add initialization of module weight for Linear Layers in x_proj

add initialization of module weight for Linear Layers in x_proj in order to avoid NaN value after being processed in mamba library

modeling_mos_mamba.py

@@ -693,6 +693,8 @@ class MoSMambaPreTrainedModel(PreTrainedModel):
             if module.bias is not None:
                 if not getattr(module.bias, "_no_reinit", False):
                     nn.init.zeros_(module.bias)
+            nn.init.uniform_(module.weight, -0.001, 0.001)
+            
         elif isinstance(module, nn.Embedding):
             nn.init.normal_(module.weight, std=self.config.initializer_range)