diff --git a/vllm/model_executor/models/lfm2.py b/vllm/model_executor/models/lfm2.py index e18edb0b36..69a0e23be4 100644 --- a/vllm/model_executor/models/lfm2.py +++ b/vllm/model_executor/models/lfm2.py @@ -258,7 +258,7 @@ class Lfm2ShortConvDecoderLayer(nn.Module): model_config=model_config, cache_config=cache_config, quant_config=quant_config, - prefix=f"{prefix}.conv", + prefix=f"{prefix}.short_conv", ) self.feed_forward = Lfm2MLP( diff --git a/vllm/model_executor/models/lfm2_moe.py b/vllm/model_executor/models/lfm2_moe.py index 0b41cb0f0f..44bf55917e 100644 --- a/vllm/model_executor/models/lfm2_moe.py +++ b/vllm/model_executor/models/lfm2_moe.py @@ -354,7 +354,7 @@ class Lfm2MoeShortConvDecoderLayer(nn.Module): model_config=model_config, cache_config=cache_config, quant_config=quant_config, - prefix=f"{prefix}.conv", + prefix=f"{prefix}.short_conv", ) if layer_idx < config.num_dense_layers: