Update Dockerfile.ci to use a new vllm wheel URL and refactor MooncakeConnector import logic in monkey_patch.py for improved compatibility. Adjust OpenAIServingChat to call the harmony request method from the correct context.

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
This commit is contained in:
tzhouam committed 2026-03-16 08:33:28 +00:00
1 parent a26e568dfc
commit de6bcc734c
3 files changed
+18 -22

No files matched your search

@@ -8,6 +8,7 @@ so the decode side references the correct KV entry.
from __future__ import annotations
import importlib
import logging
import sys
from dataclasses import dataclass
@@ -30,30 +31,23 @@ class PatchedRecvReqMeta:
def _import_mooncake_module():
"""Import MooncakeConnector module, supporting both vLLM >=0.16 and older."""
try:
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake import mooncake_connector
return mooncake_connector
except ImportError:
pass
try:
from vllm.distributed.kv_transfer.kv_connector.v1 import mooncake_connector
return mooncake_connector
except ImportError:
return None
for mod_path in (
"vllm.distributed.kv_transfer.kv_connector.v1.mooncake.mooncake_connector",
"vllm.distributed.kv_transfer.kv_connector.v1.mooncake_connector",
):
try:
return importlib.import_module(mod_path)
except (ImportError, ModuleNotFoundError):
continue
return None
def _create_patched_mooncake_connector():
"""Return a subclass of MooncakeConnector with remote_request_id support."""
try:
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.mooncake_connector import (
MooncakeConnector as _OriginalMooncakeConnector,
)
except (ImportError, AttributeError):
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake_connector import (
MooncakeConnector as _OriginalMooncakeConnector,
)
_mc_mod = _import_mooncake_module()
if _mc_mod is None:
raise ImportError("Cannot import MooncakeConnector from upstream vLLM")
_OriginalMooncakeConnector = _mc_mod.MooncakeConnector
class PatchedMooncakeConnector(_OriginalMooncakeConnector):
"""Fixes request-ID mismatch in PD disaggregation by injecting