paraphraser

This model is a fine-tuned version of t5-base on the cointegrated/ru-paraphrase-NMT-Leipzig dataset. It achieves the following results on the evaluation set:

• Loss: 0.2841

Intended uses & limitations

from transformers import T5TokenizerFast, T5ForConditionalGeneration

encoded = tokenizer.encode('перефразируй: В какой срок необходимо оповестить Вайлдберрис о закрытии?', return_tensors='pt')
output_sequences = model.generate(
    input_ids=encoded,
    max_length=128,
    temperature=0.7,
    top_k=0,
    top_p=0.9,
    repetition_penalty=1,
    do_sample=True,
    num_return_sequences=5,
    pad_token_id=0
)
decoded = tokenizer.batch_decode(output_sequences, skip_special_tokens=True)
# Когда нужно будет известить Вайлдберриса о закрытии?

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
No log	1.0	141	0.2860
No log	2.0	282	0.2858
No log	3.0	423	0.2847
0.3385	4.0	564	0.2839
0.3385	5.0	705	0.2840
0.3385	6.0	846	0.2837
0.3385	7.0	987	0.2844
0.3089	8.0	1128	0.2837
0.3089	9.0	1269	0.2840
0.3089	10.0	1410	0.2841

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.0
• Datasets 2.1.0
• Tokenizers 0.14.1