[Bugfix] Set kv_quant_mode on the generic MLA KV-cache spec (#48379)
Signed-off-by: Mikhail Kostryukov <mike@triptrack.net> Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
drakosha committed
Jul 15, 2026 at 06:36 UTC
6472131298e7b9ce81651cac90cc2dcc28963b55
1 file changed
+2
vllm/model_executor/layers/attention/mla_attention.py
+2
@@ -276,6 +276,7 @@ from vllm.v1.kv_cache_interface import (
276
AttentionSpec,
277
KVCacheSpec,
278
MLAAttentionSpec,
279
+ get_kv_quant_mode,
280
)
281
282
logger = init_logger(__name__)
@@ -1025,6 +1026,7 @@ class MLAAttention(nn.Module, AttentionLayerBase):
1026
head_size=self.head_size,
1027
dtype=kv_cache_dtype,
1028
cache_dtype_str=vllm_config.cache_config.cache_dtype,
1029
+ kv_quant_mode=get_kv_quant_mode(self.kv_cache_dtype),
1030
)
1031
1032
def _v_up_proj(self, x: torch.Tensor, out: torch.Tensor):