T5 model

1. download through script

from transformers import T5Tokenizer, T5Model

# Specify the model name
model_name = 't5-base'  # or 't5-small', 't5-large', 't5-3b', 't5-11b' depending on what you want

# Download the model and the tokenizer
model = T5Model.from_pretrained(model_name)
tokenizer = T5Tokenizer.from_pretrained(model_name)

# Save the model and the tokenizer to a directory
model.save_pretrained('T5-base')
tokenizer.save_pretrained('T5-base')

2. encode a sentence with T5

from transformers import T5Tokenizer, T5Model
import torch

# Specify the path to your locally saved model
model_dir = 'T5/t5-base'

# Load pre-trained model tokenizer (vocabulary)
tokenizer = T5Tokenizer.from_pretrained(model_dir)

# Sentence to encode
text = "Hello, my dog is cute"

# Tokenize our sentence with the T5 tokenizer.
tokenized_text = tokenizer.encode_plus(text, return_tensors="pt")

# Load pre-trained model (weights)
model = T5Model.from_pretrained(model_dir)

# Put the model in "evaluation" mode, meaning feed-forward operation.
model.eval()

# Predict hidden states features for each layer
with torch.no_grad():
    # Exclude token_type_ids for T5
    outputs = model(input_ids=tokenized_text['input_ids'], attention_mask=tokenized_text['attention_mask'])

# Get the hidden states from the model
hidden_states = outputs[0]

print(hidden_states)