t5-large-finetuned-break-qdmr-decomposition

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

• Loss: 0.1729
• Bleu: 0.2217
• Brevity Penalty: 0.2926
• Length Ratio: 0.4487
• Translation Length: 108954
• Reference Length: 242845

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: 0.0001
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 128
• 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	Bleu	Brevity Penalty	Length Ratio	Translation Length	Reference Length
No log	1.0	346	0.2217	0.2190	0.2973	0.4519	109738	242845
0.3597	2.0	692	0.1898	0.2213	0.2944	0.4499	109245	242845
0.1943	3.0	1038	0.1780	0.2213	0.2936	0.4494	109125	242845
0.1943	4.0	1385	0.1722	0.2209	0.2926	0.4486	108943	242845
0.1588	5.0	1731	0.1708	0.2221	0.2938	0.4495	109159	242845
0.1395	6.0	2077	0.1699	0.2209	0.2907	0.4473	108635	242845
0.1395	7.0	2423	0.1699	0.2219	0.2927	0.4487	108964	242845
0.1245	8.0	2770	0.1717	0.2215	0.2924	0.4485	108909	242845
0.1152	9.0	3116	0.1724	0.2215	0.2924	0.4485	108914	242845
0.1152	9.99	3460	0.1729	0.2217	0.2926	0.4487	108954	242845

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3