[GLM5.3 Bug] Fix sparse indexer attn topk backend selection (#58594)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Sep 26, 2026 at 11:35 UTC
3576691c426ea7565789da98b8440600f8ca488e
1 file changed
+1
vllm/models/deepseek_v32/attention.py
+1
@@ -495,6 +495,7 @@ class DeepseekV32Attention(MLAAttention):
495
self._vllm_config.parallel_config.cp_kv_cache_interleave_size
496
),
497
skip_topk_buffer_clear=True,
498
+ topk_backend=self.indexer.indexer_op.topk_backend,
499
)
500
self.impl.record_logical_topk_ready() # type: ignore[attr-defined]
501