add T5 as decoder only

### Model description Since T5/ByT5 is an encoder-decoder model, I created a subclass T5DecoderOnlyForCausalLM to use it as decoder only for OCR task: `from transformers.models.t5.modeling_t5 import T5PreTrainedModel, T5Stack import torch import torch.nn as nn from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions, Seq2SeqLMOutput class T5DecoderOnlyForCausalLM(T5PreTrainedModel): def __init__(self, config): super().__init__(config) self.shared = nn.Embedding(config.vocab_size, config.d_model) self.decoder = T5Stack(config, self.shared) self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False) self.is_decoder = True config.is_decoder = True config.use_cache = False def forward( self, input_ids=None, attention_mask=None, decoder_attention_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, decoder_input_ids=None, inputs_embeds=None, decoder_inputs_embeds=None, head_mask=None, use_cache=None, cross_attn_head_mask=None, past_key_values=None, output_attentions=None, output_hidden_states=None, return_dict=None, ): decoder_outputs = self.decoder( input_ids=input_ids, attention_mask=decoder_attention_mask, past_key_values=past_key_values, encoder_hidden_states=encoder_hidden_states, encoder_attention_mask=encoder_attention_mask, inputs_embeds=decoder_inputs_embeds, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) last_hidden_state = decoder_outputs.last_hidden_state logits = self.lm_head(last_hidden_state) hidden_states = decoder_outputs.hidden_states past_key_values = decoder_outputs.past_key_values attentions =…