[Model][Gemma4] Load Weights with AutoWeightsLoader (#55911)

Signed-off-by: Josephasafg <ajgard7@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Asaf Gardin
2026-09-18 12:07:30 +00:00
committed by GitHub
co-authored by Harry Mellor Claude Opus 5
parent 2b8a5495a8
commit 44dd18fe0b
8 changed files with 239 additions and 518 deletions
+2 -5
View File
@@ -46,10 +46,7 @@ from vllm.logger import init_logger
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import ReplicatedLinear
from vllm.model_executor.models.gemma3n_mm import batch_audio_features
from vllm.model_executor.models.gemma4 import (
_GEMMA4_EXPERT_PARENT_MAPPER,
Gemma4ForCausalLM,
)
from vllm.model_executor.models.gemma4 import Gemma4ForCausalLM
from vllm.model_executor.models.module_mapping import MultiModelKeys
from vllm.model_executor.models.transformers.utils import recursive_replace_linear
from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -1060,7 +1057,7 @@ class Gemma4ForConditionalGeneration(
}
# Maps checkpoint prefixes to vLLM module paths.
hf_to_vllm_mapper = _GEMMA4_EXPERT_PARENT_MAPPER | WeightsMapper(
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_prefix={
# vision tower
"model.vision_tower": "vision_tower",