[CPU] V1 support for the CPU backend (#16441)

2025-06-04 09:43:01 +08:00
parent 52dceb172d
commit 4555143ea7
15 changed files with 465 additions and 40 deletions
--- a/tests/kernels/attention/test_attention_selector.py
+++ b/tests/kernels/attention/test_attention_selector.py
@@ -85,7 +85,10 @@ def test_env(
                       CpuPlatform()):
                backend = get_attn_backend(16, torch.float16, torch.float16,
                                           block_size, False)
-            assert backend.get_name() == "TORCH_SDPA"
+            if use_v1:
+                assert backend.get_name() == "TORCH_SDPA_VLLM_V1"
+            else:
+                assert backend.get_name() == "TORCH_SDPA"

        elif device == "hip":
            with patch("vllm.attention.selector.current_platform",