base model初始化large model，造成的参数矩阵对不上权重不匹配问题

先报错没有指定文件
OSError: Error no file named pytorch_model.bin, tf_model.h5, model.ckpt.index or flax_model.msgpack found in directory

在https://huggingface.co/搜索下载后，载入预训练模型时Pytorch遇到权重不匹配的问题

raise RuntimeError(f"Error(s) in loading state_dict for {model.__class__.__name__}:\n\t{error_msg}")
RuntimeError: Error(s) in loading state_dict for PegasusForConditionalGeneration:
	size mismatch for final_logits_bias: copying a param with shape torch.Size([1, 96103]) from checkpoint, the shape in current model is torch.Size([1, 21128]).
	size mismatch for model.shared.weight: copying a param with shape torch.Size([96103, 1024]) from checkpoint, the shape in current model is torch.Size([21128, 768]).
	size mismatch for model.encoder.embed_tokens.weight: copying a param with shape torch.Size([96103, 1024]) from checkpoint, the shape in current model is torch.Size([21128, 768]).
	size mismatch for model.encoder.embed_positions.weight: copying a param with shape torch.Size([512, 1024]) from checkpoint, the shape in current model is torch.Size([1024, 768]).
	size mismatch for model.encoder.layers.0.self_attn.k_proj.weight: copying a param with shape torch.Size([1024, 1024]) from checkpoint, the shape in current model is torch.Size([768, 768]).

百度主要两个原因：
1、现为CPU，但加载了原先GPU训练的pkl
2、代码原因

首先排除代码因素，然后查看gpu状态

import torch
print(torch.cuda.is_available())

排除以上两个原因后，尝试删除.pkl缓存文件，重新生成.pkl文件

依旧报错

请教学弟，
可能原因：
arg.那个值，可能用base model初始化了一个large model，所以参数矩阵对不上

解决方案1:在config里面修改参数
确认了没问题
解决方案2：pytorch model文件大了
下载的是large model ，但初始化用的是base model

重新下载小点的文件

解决啦hhh