biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

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

Browse Source 
Signed-off-by: Hollow Man <hollowman@opensuse.org>
This commit is contained in:
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
2026-01-07 08:49:39 +02:00
committed by GitHub
parent efeaac92f2
commit 482914849c
35 changed files with 46 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/llama4.py
View File

@@ -539,6 +539,7 @@ class Llama4Model(LlamaModel):
# Expert parameter mapping for the case where the expert weights are
# not fused into a single weight tensor.
expert_params_mapping = SharedFusedMoE.make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",
ckpt_up_proj_name="up_proj",
@@ -548,6 +549,7 @@ class Llama4Model(LlamaModel):
# Expert parameter mapping for the case where the expert weights are
# fused into a single weight tensor.
expert_params_mapping_fused = SharedFusedMoE.make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_up_proj",
ckpt_down_proj_name="down_proj",
ckpt_up_proj_name="gate_up_proj",