mrm8488/t5-base-iterater

T5 (base) fine-tuned on IteraTeR

This model is a fine-tuned version of t5-base on an IteraTeR dataset. It achieves the following results on the evaluation set:

• Loss: 0.2580

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.3286	0.09	2000	0.3010
0.3194	0.18	4000	0.2872
0.3208	0.27	6000	0.2792
0.3091	0.36	8000	0.2731
0.3164	0.45	10000	0.2678
0.2941	0.54	12000	0.2682
0.2981	0.63	14000	0.2696
0.2975	0.72	16000	0.2643
0.3109	0.81	18000	0.2624
0.2965	0.9	20000	0.2648
0.3053	0.99	22000	0.2627
0.2779	1.08	24000	0.2632
0.2692	1.17	26000	0.2608
0.2755	1.26	28000	0.2600
0.2771	1.35	30000	0.2584
0.2774	1.44	32000	0.2609
0.2976	1.53	34000	0.2593
0.2646	1.62	36000	0.2616
0.2705	1.71	38000	0.2574
0.2714	1.8	40000	0.2577
0.2857	1.9	42000	0.2576
0.2832	1.99	44000	0.2580

How to use

from transformers import T5ForConditionalGeneration, T5TokenizerFast
MODEL_CKPT = 'mrm8488/t5-base-iterater'

tokenizer = T5TokenizerFast.from_pretrained(MODEL_CKPT)
model = T5ForConditionalGeneration.from_pretrained(MODEL_CKPT)

def predict(intent, text):
   input_text =  f"<{intent}>  {text}"
   features = tokenizer([input_text], return_tensors='pt')
   output = model.generate(input_ids=features['input_ids'], 
               attention_mask=features['attention_mask'], max_length=128, num_beams=8)
   return tokenizer.decode(output[0], skip_special_tokens=True)
   
text = "Delay-based schemes have the potential to resolve this last packet problem by scheduling the link based on the delay for the packet has encountered."
intent = "clarity"

predict(intent, text)
# Delay-based schemes have the potential to resolve this last packet problem by scheduling the link based on the delay the packet has encountered.

Framework versions

• Transformers 4.18.0.dev0
• Pytorch 1.10.0+cu111
• Datasets 2.0.0
• Tokenizers 0.11.6