yields4-node cluster serving GLM-5.3-Flash at ~84 tok/s decode, 393K context, 2.25M-token KV pool