[Bugfix] LoRA: extend expert base_layer loading to Qwen3.5 and Step3.x (#37114)
Signed-off-by: Hollow Man <hollowman@opensuse.org>
This commit is contained in:
@@ -306,9 +306,12 @@ class Qwen3_5Model(Qwen3NextModel):
|
||||
loaded_params: set[str] = set()
|
||||
expert_params_mapping = self.get_expert_mapping()
|
||||
is_fused_expert = False
|
||||
base_layer = (
|
||||
"base_layer." if any(".base_layer." in name for name in params_dict) else ""
|
||||
)
|
||||
fused_expert_params_mapping = [
|
||||
("experts.w13_weight", "experts.gate_up_proj", 0, "w1"),
|
||||
("experts.w2_weight", "experts.down_proj", 0, "w2"),
|
||||
(f"experts.{base_layer}w13_weight", "experts.gate_up_proj", 0, "w1"),
|
||||
(f"experts.{base_layer}w2_weight", "experts.down_proj", 0, "w2"),
|
||||
]
|
||||
num_experts = (
|
||||
self.config.num_experts if hasattr(self.config, "num_experts") else 0
|
||||
|
||||
Reference in New Issue
Block a user