synpre_mix_v1_1M_t5-small

This model is a fine-tuned version of t5-small on the tyzhu/synpre_mix_v1_1M dataset. It achieves the following results on the evaluation set:

• Loss: 0.0410
• Bleu: 96.6156
• Gen Len: 77.3534

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: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 10000
• training_steps: 80000

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
9.4619	0.64	5000	9.3641	0.1465	180.5239
9.0544	1.28	10000	8.9063	1.0907	62.4499
7.7965	1.92	15000	6.9912	3.0144	42.4833
0.5652	2.56	20000	0.2051	80.5252	86.4282
0.0696	3.2	25000	0.0410	96.6156	77.3534

Framework versions

• Transformers 4.34.0
• Pytorch 2.1.0+cu121
• Datasets 2.14.5
• Tokenizers 0.14.1