fix(disagg): fix sending KV cache in case of MLA for NIXL backend (#10673)

2025-09-20 10:30:10 +03:00
parent f111649580
commit 2b7417bf6a
1 changed files with 1 additions and 1 deletions
--- a/python/sglang/srt/disaggregation/nixl/conn.py
+++ b/python/sglang/srt/disaggregation/nixl/conn.py
@@ -547,7 +547,7 @@ class NixlKVManager(CommonKVManager):
            notif = "_".join([str(req.room), "kv", str(chunk_id), str(int(is_last))])
            decode_tp_size = self.decode_kv_args_table[req.agent_name].decode_tp_size

-            if decode_tp_size == self.attn_tp_size:
+            if self.is_mla_backend or (decode_tp_size == self.attn_tp_size):
                kv_xfer_handle = self.send_kvcache(
                    req.agent_name,
                    kv_indices,