From 058f99c561065238bc7ce75729a00c1d47d33cd9 Mon Sep 17 00:00:00 2001 From: Haotong Zou Date: Thu, 27 Aug 2026 01:42:36 +0000 Subject: [PATCH] [CPU] Fix truncated KV prefix in intel_amx spec verify (#36430) --- python/sglang/srt/layers/attention/intel_amx_backend.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/python/sglang/srt/layers/attention/intel_amx_backend.py b/python/sglang/srt/layers/attention/intel_amx_backend.py index 3ab03afdf..8dc211d1b 100644 --- a/python/sglang/srt/layers/attention/intel_amx_backend.py +++ b/python/sglang/srt/layers/attention/intel_amx_backend.py @@ -215,7 +215,6 @@ class IntelAMXAttnBackend(AttentionBackend): seq_lens, extend_seq_lens, extend_start_loc, tree_mask = self.extend_metadata _, max_extend_len = self.forward_metadata - seq_lens = forward_batch.seq_lens if seq_lens.dtype != torch.int64: seq_lens = seq_lens.to(torch.int64) @@ -262,7 +261,6 @@ class IntelAMXAttnBackend(AttentionBackend): seq_lens = forward_batch.seq_lens q = q.reshape(-1, layer.tp_q_head_num * layer.qk_head_dim) - seq_lens = forward_batch.seq_lens if seq_lens.dtype != torch.int64: seq_lens = seq_lens.to(torch.int64)