From 39724758e7b74e7ca2c4416b51dd02d1da6d5ad8 Mon Sep 17 00:00:00 2001 From: Yifan Qiao Date: Thu, 24 Sep 2026 17:13:55 -0700 Subject: [PATCH] [Bugfix] GLM-5.3-Flash: launch the kpool paged MQA logits in the varlen mode its schedule was built with (#55270) Signed-off-by: Yifan Qiao Co-authored-by: Claude Fable 5.1 --- vllm/models/glm5next/nvidia/sparse_indexer.py | 1 + 1 file changed, 1 insertion(+) diff --git a/vllm/models/glm5next/nvidia/sparse_indexer.py b/vllm/models/glm5next/nvidia/sparse_indexer.py index c80e442015..00f51cc6a7 100644 --- a/vllm/models/glm5next/nvidia/sparse_indexer.py +++ b/vllm/models/glm5next/nvidia/sparse_indexer.py @@ -558,6 +558,7 @@ def sparse_attn_indexer_kpool( decode_metadata.schedule_metadata, max_model_len=max_pool_len, clean_logits=False, + indices=decode_metadata.indices, ) num_rows = logits.shape[0] # kpool: logits are pool-granular -> select topk_tokens//kpool pools,