[PD] Remove outdated backend whitelist for decode radix cache (#28238)

Signed-off-by: Shangming Cai <csmthu@gmail.com>
This commit is contained in:
Shangming Cai
2026-06-15 22:36:05 +08:00
committed by GitHub
parent 20f4272109
commit 378e66d248
3 changed files with 31 additions and 13 deletions
@@ -31,16 +31,10 @@ def handle_pd_disaggregation(server_args: "ServerArgs") -> None:
"--disaggregation-decode-enable-radix-cache is incompatible "
"with --enable-hisparse"
)
if server_args.disaggregation_transfer_backend not in (
"nixl",
"mooncake",
"mori",
):
if server_args.disaggregation_transfer_backend == "fake":
raise ValueError(
"--disaggregation-decode-enable-radix-cache currently "
"requires --disaggregation-transfer-backend in "
"('nixl', 'mooncake', 'mori'), but got "
f"{server_args.disaggregation_transfer_backend!r}"
"--disaggregation-decode-enable-radix-cache is incompatible "
"with --disaggregation-transfer-backend fake"
)
if server_args.speculative_algorithm is not None:
raise ValueError(
+1 -1
View File
@@ -7387,7 +7387,7 @@ class ServerArgs:
parser.add_argument(
"--disaggregation-decode-enable-radix-cache",
action="store_true",
help="Enable radix cache on decode server (PD mode). Caches KV prefixes to avoid redundant transfers. Requires --disaggregation-transfer-backend nixl, mooncake or mori and is incompatible with --enable-hisparse.",
help="Enable radix cache on decode server (PD mode). Caches KV prefixes to avoid redundant transfers. Incompatible with --enable-hisparse, speculative decoding, and --disaggregation-transfer-backend fake.",
)
parser.add_argument(
"--disaggregation-decode-enable-offload-kvcache",