philosophy_model

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

• Loss: 0.0005
• Rouge1: 0.81
• Rouge2: 0.8004
• Rougel: 0.8107
• Rougelsum: 0.809
• Gen Len: 18.5

Model description

This model summarises passages on Indian philosophy. Enter snippet from Hindu philosophy in text box on right. Click compute.

Intended uses & limitations

More information needed

Training and evaluation data

Dataset:130, train:100, test:30

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0056
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-06
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
No log	1.0	13	2.2462	0.3632	0.1462	0.3114	0.3126	18.3333
No log	2.0	26	1.4611	0.459	0.3039	0.4178	0.4178	18.5667
No log	3.0	39	0.8323	0.5465	0.4259	0.5247	0.5208	17.1333
No log	4.0	52	0.4723	0.6161	0.5176	0.601	0.6004	18.3667
No log	5.0	65	0.3121	0.6812	0.6078	0.6747	0.6714	18.2333
No log	6.0	78	0.1546	0.7418	0.7023	0.7338	0.7313	18.0667
No log	7.0	91	0.1121	0.7832	0.763	0.7802	0.7789	18.5
No log	8.0	104	0.0699	0.8014	0.7882	0.8027	0.8009	18.5333
No log	9.0	117	0.0459	0.7958	0.7805	0.7946	0.7917	18.5
No log	10.0	130	0.0517	0.8091	0.7958	0.8105	0.809	18.4667
No log	11.0	143	0.0358	0.7994	0.7852	0.7973	0.7946	18.5
No log	12.0	156	0.0418	0.7799	0.7548	0.7747	0.7732	18.2667
No log	13.0	169	0.0107	0.81	0.8004	0.8107	0.809	18.5
No log	14.0	182	0.0065	0.8077	0.7971	0.8094	0.8075	18.5
No log	15.0	195	0.0178	0.808	0.796	0.8094	0.8075	18.3667
No log	16.0	208	0.0017	0.81	0.8004	0.8107	0.809	18.5
No log	17.0	221	0.0055	0.81	0.8004	0.8107	0.809	18.5
No log	18.0	234	0.0020	0.81	0.8004	0.8107	0.809	18.5
No log	19.0	247	0.0006	0.81	0.8004	0.8107	0.809	18.5
No log	20.0	260	0.0005	0.81	0.8004	0.8107	0.809	18.5

Framework versions

• Transformers 4.33.2
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3