gemma3n lora

Signed-off-by: NickLucche <[email protected]>

Author: NickLucche

vllm/model_executor/models/gemma3n.py

@@ -20,8 +20,8 @@
 
 import torch
 from torch import nn
-from transformers.models.gemma3n.configuration_gemma3n import Gemma3nTextConfig
 
+from transformers.models.gemma3n.configuration_gemma3n import Gemma3nTextConfig
 from vllm.attention import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig
@@ -47,7 +47,7 @@
 from vllm.sequence import IntermediateTensors
 from vllm.v1.attention.backends.utils import KVSharingFastPrefillMetadata
 
-from .interfaces import SupportsQuant
+from .interfaces import SupportsLoRA, SupportsQuant
 from .utils import (AutoWeightsLoader, extract_layer_index,
                     is_pp_missing_parameter, make_layers, maybe_prefix)
 
@@ -1047,7 +1047,7 @@ def load_weights(self, weights: Iterable[tuple[str,
         return loaded_params
 
 
-class Gemma3nForCausalLM(nn.Module):
+class Gemma3nForCausalLM(nn.Module, SupportsLoRA):
     packed_modules_mapping = {
         "qkv_proj": [
             "q_proj",