疑问

[nagara214]

加载权重后输出这些信息正常吗

Some weights of the model checkpoint at ./ckpt/Lenna-7B were not used when initializing LennaForCausalLM: ['model.visual_model.encoder.fusion_layers.1.weight_l', 'model.visual_model.encoder.fusion_layers.3.weight_v', 'model.visual_model.encoder.fusion_layers.4.weight_v', 'model.visual_model.encoder.fusion_layers.2.weight_l', 'model.visual_model.encoder.fusion_layers.5.weight_v', 'model.visual_model.encoder.fusion_layers.0.weight_l', 'model.visual_model.encoder.fusion_layers.3.weight_l', 'model.visual_model.encoder.fusion_layers.4.weight_l', 'model.visual_model.encoder.fusion_layers.1.weight_v', 'model.visual_model.encoder.fusion_layers.5.weight_l', 'model.visual_model.encoder.fusion_layers.2.weight_v', 'model.visual_model.encoder.fusion_layers.0.weight_v']
- This IS expected if you are initializing LennaForCausalLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing LennaForCausalLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of LennaForCausalLM were not initialized from the model checkpoint at ./ckpt/Lenna-7B and are newly initialized: ['model.visual_model.encoder.fusion_layers.4.gamma_l', 'model.visual_model.encoder.fusion_layers.1.gamma_l', 'model.visual_model.encoder.fusion_layers.0.gamma_v', 'model.visual_model.encoder.fusion_layers.3.gamma_l', 'model.visual_model.encoder.fusion_layers.4.gamma_v', 'model.visual_model.encoder.fusion_layers.5.gamma_l', 'model.visual_model.encoder.fusion_layers.2.gamma_l', 'model.visual_model.encoder.fusion_layers.1.gamma_v', 'model.visual_model.encoder.fusion_layers.3.gamma_v', 'model.visual_model.encoder.fusion_layers.0.gamma_l', 'model.visual_model.encoder.fusion_layers.5.gamma_v', 'model.visual_model.encoder.fusion_layers.2.gamma_v']