import torch
import transformers

use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")

t5_tokenizer = transformers.GPT2Tokenizer.from_pretrained("AlexWortega/FlanFred")
t5_model = transformers.T5ForConditionalGeneration.from_pretrained("AlexWortega/FlanFred")

def generate_text(input_str, tokenizer, model, device, max_length=50):
    # encode the input string to model's input_ids
    input_ids = tokenizer.encode(input_str, return_tensors='pt').to(device)
    
    # generate text
    with torch.no_grad():
        outputs = model.generate(input_ids=input_ids, max_length=max_length, num_return_sequences=1, temperature=0.7, do_sample=True)
    
    # decode the output and return the text
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# usage:
input_str = "Hello, how are you?"
print(generate_text(input_str, t5_tokenizer, t5_model, device))

Metrics:

| Metric        | flanfred | siberianfred  | fred  |
| ------------- | ----- |------ |----- |
| xnli_en       | 0.51   |0.49  |0.041 |
| xnli_ru       | 0.71   |0.62 |0.55 |
| xwinograd_ru  | 0.66   |0.51 |0.54 |

Citation

@MISC{AlexWortega/flan_translated_300k,
    author  = {Pavel Ilin, Ksenia Zolian,Ilya kuleshov, Egor Kokush, Aleksandr Nikolich},
    title   = {Russian Flan translated},
    url     = {https://huggingface.co/datasets/AlexWortega/flan_translated_300k},
    year    = 2023
}