[Kernel] [CPU] refactor cpu_attn.py:_run_sdpa_forward for better memory access (vllm-project#24701)

Signed-off-by: ignaciosica <mignacio.sica@gmail.com>

Author: ignaciosica

vllm/v1/attention/backends/cpu_attn.py

@@ -641,10 +641,6 @@ def _run_sdpa_forward(
         attn_metadata: TorchSDPAMetadata,
         attn_type: str = AttentionType.DECODER,
     ) -> None:
-        if self.num_kv_heads != self.num_heads:
-            key = key.repeat_interleave(self.num_queries_per_kv, dim=1)
-            value = value.repeat_interleave(self.num_queries_per_kv, dim=1)
-
         attn_masks = attn_metadata.get_attn_bias(attn_type)
         if attn_masks is None:
             if self.alibi_slopes is not None:
@@ -665,6 +661,10 @@ def _run_sdpa_forward(
         key = key.movedim(0, key.dim() - 2)
         value = value.movedim(0, value.dim() - 2)
 
+        if self.num_kv_heads != self.num_heads:
+            key = key.repeat_interleave(self.num_queries_per_kv, dim=-3)
+            value = value.repeat_interleave(self.num_queries_per_kv, dim=-3)
+
         causal_attn = (attn_type == AttentionType.DECODER)
 
         seq_lens_q, seq_lens_kv = attn_metadata.get_seq_lens(attn_type)