GLM-5.2
Z.ai (Zhipu AI)We utilized a hybrid approach combining Attention Data Parallelism (DP) and MoE Expert Parallelism (EP)
1 more from this model
DP-attention is primarily introduced to prevent copying KV across different ranks
specific method · filed under inference & serving
DP-attention is a serving parallelism scheme that runs attention under data parallelism across ranks.1 Its purpose is to avoid copying KV cache between ranks while provisioning distributed KV-cache capacity.2
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
multi-node inference with DP-attention for MLA— GLM-5: from Vibe Coding to Agentic Engineering
DP-attention is primarily introduced to prevent copying KV across different ranks— GLM-5: from Vibe Coding to Agentic Engineering
Also called Attention Data Parallelism (DP), Data Parallel attention, --enable-dp-attention, DP-attention.
6 quotes from 5 sources, one card per model, each opening with its strongest.
We utilized a hybrid approach combining Attention Data Parallelism (DP) and MoE Expert Parallelism (EP)
DP-attention is primarily introduced to prevent copying KV across different ranks
enable_attention_dp: true
--enable-dp-attention
For more optimized configurations, including MegaMoE, attention data parallelism, and HiCache, see the [SGLang Cookbook for Step 5 Preview]
For more optimized configurations, including MegaMoE, attention data parallelism and HiCache, see the SGLang Cookbook for Step 5 Preview.
Picked by hand, not extracted: where to read more, not evidence for anything on this page.