[XPU] support HND layout (#47975)
Signed-off-by: zhenwei-intel <zhenwei.liu@intel.com>
liuzhenwei committed
Jul 17, 2026 at 18:54 UTC
850295881041754184717804104fcaadd2b2129e
1 file changed
-8
vllm/platforms/xpu.py
-8
@@ -124,14 +124,6 @@ class XPUPlatform(Platform):
124
attn_selector_config: "AttentionSelectorConfig",
125
num_heads: int | None = None,
126
) -> str:
127
- from vllm.v1.attention.backends.utils import set_kv_cache_layout
128
-
129
- set_kv_cache_layout("NHD")
130
- logger.info_once(
131
- "Setting VLLM_KV_CACHE_LAYOUT to 'NHD' for XPU; "
132
- "only NHD layout is supported by XPU attention kernels."
133
- )
134
-
127
# TurboQuant KV cache: route directly to TQ backend
128
kv_cache_dtype = attn_selector_config.kv_cache_dtype
129
if kv_cache_dtype is not None and kv_cache_dtype.startswith("turboquant_"):