[kv_offload+HMA][7/N]: Support register_kv_caches for hybrid models (#37853)

Signed-off-by: Or Ozeri <oro@il.ibm.com>
2026-03-27 08:38:33 +03:00
parent 999dfc1622
commit 7cc302dd87
13 changed files with 1476 additions and 758 deletions
--- a/vllm/v1/attention/backends/utils.py
+++ b/vllm/v1/attention/backends/utils.py
@@ -78,9 +78,10 @@ def get_kv_cache_layout():
    return cache_layout


-def set_kv_cache_layout(cache_layout: KVCacheLayoutType):
+def set_kv_cache_layout(cache_layout: KVCacheLayoutType | None):
    global _KV_CACHE_LAYOUT_OVERRIDE
    _KV_CACHE_LAYOUT_OVERRIDE = cache_layout
+    get_kv_cache_layout.cache_clear()


@dataclass