[BugFix] LoRA: Support loading base_layer of experts (#31104)

Signed-off-by: Hollow Man <hollowman@opensuse.org>
2026-01-07 08:49:39 +02:00
parent efeaac92f2
commit 482914849c
35 changed files with 46 additions and 3 deletions
--- a/vllm/model_executor/models/nemotron_h.py
+++ b/vllm/model_executor/models/nemotron_h.py
@@ -636,6 +636,7 @@ class NemotronHModel(nn.Module):
                #   what the activation is applied to
                # - FusedMoe.w3 (aka up_proj) should be ignored since we're
                #   using non-gated MoE
+                self,
                ckpt_gate_proj_name="up_proj",
                ckpt_down_proj_name="down_proj",
                ckpt_up_proj_name="",