[Bugfix] Fix Llama4 weight loading (#45047)

Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Tyler Michael Smith
2026-06-10 13:40:45 -04:00
committed by GitHub
co-authored by Claude Opus 4.6
parent d1bcb4b44c
commit fa8c868a3c
4 changed files with 18 additions and 6 deletions
+2
View File
@@ -51,6 +51,7 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.model_loader.weight_utils import (
default_weight_loader,
maybe_remap_kv_scale_name,
maybe_remap_moe_expert_param_name,
)
from vllm.model_executor.models.interfaces import MixtureOfExperts
from vllm.model_executor.models.utils import sequence_parallel_chunk
@@ -662,6 +663,7 @@ class Llama4Model(LlamaModel):
if "experts." in name and any(
scale_name in name for scale_name in scale_names
):
name = maybe_remap_moe_expert_param_name(name, params_dict)
param = params_dict[name]
weight_loader = getattr(
param, "weight_loader", default_weight_loader