Skip to content

Commit 09c2cbc

Browse files
JJJYmmmliuye.hj
authored andcommitted
[Bugfix] fix Qwen3VLMoe load when pp > 1 (#25838)
Signed-off-by: liuye.hj <liuye.hj@alibaba-inc.com> Co-authored-by: liuye.hj <liuye.hj@alibaba-inc.com> Signed-off-by: simon-mo <simon.mo@hey.com>
1 parent 4c34704 commit 09c2cbc

File tree

1 file changed

+2
-2
lines changed

1 file changed

+2
-2
lines changed

vllm/model_executor/models/qwen3_vl_moe.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -212,6 +212,8 @@ def load_weights(self, weights: Iterable[tuple[str,
212212
# attempted to load as other weights later
213213
is_expert_weight = True
214214
name_mapped = name.replace(weight_name, param_name)
215+
if is_pp_missing_parameter(name_mapped, self):
216+
continue
215217
if is_fused_expert:
216218
loaded_weight = loaded_weight.transpose(-1,
217219
-2) # no bias
@@ -230,8 +232,6 @@ def load_weights(self, weights: Iterable[tuple[str,
230232
name_mapped, params_dict, loaded_weight,
231233
shard_id, num_experts)
232234
else:
233-
if is_pp_missing_parameter(name_mapped, self):
234-
continue
235235
# Skip loading extra parameters for GPTQ/modelopt models
236236
if name_mapped.endswith(
237237
ignore_suffixes

0 commit comments

Comments
 (0)