EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

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

Browse Source
This commit is contained in:
dmitrygx
2025-09-20 10:30:10 +03:00
committed by GitHub
parent f111649580
commit 2b7417bf6a
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/disaggregation/nixl/conn.py
View File

@@ -547,7 +547,7 @@ class NixlKVManager(CommonKVManager):
notif = "_".join([str(req.room), "kv", str(chunk_id), str(int(is_last))]) 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 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( kv_xfer_handle = self.send_kvcache(
req.agent_name, req.agent_name,
kv_indices, kv_indices,