import torch
from transformers import GPT2Tokenizer, GPT2LMHeadModel
model_name = "gpt2" #output_dir #
tokenizer = GPT2Tokenizer.from_pretrained(model_name, padding_side='left', add_eos_token=True) #gpt2
llm_model = GPT2LMHeadModel.from_pretrained(model_name) # gpt2
device = torch.device('cuda')
llm_model.to(device)
tokenizer.pad_token = tokenizer.eos_token
# Configure the model
llm_model.config.pad_token_id = tokenizer.eos_token_id
llm_model.config.eos_token_id = tokenizer.eos_token_id
llm_model.config.vocab_size = llm_model.config.vocab_size + len(tokenizer.get_added_vocab())
llm_model.resize_token_embeddings(len(tokenizer))
llm_model.config.pad_token_id = tokenizer.pad_token_id
print(tokenizer.eos_token_id)
result
github issue
